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
  29 #include <linux/pci.h>
  30 #include <linux/seq_file.h>
  31 #include <linux/slab.h>
  32
  33 #include <drm/drm.h>
  34 #include <drm/drm_device.h>
  35 #include <drm/drm_file.h>
  36 #include <drm/radeon_drm.h>
  37
  38 #include "r100_track.h"
  39 #include "r300_reg_safe.h"
  40 #include "r300d.h"
  41 #include "radeon.h"
  42 #include "radeon_asic.h"
  43 #include "radeon_reg.h"
  44 #include "rv350d.h"
  45
  46 /* This files gather functions specifics to: r300,r350,rv350,rv370,rv380
  47  *
  48  * GPU Errata:
  49  * - HOST_PATH_CNTL: r300 family seems to dislike write to HOST_PATH_CNTL
  50  *   using MMIO to flush host path read cache, this lead to HARDLOCKUP.
  51  *   However, scheduling such write to the ring seems harmless, i suspect
  52  *   the CP read collide with the flush somehow, or maybe the MC, hard to
  53  *   tell. (Jerome Glisse)
  54  */
  55
  56 /*
  57  * Indirect registers accessor
  58  */
  59 uint32_t rv370_pcie_rreg(struct radeon_device *rdev, uint32_t reg)
  60 {
  61         unsigned long flags;
  62         uint32_t r;
  63
  64         spin_lock_irqsave(&rdev->pcie_idx_lock, flags);
  65         WREG32(RADEON_PCIE_INDEX, ((reg) & rdev->pcie_reg_mask));
  66         r = RREG32(RADEON_PCIE_DATA);
  67         spin_unlock_irqrestore(&rdev->pcie_idx_lock, flags);
  68         return r;
  69 }
  70
  71 void rv370_pcie_wreg(struct radeon_device *rdev, uint32_t reg, uint32_t v)
  72 {
  73         unsigned long flags;
  74
  75         spin_lock_irqsave(&rdev->pcie_idx_lock, flags);
  76         WREG32(RADEON_PCIE_INDEX, ((reg) & rdev->pcie_reg_mask));
  77         WREG32(RADEON_PCIE_DATA, (v));
  78         spin_unlock_irqrestore(&rdev->pcie_idx_lock, flags);
  79 }
  80
  81 /*
  82  * rv370,rv380 PCIE GART
  83  */
  84 static void rv370_debugfs_pcie_gart_info_init(struct radeon_device *rdev);
  85
  86 void rv370_pcie_gart_tlb_flush(struct radeon_device *rdev)
  87 {
  88         uint32_t tmp;
  89         int i;
  90
  91         /* Workaround HW bug do flush 2 times */
  92         for (i = 0; i < 2; i++) {
  93                 tmp = RREG32_PCIE(RADEON_PCIE_TX_GART_CNTL);
  94                 WREG32_PCIE(RADEON_PCIE_TX_GART_CNTL, tmp | RADEON_PCIE_TX_GART_INVALIDATE_TLB);
  95                 (void)RREG32_PCIE(RADEON_PCIE_TX_GART_CNTL);
  96                 WREG32_PCIE(RADEON_PCIE_TX_GART_CNTL, tmp);
  97         }
  98         mb();
  99 }
 100
 101 #define R300_PTE_UNSNOOPED (1 << 0)
 102 #define R300_PTE_WRITEABLE (1 << 2)
 103 #define R300_PTE_READABLE  (1 << 3)
 104
 105 uint64_t rv370_pcie_gart_get_page_entry(uint64_t addr, uint32_t flags)
 106 {
 107         addr = (lower_32_bits(addr) >> 8) |
 108                 ((upper_32_bits(addr) & 0xff) << 24);
 109         if (flags & RADEON_GART_PAGE_READ)
 110                 addr |= R300_PTE_READABLE;
 111         if (flags & RADEON_GART_PAGE_WRITE)
 112                 addr |= R300_PTE_WRITEABLE;
 113         if (!(flags & RADEON_GART_PAGE_SNOOP))
 114                 addr |= R300_PTE_UNSNOOPED;
 115         return addr;
 116 }
 117
 118 void rv370_pcie_gart_set_page(struct radeon_device *rdev, unsigned i,
 119                               uint64_t entry)
 120 {
 121         void __iomem *ptr = rdev->gart.ptr;
 122
 123         /* on x86 we want this to be CPU endian, on powerpc
 124          * on powerpc without HW swappers, it'll get swapped on way
 125          * into VRAM - so no need for cpu_to_le32 on VRAM tables */
 126         writel(entry, ((void __iomem *)ptr) + (i * 4));
 127 }
 128
 129 int rv370_pcie_gart_init(struct radeon_device *rdev)
 130 {
 131         int r;
 132
 133         if (rdev->gart.robj) {
 134                 WARN(1, "RV370 PCIE GART already initialized\n");
 135                 return 0;
 136         }
 137         /* Initialize common gart structure */
 138         r = radeon_gart_init(rdev);
 139         if (r)
 140                 return r;
 141         rv370_debugfs_pcie_gart_info_init(rdev);
 142
 143         rdev->gart.table_size = rdev->gart.num_gpu_pages * 4;
 144         rdev->asic->gart.tlb_flush = &rv370_pcie_gart_tlb_flush;
 145         rdev->asic->gart.get_page_entry = &rv370_pcie_gart_get_page_entry;
 146         rdev->asic->gart.set_page = &rv370_pcie_gart_set_page;
 147         return radeon_gart_table_vram_alloc(rdev);
 148 }
 149
 150 int rv370_pcie_gart_enable(struct radeon_device *rdev)
 151 {
 152         uint32_t table_addr;
 153         uint32_t tmp;
 154         int r;
 155
 156         if (rdev->gart.robj == NULL) {
 157                 dev_err(rdev->dev, "No VRAM object for PCIE GART.\n");
 158                 return -EINVAL;
 159         }
 160         r = radeon_gart_table_vram_pin(rdev);
 161         if (r)
 162                 return r;
 163         /* discard memory request outside of configured range */
 164         tmp = RADEON_PCIE_TX_GART_UNMAPPED_ACCESS_DISCARD;
 165         WREG32_PCIE(RADEON_PCIE_TX_GART_CNTL, tmp);
 166         WREG32_PCIE(RADEON_PCIE_TX_GART_START_LO, rdev->mc.gtt_start);
 167         tmp = rdev->mc.gtt_end & ~RADEON_GPU_PAGE_MASK;
 168         WREG32_PCIE(RADEON_PCIE_TX_GART_END_LO, tmp);
 169         WREG32_PCIE(RADEON_PCIE_TX_GART_START_HI, 0);
 170         WREG32_PCIE(RADEON_PCIE_TX_GART_END_HI, 0);
 171         table_addr = rdev->gart.table_addr;
 172         WREG32_PCIE(RADEON_PCIE_TX_GART_BASE, table_addr);
 173         /* FIXME: setup default page */
 174         WREG32_PCIE(RADEON_PCIE_TX_DISCARD_RD_ADDR_LO, rdev->mc.vram_start);
 175         WREG32_PCIE(RADEON_PCIE_TX_DISCARD_RD_ADDR_HI, 0);
 176         /* Clear error */
 177         WREG32_PCIE(RADEON_PCIE_TX_GART_ERROR, 0);
 178         tmp = RREG32_PCIE(RADEON_PCIE_TX_GART_CNTL);
 179         tmp |= RADEON_PCIE_TX_GART_EN;
 180         tmp |= RADEON_PCIE_TX_GART_UNMAPPED_ACCESS_DISCARD;
 181         WREG32_PCIE(RADEON_PCIE_TX_GART_CNTL, tmp);
 182         rv370_pcie_gart_tlb_flush(rdev);
 183         DRM_INFO("PCIE GART of %uM enabled (table at 0x%016llX).\n",
 184                  (unsigned)(rdev->mc.gtt_size >> 20),
 185                  (unsigned long long)table_addr);
 186         rdev->gart.ready = true;
 187         return 0;
 188 }
 189
 190 void rv370_pcie_gart_disable(struct radeon_device *rdev)
 191 {
 192         u32 tmp;
 193
 194         WREG32_PCIE(RADEON_PCIE_TX_GART_START_LO, 0);
 195         WREG32_PCIE(RADEON_PCIE_TX_GART_END_LO, 0);
 196         WREG32_PCIE(RADEON_PCIE_TX_GART_START_HI, 0);
 197         WREG32_PCIE(RADEON_PCIE_TX_GART_END_HI, 0);
 198         tmp = RREG32_PCIE(RADEON_PCIE_TX_GART_CNTL);
 199         tmp |= RADEON_PCIE_TX_GART_UNMAPPED_ACCESS_DISCARD;
 200         WREG32_PCIE(RADEON_PCIE_TX_GART_CNTL, tmp & ~RADEON_PCIE_TX_GART_EN);
 201         radeon_gart_table_vram_unpin(rdev);
 202 }
 203
 204 void rv370_pcie_gart_fini(struct radeon_device *rdev)
 205 {
 206         radeon_gart_fini(rdev);
 207         rv370_pcie_gart_disable(rdev);
 208         radeon_gart_table_vram_free(rdev);
 209 }
 210
 211 void r300_fence_ring_emit(struct radeon_device *rdev,
 212                           struct radeon_fence *fence)
 213 {
 214         struct radeon_ring *ring = &rdev->ring[fence->ring];
 215
 216         /* Who ever call radeon_fence_emit should call ring_lock and ask
 217          * for enough space (today caller are ib schedule and buffer move) */
 218         /* Write SC register so SC & US assert idle */
 219         radeon_ring_write(ring, PACKET0(R300_RE_SCISSORS_TL, 0));
 220         radeon_ring_write(ring, 0);
 221         radeon_ring_write(ring, PACKET0(R300_RE_SCISSORS_BR, 0));
 222         radeon_ring_write(ring, 0);
 223         /* Flush 3D cache */
 224         radeon_ring_write(ring, PACKET0(R300_RB3D_DSTCACHE_CTLSTAT, 0));
 225         radeon_ring_write(ring, R300_RB3D_DC_FLUSH);
 226         radeon_ring_write(ring, PACKET0(R300_RB3D_ZCACHE_CTLSTAT, 0));
 227         radeon_ring_write(ring, R300_ZC_FLUSH);
 228         /* Wait until IDLE & CLEAN */
 229         radeon_ring_write(ring, PACKET0(RADEON_WAIT_UNTIL, 0));
 230         radeon_ring_write(ring, (RADEON_WAIT_3D_IDLECLEAN |
 231                                  RADEON_WAIT_2D_IDLECLEAN |
 232                                  RADEON_WAIT_DMA_GUI_IDLE));
 233         radeon_ring_write(ring, PACKET0(RADEON_HOST_PATH_CNTL, 0));
 234         radeon_ring_write(ring, rdev->config.r300.hdp_cntl |
 235                                 RADEON_HDP_READ_BUFFER_INVALIDATE);
 236         radeon_ring_write(ring, PACKET0(RADEON_HOST_PATH_CNTL, 0));
 237         radeon_ring_write(ring, rdev->config.r300.hdp_cntl);
 238         /* Emit fence sequence & fire IRQ */
 239         radeon_ring_write(ring, PACKET0(rdev->fence_drv[fence->ring].scratch_reg, 0));
 240         radeon_ring_write(ring, fence->seq);
 241         radeon_ring_write(ring, PACKET0(RADEON_GEN_INT_STATUS, 0));
 242         radeon_ring_write(ring, RADEON_SW_INT_FIRE);
 243 }
 244
 245 void r300_ring_start(struct radeon_device *rdev, struct radeon_ring *ring)
 246 {
 247         unsigned gb_tile_config;
 248         int r;
 249
 250         /* Sub pixel 1/12 so we can have 4K rendering according to doc */
 251         gb_tile_config = (R300_ENABLE_TILING | R300_TILE_SIZE_16);
 252         switch (rdev->num_gb_pipes) {
 253         case 2:
 254                 gb_tile_config |= R300_PIPE_COUNT_R300;
 255                 break;
 256         case 3:
 257                 gb_tile_config |= R300_PIPE_COUNT_R420_3P;
 258                 break;
 259         case 4:
 260                 gb_tile_config |= R300_PIPE_COUNT_R420;
 261                 break;
 262         case 1:
 263         default:
 264                 gb_tile_config |= R300_PIPE_COUNT_RV350;
 265                 break;
 266         }
 267
 268         r = radeon_ring_lock(rdev, ring, 64);
 269         if (r) {
 270                 return;
 271         }
 272         radeon_ring_write(ring, PACKET0(RADEON_ISYNC_CNTL, 0));
 273         radeon_ring_write(ring,
 274                           RADEON_ISYNC_ANY2D_IDLE3D |
 275                           RADEON_ISYNC_ANY3D_IDLE2D |
 276                           RADEON_ISYNC_WAIT_IDLEGUI |
 277                           RADEON_ISYNC_CPSCRATCH_IDLEGUI);
 278         radeon_ring_write(ring, PACKET0(R300_GB_TILE_CONFIG, 0));
 279         radeon_ring_write(ring, gb_tile_config);
 280         radeon_ring_write(ring, PACKET0(RADEON_WAIT_UNTIL, 0));
 281         radeon_ring_write(ring,
 282                           RADEON_WAIT_2D_IDLECLEAN |
 283                           RADEON_WAIT_3D_IDLECLEAN);
 284         radeon_ring_write(ring, PACKET0(R300_DST_PIPE_CONFIG, 0));
 285         radeon_ring_write(ring, R300_PIPE_AUTO_CONFIG);
 286         radeon_ring_write(ring, PACKET0(R300_GB_SELECT, 0));
 287         radeon_ring_write(ring, 0);
 288         radeon_ring_write(ring, PACKET0(R300_GB_ENABLE, 0));
 289         radeon_ring_write(ring, 0);
 290         radeon_ring_write(ring, PACKET0(R300_RB3D_DSTCACHE_CTLSTAT, 0));
 291         radeon_ring_write(ring, R300_RB3D_DC_FLUSH | R300_RB3D_DC_FREE);
 292         radeon_ring_write(ring, PACKET0(R300_RB3D_ZCACHE_CTLSTAT, 0));
 293         radeon_ring_write(ring, R300_ZC_FLUSH | R300_ZC_FREE);
 294         radeon_ring_write(ring, PACKET0(RADEON_WAIT_UNTIL, 0));
 295         radeon_ring_write(ring,
 296                           RADEON_WAIT_2D_IDLECLEAN |
 297                           RADEON_WAIT_3D_IDLECLEAN);
 298         radeon_ring_write(ring, PACKET0(R300_GB_AA_CONFIG, 0));
 299         radeon_ring_write(ring, 0);
 300         radeon_ring_write(ring, PACKET0(R300_RB3D_DSTCACHE_CTLSTAT, 0));
 301         radeon_ring_write(ring, R300_RB3D_DC_FLUSH | R300_RB3D_DC_FREE);
 302         radeon_ring_write(ring, PACKET0(R300_RB3D_ZCACHE_CTLSTAT, 0));
 303         radeon_ring_write(ring, R300_ZC_FLUSH | R300_ZC_FREE);
 304         radeon_ring_write(ring, PACKET0(R300_GB_MSPOS0, 0));
 305         radeon_ring_write(ring,
 306                           ((6 << R300_MS_X0_SHIFT) |
 307                            (6 << R300_MS_Y0_SHIFT) |
 308                            (6 << R300_MS_X1_SHIFT) |
 309                            (6 << R300_MS_Y1_SHIFT) |
 310                            (6 << R300_MS_X2_SHIFT) |
 311                            (6 << R300_MS_Y2_SHIFT) |
 312                            (6 << R300_MSBD0_Y_SHIFT) |
 313                            (6 << R300_MSBD0_X_SHIFT)));
 314         radeon_ring_write(ring, PACKET0(R300_GB_MSPOS1, 0));
 315         radeon_ring_write(ring,
 316                           ((6 << R300_MS_X3_SHIFT) |
 317                            (6 << R300_MS_Y3_SHIFT) |
 318                            (6 << R300_MS_X4_SHIFT) |
 319                            (6 << R300_MS_Y4_SHIFT) |
 320                            (6 << R300_MS_X5_SHIFT) |
 321                            (6 << R300_MS_Y5_SHIFT) |
 322                            (6 << R300_MSBD1_SHIFT)));
 323         radeon_ring_write(ring, PACKET0(R300_GA_ENHANCE, 0));
 324         radeon_ring_write(ring, R300_GA_DEADLOCK_CNTL | R300_GA_FASTSYNC_CNTL);
 325         radeon_ring_write(ring, PACKET0(R300_GA_POLY_MODE, 0));
 326         radeon_ring_write(ring,
 327                           R300_FRONT_PTYPE_TRIANGE | R300_BACK_PTYPE_TRIANGE);
 328         radeon_ring_write(ring, PACKET0(R300_GA_ROUND_MODE, 0));
 329         radeon_ring_write(ring,
 330                           R300_GEOMETRY_ROUND_NEAREST |
 331                           R300_COLOR_ROUND_NEAREST);
 332         radeon_ring_unlock_commit(rdev, ring, false);
 333 }
 334
 335 static void r300_errata(struct radeon_device *rdev)
 336 {
 337         rdev->pll_errata = 0;
 338
 339         if (rdev->family == CHIP_R300 &&
 340             (RREG32(RADEON_CONFIG_CNTL) & RADEON_CFG_ATI_REV_ID_MASK) == RADEON_CFG_ATI_REV_A11) {
 341                 rdev->pll_errata |= CHIP_ERRATA_R300_CG;
 342         }
 343 }
 344
 345 int r300_mc_wait_for_idle(struct radeon_device *rdev)
 346 {
 347         unsigned i;
 348         uint32_t tmp;
 349
 350         for (i = 0; i < rdev->usec_timeout; i++) {
 351                 /* read MC_STATUS */
 352                 tmp = RREG32(RADEON_MC_STATUS);
 353                 if (tmp & R300_MC_IDLE) {
 354                         return 0;
 355                 }
 356                 udelay(1);
 357         }
 358         return -1;
 359 }
 360
 361 static void r300_gpu_init(struct radeon_device *rdev)
 362 {
 363         uint32_t gb_tile_config, tmp;
 364
 365         if ((rdev->family == CHIP_R300 && rdev->pdev->device != 0x4144) ||
 366             (rdev->family == CHIP_R350 && rdev->pdev->device != 0x4148)) {
 367                 /* r300,r350 */
 368                 rdev->num_gb_pipes = 2;
 369         } else {
 370                 /* rv350,rv370,rv380,r300 AD, r350 AH */
 371                 rdev->num_gb_pipes = 1;
 372         }
 373         rdev->num_z_pipes = 1;
 374         gb_tile_config = (R300_ENABLE_TILING | R300_TILE_SIZE_16);
 375         switch (rdev->num_gb_pipes) {
 376         case 2:
 377                 gb_tile_config |= R300_PIPE_COUNT_R300;
 378                 break;
 379         case 3:
 380                 gb_tile_config |= R300_PIPE_COUNT_R420_3P;
 381                 break;
 382         case 4:
 383                 gb_tile_config |= R300_PIPE_COUNT_R420;
 384                 break;
 385         default:
 386         case 1:
 387                 gb_tile_config |= R300_PIPE_COUNT_RV350;
 388                 break;
 389         }
 390         WREG32(R300_GB_TILE_CONFIG, gb_tile_config);
 391
 392         if (r100_gui_wait_for_idle(rdev)) {
 393                 pr_warn("Failed to wait GUI idle while programming pipes. Bad things might happen.\n");
 394         }
 395
 396         tmp = RREG32(R300_DST_PIPE_CONFIG);
 397         WREG32(R300_DST_PIPE_CONFIG, tmp | R300_PIPE_AUTO_CONFIG);
 398
 399         WREG32(R300_RB2D_DSTCACHE_MODE,
 400                R300_DC_AUTOFLUSH_ENABLE |
 401                R300_DC_DC_DISABLE_IGNORE_PE);
 402
 403         if (r100_gui_wait_for_idle(rdev)) {
 404                 pr_warn("Failed to wait GUI idle while programming pipes. Bad things might happen.\n");
 405         }
 406         if (r300_mc_wait_for_idle(rdev)) {
 407                 pr_warn("Failed to wait MC idle while programming pipes. Bad things might happen.\n");
 408         }
 409         DRM_INFO("radeon: %d quad pipes, %d Z pipes initialized\n",
 410                  rdev->num_gb_pipes, rdev->num_z_pipes);
 411 }
 412
 413 int r300_asic_reset(struct radeon_device *rdev, bool hard)
 414 {
 415         struct r100_mc_save save;
 416         u32 status, tmp;
 417         int ret = 0;
 418
 419         status = RREG32(R_000E40_RBBM_STATUS);
 420         if (!G_000E40_GUI_ACTIVE(status)) {
 421                 return 0;
 422         }
 423         r100_mc_stop(rdev, &save);
 424         status = RREG32(R_000E40_RBBM_STATUS);
 425         dev_info(rdev->dev, "(%s:%d) RBBM_STATUS=0x%08X\n", __func__, __LINE__, status);
 426         /* stop CP */
 427         WREG32(RADEON_CP_CSQ_CNTL, 0);
 428         tmp = RREG32(RADEON_CP_RB_CNTL);
 429         WREG32(RADEON_CP_RB_CNTL, tmp | RADEON_RB_RPTR_WR_ENA);
 430         WREG32(RADEON_CP_RB_RPTR_WR, 0);
 431         WREG32(RADEON_CP_RB_WPTR, 0);
 432         WREG32(RADEON_CP_RB_CNTL, tmp);
 433         /* save PCI state */
 434         pci_save_state(rdev->pdev);
 435         /* disable bus mastering */
 436         r100_bm_disable(rdev);
 437         WREG32(R_0000F0_RBBM_SOFT_RESET, S_0000F0_SOFT_RESET_VAP(1) |
 438                                         S_0000F0_SOFT_RESET_GA(1));
 439         RREG32(R_0000F0_RBBM_SOFT_RESET);
 440         mdelay(500);
 441         WREG32(R_0000F0_RBBM_SOFT_RESET, 0);
 442         mdelay(1);
 443         status = RREG32(R_000E40_RBBM_STATUS);
 444         dev_info(rdev->dev, "(%s:%d) RBBM_STATUS=0x%08X\n", __func__, __LINE__, status);
 445         /* resetting the CP seems to be problematic sometimes it end up
 446          * hard locking the computer, but it's necessary for successful
 447          * reset more test & playing is needed on R3XX/R4XX to find a
 448          * reliable (if any solution)
 449          */
 450         WREG32(R_0000F0_RBBM_SOFT_RESET, S_0000F0_SOFT_RESET_CP(1));
 451         RREG32(R_0000F0_RBBM_SOFT_RESET);
 452         mdelay(500);
 453         WREG32(R_0000F0_RBBM_SOFT_RESET, 0);
 454         mdelay(1);
 455         status = RREG32(R_000E40_RBBM_STATUS);
 456         dev_info(rdev->dev, "(%s:%d) RBBM_STATUS=0x%08X\n", __func__, __LINE__, status);
 457         /* restore PCI & busmastering */
 458         pci_restore_state(rdev->pdev);
 459         r100_enable_bm(rdev);
 460         /* Check if GPU is idle */
 461         if (G_000E40_GA_BUSY(status) || G_000E40_VAP_BUSY(status)) {
 462                 dev_err(rdev->dev, "failed to reset GPU\n");
 463                 ret = -1;
 464         } else
 465                 dev_info(rdev->dev, "GPU reset succeed\n");
 466         r100_mc_resume(rdev, &save);
 467         return ret;
 468 }
 469
 470 /*
 471  * r300,r350,rv350,rv380 VRAM info
 472  */
 473 void r300_mc_init(struct radeon_device *rdev)
 474 {
 475         u64 base;
 476         u32 tmp;
 477
 478         /* DDR for all card after R300 & IGP */
 479         rdev->mc.vram_is_ddr = true;
 480         tmp = RREG32(RADEON_MEM_CNTL);
 481         tmp &= R300_MEM_NUM_CHANNELS_MASK;
 482         switch (tmp) {
 483         case 0: rdev->mc.vram_width = 64; break;
 484         case 1: rdev->mc.vram_width = 128; break;
 485         case 2: rdev->mc.vram_width = 256; break;
 486         default:  rdev->mc.vram_width = 128; break;
 487         }
 488         r100_vram_init_sizes(rdev);
 489         base = rdev->mc.aper_base;
 490         if (rdev->flags & RADEON_IS_IGP)
 491                 base = (RREG32(RADEON_NB_TOM) & 0xffff) << 16;
 492         radeon_vram_location(rdev, &rdev->mc, base);
 493         rdev->mc.gtt_base_align = 0;
 494         if (!(rdev->flags & RADEON_IS_AGP))
 495                 radeon_gtt_location(rdev, &rdev->mc);
 496         radeon_update_bandwidth_info(rdev);
 497 }
 498
 499 void rv370_set_pcie_lanes(struct radeon_device *rdev, int lanes)
 500 {
 501         uint32_t link_width_cntl, mask;
 502
 503         if (rdev->flags & RADEON_IS_IGP)
 504                 return;
 505
 506         if (!(rdev->flags & RADEON_IS_PCIE))
 507                 return;
 508
 509         /* FIXME wait for idle */
 510
 511         switch (lanes) {
 512         case 0:
 513                 mask = RADEON_PCIE_LC_LINK_WIDTH_X0;
 514                 break;
 515         case 1:
 516                 mask = RADEON_PCIE_LC_LINK_WIDTH_X1;
 517                 break;
 518         case 2:
 519                 mask = RADEON_PCIE_LC_LINK_WIDTH_X2;
 520                 break;
 521         case 4:
 522                 mask = RADEON_PCIE_LC_LINK_WIDTH_X4;
 523                 break;
 524         case 8:
 525                 mask = RADEON_PCIE_LC_LINK_WIDTH_X8;
 526                 break;
 527         case 12:
 528                 mask = RADEON_PCIE_LC_LINK_WIDTH_X12;
 529                 break;
 530         case 16:
 531         default:
 532                 mask = RADEON_PCIE_LC_LINK_WIDTH_X16;
 533                 break;
 534         }
 535
 536         link_width_cntl = RREG32_PCIE(RADEON_PCIE_LC_LINK_WIDTH_CNTL);
 537
 538         if ((link_width_cntl & RADEON_PCIE_LC_LINK_WIDTH_RD_MASK) ==
 539             (mask << RADEON_PCIE_LC_LINK_WIDTH_RD_SHIFT))
 540                 return;
 541
 542         link_width_cntl &= ~(RADEON_PCIE_LC_LINK_WIDTH_MASK |
 543                              RADEON_PCIE_LC_RECONFIG_NOW |
 544                              RADEON_PCIE_LC_RECONFIG_LATER |
 545                              RADEON_PCIE_LC_SHORT_RECONFIG_EN);
 546         link_width_cntl |= mask;
 547         WREG32_PCIE(RADEON_PCIE_LC_LINK_WIDTH_CNTL, link_width_cntl);
 548         WREG32_PCIE(RADEON_PCIE_LC_LINK_WIDTH_CNTL, (link_width_cntl |
 549                                                      RADEON_PCIE_LC_RECONFIG_NOW));
 550
 551         /* wait for lane set to complete */
 552         link_width_cntl = RREG32_PCIE(RADEON_PCIE_LC_LINK_WIDTH_CNTL);
 553         while (link_width_cntl == 0xffffffff)
 554                 link_width_cntl = RREG32_PCIE(RADEON_PCIE_LC_LINK_WIDTH_CNTL);
 555
 556 }
 557
 558 int rv370_get_pcie_lanes(struct radeon_device *rdev)
 559 {
 560         u32 link_width_cntl;
 561
 562         if (rdev->flags & RADEON_IS_IGP)
 563                 return 0;
 564
 565         if (!(rdev->flags & RADEON_IS_PCIE))
 566                 return 0;
 567
 568         /* FIXME wait for idle */
 569
 570         link_width_cntl = RREG32_PCIE(RADEON_PCIE_LC_LINK_WIDTH_CNTL);
 571
 572         switch ((link_width_cntl & RADEON_PCIE_LC_LINK_WIDTH_RD_MASK) >> RADEON_PCIE_LC_LINK_WIDTH_RD_SHIFT) {
 573         case RADEON_PCIE_LC_LINK_WIDTH_X0:
 574                 return 0;
 575         case RADEON_PCIE_LC_LINK_WIDTH_X1:
 576                 return 1;
 577         case RADEON_PCIE_LC_LINK_WIDTH_X2:
 578                 return 2;
 579         case RADEON_PCIE_LC_LINK_WIDTH_X4:
 580                 return 4;
 581         case RADEON_PCIE_LC_LINK_WIDTH_X8:
 582                 return 8;
 583         case RADEON_PCIE_LC_LINK_WIDTH_X16:
 584         default:
 585                 return 16;
 586         }
 587 }
 588
 589 #if defined(CONFIG_DEBUG_FS)
 590 static int rv370_debugfs_pcie_gart_info_show(struct seq_file *m, void *unused)
 591 {
 592         struct radeon_device *rdev = m->private;
 593         uint32_t tmp;
 594
 595         tmp = RREG32_PCIE(RADEON_PCIE_TX_GART_CNTL);
 596         seq_printf(m, "PCIE_TX_GART_CNTL 0x%08x\n", tmp);
 597         tmp = RREG32_PCIE(RADEON_PCIE_TX_GART_BASE);
 598         seq_printf(m, "PCIE_TX_GART_BASE 0x%08x\n", tmp);
 599         tmp = RREG32_PCIE(RADEON_PCIE_TX_GART_START_LO);
 600         seq_printf(m, "PCIE_TX_GART_START_LO 0x%08x\n", tmp);
 601         tmp = RREG32_PCIE(RADEON_PCIE_TX_GART_START_HI);
 602         seq_printf(m, "PCIE_TX_GART_START_HI 0x%08x\n", tmp);
 603         tmp = RREG32_PCIE(RADEON_PCIE_TX_GART_END_LO);
 604         seq_printf(m, "PCIE_TX_GART_END_LO 0x%08x\n", tmp);
 605         tmp = RREG32_PCIE(RADEON_PCIE_TX_GART_END_HI);
 606         seq_printf(m, "PCIE_TX_GART_END_HI 0x%08x\n", tmp);
 607         tmp = RREG32_PCIE(RADEON_PCIE_TX_GART_ERROR);
 608         seq_printf(m, "PCIE_TX_GART_ERROR 0x%08x\n", tmp);
 609         return 0;
 610 }
 611
 612 DEFINE_SHOW_ATTRIBUTE(rv370_debugfs_pcie_gart_info);
 613 #endif
 614
 615 static void rv370_debugfs_pcie_gart_info_init(struct radeon_device *rdev)
 616 {
 617 #if defined(CONFIG_DEBUG_FS)
 618         struct dentry *root = rdev->ddev->primary->debugfs_root;
 619
 620         debugfs_create_file("rv370_pcie_gart_info", 0444, root, rdev,
 621                             &rv370_debugfs_pcie_gart_info_fops);
 622 #endif
 623 }
 624
 625 static int r300_packet0_check(struct radeon_cs_parser *p,
 626                 struct radeon_cs_packet *pkt,
 627                 unsigned idx, unsigned reg)
 628 {
 629         struct radeon_bo_list *reloc;
 630         struct r100_cs_track *track;
 631         volatile uint32_t *ib;
 632         uint32_t tmp, tile_flags = 0;
 633         unsigned i;
 634         int r;
 635         u32 idx_value;
 636
 637         ib = p->ib.ptr;
 638         track = (struct r100_cs_track *)p->track;
 639         idx_value = radeon_get_ib_value(p, idx);
 640
 641         switch (reg) {
 642         case AVIVO_D1MODE_VLINE_START_END:
 643         case RADEON_CRTC_GUI_TRIG_VLINE:
 644                 r = r100_cs_packet_parse_vline(p);
 645                 if (r) {
 646                         DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
 647                                         idx, reg);
 648                         radeon_cs_dump_packet(p, pkt);
 649                         return r;
 650                 }
 651                 break;
 652         case RADEON_DST_PITCH_OFFSET:
 653         case RADEON_SRC_PITCH_OFFSET:
 654                 r = r100_reloc_pitch_offset(p, pkt, idx, reg);
 655                 if (r)
 656                         return r;
 657                 break;
 658         case R300_RB3D_COLOROFFSET0:
 659         case R300_RB3D_COLOROFFSET1:
 660         case R300_RB3D_COLOROFFSET2:
 661         case R300_RB3D_COLOROFFSET3:
 662                 i = (reg - R300_RB3D_COLOROFFSET0) >> 2;
 663                 r = radeon_cs_packet_next_reloc(p, &reloc, 0);
 664                 if (r) {
 665                         DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
 666                                         idx, reg);
 667                         radeon_cs_dump_packet(p, pkt);
 668                         return r;
 669                 }
 670                 track->cb[i].robj = reloc->robj;
 671                 track->cb[i].offset = idx_value;
 672                 track->cb_dirty = true;
 673                 ib[idx] = idx_value + ((u32)reloc->gpu_offset);
 674                 break;
 675         case R300_ZB_DEPTHOFFSET:
 676                 r = radeon_cs_packet_next_reloc(p, &reloc, 0);
 677                 if (r) {
 678                         DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
 679                                         idx, reg);
 680                         radeon_cs_dump_packet(p, pkt);
 681                         return r;
 682                 }
 683                 track->zb.robj = reloc->robj;
 684                 track->zb.offset = idx_value;
 685                 track->zb_dirty = true;
 686                 ib[idx] = idx_value + ((u32)reloc->gpu_offset);
 687                 break;
 688         case R300_TX_OFFSET_0:
 689         case R300_TX_OFFSET_0+4:
 690         case R300_TX_OFFSET_0+8:
 691         case R300_TX_OFFSET_0+12:
 692         case R300_TX_OFFSET_0+16:
 693         case R300_TX_OFFSET_0+20:
 694         case R300_TX_OFFSET_0+24:
 695         case R300_TX_OFFSET_0+28:
 696         case R300_TX_OFFSET_0+32:
 697         case R300_TX_OFFSET_0+36:
 698         case R300_TX_OFFSET_0+40:
 699         case R300_TX_OFFSET_0+44:
 700         case R300_TX_OFFSET_0+48:
 701         case R300_TX_OFFSET_0+52:
 702         case R300_TX_OFFSET_0+56:
 703         case R300_TX_OFFSET_0+60:
 704                 i = (reg - R300_TX_OFFSET_0) >> 2;
 705                 r = radeon_cs_packet_next_reloc(p, &reloc, 0);
 706                 if (r) {
 707                         DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
 708                                         idx, reg);
 709                         radeon_cs_dump_packet(p, pkt);
 710                         return r;
 711                 }
 712
 713                 if (p->cs_flags & RADEON_CS_KEEP_TILING_FLAGS) {
 714                         ib[idx] = (idx_value & 31) | /* keep the 1st 5 bits */
 715                                   ((idx_value & ~31) + (u32)reloc->gpu_offset);
 716                 } else {
 717                         if (reloc->tiling_flags & RADEON_TILING_MACRO)
 718                                 tile_flags |= R300_TXO_MACRO_TILE;
 719                         if (reloc->tiling_flags & RADEON_TILING_MICRO)
 720                                 tile_flags |= R300_TXO_MICRO_TILE;
 721                         else if (reloc->tiling_flags & RADEON_TILING_MICRO_SQUARE)
 722                                 tile_flags |= R300_TXO_MICRO_TILE_SQUARE;
 723
 724                         tmp = idx_value + ((u32)reloc->gpu_offset);
 725                         tmp |= tile_flags;
 726                         ib[idx] = tmp;
 727                 }
 728                 track->textures[i].robj = reloc->robj;
 729                 track->tex_dirty = true;
 730                 break;
 731         /* Tracked registers */
 732         case 0x2084:
 733                 /* VAP_VF_CNTL */
 734                 track->vap_vf_cntl = idx_value;
 735                 break;
 736         case 0x20B4:
 737                 /* VAP_VTX_SIZE */
 738                 track->vtx_size = idx_value & 0x7F;
 739                 break;
 740         case 0x2134:
 741                 /* VAP_VF_MAX_VTX_INDX */
 742                 track->max_indx = idx_value & 0x00FFFFFFUL;
 743                 break;
 744         case 0x2088:
 745                 /* VAP_ALT_NUM_VERTICES - only valid on r500 */
 746                 if (p->rdev->family < CHIP_RV515)
 747                         goto fail;
 748                 track->vap_alt_nverts = idx_value & 0xFFFFFF;
 749                 break;
 750         case 0x43E4:
 751                 /* SC_SCISSOR1 */
 752                 track->maxy = ((idx_value >> 13) & 0x1FFF) + 1;
 753                 if (p->rdev->family < CHIP_RV515) {
 754                         track->maxy -= 1440;
 755                 }
 756                 track->cb_dirty = true;
 757                 track->zb_dirty = true;
 758                 break;
 759         case 0x4E00:
 760                 /* RB3D_CCTL */
 761                 if ((idx_value & (1 << 10)) && /* CMASK_ENABLE */
 762                     p->rdev->cmask_filp != p->filp) {
 763                         DRM_ERROR("Invalid RB3D_CCTL: Cannot enable CMASK.\n");
 764                         return -EINVAL;
 765                 }
 766                 track->num_cb = ((idx_value >> 5) & 0x3) + 1;
 767                 track->cb_dirty = true;
 768                 break;
 769         case 0x4E38:
 770         case 0x4E3C:
 771         case 0x4E40:
 772         case 0x4E44:
 773                 /* RB3D_COLORPITCH0 */
 774                 /* RB3D_COLORPITCH1 */
 775                 /* RB3D_COLORPITCH2 */
 776                 /* RB3D_COLORPITCH3 */
 777                 if (!(p->cs_flags & RADEON_CS_KEEP_TILING_FLAGS)) {
 778                         r = radeon_cs_packet_next_reloc(p, &reloc, 0);
 779                         if (r) {
 780                                 DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
 781                                           idx, reg);
 782                                 radeon_cs_dump_packet(p, pkt);
 783                                 return r;
 784                         }
 785
 786                         if (reloc->tiling_flags & RADEON_TILING_MACRO)
 787                                 tile_flags |= R300_COLOR_TILE_ENABLE;
 788                         if (reloc->tiling_flags & RADEON_TILING_MICRO)
 789                                 tile_flags |= R300_COLOR_MICROTILE_ENABLE;
 790                         else if (reloc->tiling_flags & RADEON_TILING_MICRO_SQUARE)
 791                                 tile_flags |= R300_COLOR_MICROTILE_SQUARE_ENABLE;
 792
 793                         tmp = idx_value & ~(0x7 << 16);
 794                         tmp |= tile_flags;
 795                         ib[idx] = tmp;
 796                 }
 797                 i = (reg - 0x4E38) >> 2;
 798                 track->cb[i].pitch = idx_value & 0x3FFE;
 799                 switch (((idx_value >> 21) & 0xF)) {
 800                 case 9:
 801                 case 11:
 802                 case 12:
 803                         track->cb[i].cpp = 1;
 804                         break;
 805                 case 3:
 806                 case 4:
 807                 case 13:
 808                 case 15:
 809                         track->cb[i].cpp = 2;
 810                         break;
 811                 case 5:
 812                         if (p->rdev->family < CHIP_RV515) {
 813                                 DRM_ERROR("Invalid color buffer format (%d)!\n",
 814                                           ((idx_value >> 21) & 0xF));
 815                                 return -EINVAL;
 816                         }
 817                         fallthrough;
 818                 case 6:
 819                         track->cb[i].cpp = 4;
 820                         break;
 821                 case 10:
 822                         track->cb[i].cpp = 8;
 823                         break;
 824                 case 7:
 825                         track->cb[i].cpp = 16;
 826                         break;
 827                 default:
 828                         DRM_ERROR("Invalid color buffer format (%d) !\n",
 829                                   ((idx_value >> 21) & 0xF));
 830                         return -EINVAL;
 831                 }
 832                 track->cb_dirty = true;
 833                 break;
 834         case 0x4F00:
 835                 /* ZB_CNTL */
 836                 if (idx_value & 2) {
 837                         track->z_enabled = true;
 838                 } else {
 839                         track->z_enabled = false;
 840                 }
 841                 track->zb_dirty = true;
 842                 break;
 843         case 0x4F10:
 844                 /* ZB_FORMAT */
 845                 switch ((idx_value & 0xF)) {
 846                 case 0:
 847                 case 1:
 848                         track->zb.cpp = 2;
 849                         break;
 850                 case 2:
 851                         track->zb.cpp = 4;
 852                         break;
 853                 default:
 854                         DRM_ERROR("Invalid z buffer format (%d) !\n",
 855                                   (idx_value & 0xF));
 856                         return -EINVAL;
 857                 }
 858                 track->zb_dirty = true;
 859                 break;
 860         case 0x4F24:
 861                 /* ZB_DEPTHPITCH */
 862                 if (!(p->cs_flags & RADEON_CS_KEEP_TILING_FLAGS)) {
 863                         r = radeon_cs_packet_next_reloc(p, &reloc, 0);
 864                         if (r) {
 865                                 DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
 866                                           idx, reg);
 867                                 radeon_cs_dump_packet(p, pkt);
 868                                 return r;
 869                         }
 870
 871                         if (reloc->tiling_flags & RADEON_TILING_MACRO)
 872                                 tile_flags |= R300_DEPTHMACROTILE_ENABLE;
 873                         if (reloc->tiling_flags & RADEON_TILING_MICRO)
 874                                 tile_flags |= R300_DEPTHMICROTILE_TILED;
 875                         else if (reloc->tiling_flags & RADEON_TILING_MICRO_SQUARE)
 876                                 tile_flags |= R300_DEPTHMICROTILE_TILED_SQUARE;
 877
 878                         tmp = idx_value & ~(0x7 << 16);
 879                         tmp |= tile_flags;
 880                         ib[idx] = tmp;
 881                 }
 882                 track->zb.pitch = idx_value & 0x3FFC;
 883                 track->zb_dirty = true;
 884                 break;
 885         case 0x4104:
 886                 /* TX_ENABLE */
 887                 for (i = 0; i < 16; i++) {
 888                         bool enabled;
 889
 890                         enabled = !!(idx_value & (1 << i));
 891                         track->textures[i].enabled = enabled;
 892                 }
 893                 track->tex_dirty = true;
 894                 break;
 895         case 0x44C0:
 896         case 0x44C4:
 897         case 0x44C8:
 898         case 0x44CC:
 899         case 0x44D0:
 900         case 0x44D4:
 901         case 0x44D8:
 902         case 0x44DC:
 903         case 0x44E0:
 904         case 0x44E4:
 905         case 0x44E8:
 906         case 0x44EC:
 907         case 0x44F0:
 908         case 0x44F4:
 909         case 0x44F8:
 910         case 0x44FC:
 911                 /* TX_FORMAT1_[0-15] */
 912                 i = (reg - 0x44C0) >> 2;
 913                 tmp = (idx_value >> 25) & 0x3;
 914                 track->textures[i].tex_coord_type = tmp;
 915                 switch ((idx_value & 0x1F)) {
 916                 case R300_TX_FORMAT_X8:
 917                 case R300_TX_FORMAT_Y4X4:
 918                 case R300_TX_FORMAT_Z3Y3X2:
 919                         track->textures[i].cpp = 1;
 920                         track->textures[i].compress_format = R100_TRACK_COMP_NONE;
 921                         break;
 922                 case R300_TX_FORMAT_X16:
 923                 case R300_TX_FORMAT_FL_I16:
 924                 case R300_TX_FORMAT_Y8X8:
 925                 case R300_TX_FORMAT_Z5Y6X5:
 926                 case R300_TX_FORMAT_Z6Y5X5:
 927                 case R300_TX_FORMAT_W4Z4Y4X4:
 928                 case R300_TX_FORMAT_W1Z5Y5X5:
 929                 case R300_TX_FORMAT_D3DMFT_CxV8U8:
 930                 case R300_TX_FORMAT_B8G8_B8G8:
 931                 case R300_TX_FORMAT_G8R8_G8B8:
 932                         track->textures[i].cpp = 2;
 933                         track->textures[i].compress_format = R100_TRACK_COMP_NONE;
 934                         break;
 935                 case R300_TX_FORMAT_Y16X16:
 936                 case R300_TX_FORMAT_FL_I16A16:
 937                 case R300_TX_FORMAT_Z11Y11X10:
 938                 case R300_TX_FORMAT_Z10Y11X11:
 939                 case R300_TX_FORMAT_W8Z8Y8X8:
 940                 case R300_TX_FORMAT_W2Z10Y10X10:
 941                 case 0x17:
 942                 case R300_TX_FORMAT_FL_I32:
 943                 case 0x1e:
 944                         track->textures[i].cpp = 4;
 945                         track->textures[i].compress_format = R100_TRACK_COMP_NONE;
 946                         break;
 947                 case R300_TX_FORMAT_W16Z16Y16X16:
 948                 case R300_TX_FORMAT_FL_R16G16B16A16:
 949                 case R300_TX_FORMAT_FL_I32A32:
 950                         track->textures[i].cpp = 8;
 951                         track->textures[i].compress_format = R100_TRACK_COMP_NONE;
 952                         break;
 953                 case R300_TX_FORMAT_FL_R32G32B32A32:
 954                         track->textures[i].cpp = 16;
 955                         track->textures[i].compress_format = R100_TRACK_COMP_NONE;
 956                         break;
 957                 case R300_TX_FORMAT_DXT1:
 958                         track->textures[i].cpp = 1;
 959                         track->textures[i].compress_format = R100_TRACK_COMP_DXT1;
 960                         break;
 961                 case R300_TX_FORMAT_ATI2N:
 962                         if (p->rdev->family < CHIP_R420) {
 963                                 DRM_ERROR("Invalid texture format %u\n",
 964                                           (idx_value & 0x1F));
 965                                 return -EINVAL;
 966                         }
 967                         /* The same rules apply as for DXT3/5. */
 968                         fallthrough;
 969                 case R300_TX_FORMAT_DXT3:
 970                 case R300_TX_FORMAT_DXT5:
 971                         track->textures[i].cpp = 1;
 972                         track->textures[i].compress_format = R100_TRACK_COMP_DXT35;
 973                         break;
 974                 default:
 975                         DRM_ERROR("Invalid texture format %u\n",
 976                                   (idx_value & 0x1F));
 977                         return -EINVAL;
 978                 }
 979                 track->tex_dirty = true;
 980                 break;
 981         case 0x4400:
 982         case 0x4404:
 983         case 0x4408:
 984         case 0x440C:
 985         case 0x4410:
 986         case 0x4414:
 987         case 0x4418:
 988         case 0x441C:
 989         case 0x4420:
 990         case 0x4424:
 991         case 0x4428:
 992         case 0x442C:
 993         case 0x4430:
 994         case 0x4434:
 995         case 0x4438:
 996         case 0x443C:
 997                 /* TX_FILTER0_[0-15] */
 998                 i = (reg - 0x4400) >> 2;
 999                 tmp = idx_value & 0x7;
1000                 if (tmp == 2 || tmp == 4 || tmp == 6) {
1001                         track->textures[i].roundup_w = false;
1002                 }
1003                 tmp = (idx_value >> 3) & 0x7;
1004                 if (tmp == 2 || tmp == 4 || tmp == 6) {
1005                         track->textures[i].roundup_h = false;
1006                 }
1007                 track->tex_dirty = true;
1008                 break;
1009         case 0x4500:
1010         case 0x4504:
1011         case 0x4508:
1012         case 0x450C:
1013         case 0x4510:
1014         case 0x4514:
1015         case 0x4518:
1016         case 0x451C:
1017         case 0x4520:
1018         case 0x4524:
1019         case 0x4528:
1020         case 0x452C:
1021         case 0x4530:
1022         case 0x4534:
1023         case 0x4538:
1024         case 0x453C:
1025                 /* TX_FORMAT2_[0-15] */
1026                 i = (reg - 0x4500) >> 2;
1027                 tmp = idx_value & 0x3FFF;
1028                 track->textures[i].pitch = tmp + 1;
1029                 if (p->rdev->family >= CHIP_RV515) {
1030                         tmp = ((idx_value >> 15) & 1) << 11;
1031                         track->textures[i].width_11 = tmp;
1032                         tmp = ((idx_value >> 16) & 1) << 11;
1033                         track->textures[i].height_11 = tmp;
1034
1035                         /* ATI1N */
1036                         if (idx_value & (1 << 14)) {
1037                                 /* The same rules apply as for DXT1. */
1038                                 track->textures[i].compress_format =
1039                                         R100_TRACK_COMP_DXT1;
1040                         }
1041                 } else if (idx_value & (1 << 14)) {
1042                         DRM_ERROR("Forbidden bit TXFORMAT_MSB\n");
1043                         return -EINVAL;
1044                 }
1045                 track->tex_dirty = true;
1046                 break;
1047         case 0x4480:
1048         case 0x4484:
1049         case 0x4488:
1050         case 0x448C:
1051         case 0x4490:
1052         case 0x4494:
1053         case 0x4498:
1054         case 0x449C:
1055         case 0x44A0:
1056         case 0x44A4:
1057         case 0x44A8:
1058         case 0x44AC:
1059         case 0x44B0:
1060         case 0x44B4:
1061         case 0x44B8:
1062         case 0x44BC:
1063                 /* TX_FORMAT0_[0-15] */
1064                 i = (reg - 0x4480) >> 2;
1065                 tmp = idx_value & 0x7FF;
1066                 track->textures[i].width = tmp + 1;
1067                 tmp = (idx_value >> 11) & 0x7FF;
1068                 track->textures[i].height = tmp + 1;
1069                 tmp = (idx_value >> 26) & 0xF;
1070                 track->textures[i].num_levels = tmp;
1071                 tmp = idx_value & (1 << 31);
1072                 track->textures[i].use_pitch = !!tmp;
1073                 tmp = (idx_value >> 22) & 0xF;
1074                 track->textures[i].txdepth = tmp;
1075                 track->tex_dirty = true;
1076                 break;
1077         case R300_ZB_ZPASS_ADDR:
1078                 r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1079                 if (r) {
1080                         DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1081                                         idx, reg);
1082                         radeon_cs_dump_packet(p, pkt);
1083                         return r;
1084                 }
1085                 ib[idx] = idx_value + ((u32)reloc->gpu_offset);
1086                 break;
1087         case 0x4e0c:
1088                 /* RB3D_COLOR_CHANNEL_MASK */
1089                 track->color_channel_mask = idx_value;
1090                 track->cb_dirty = true;
1091                 break;
1092         case 0x43a4:
1093                 /* SC_HYPERZ_EN */
1094                 /* r300c emits this register - we need to disable hyperz for it
1095                  * without complaining */
1096                 if (p->rdev->hyperz_filp != p->filp) {
1097                         if (idx_value & 0x1)
1098                                 ib[idx] = idx_value & ~1;
1099                 }
1100                 break;
1101         case 0x4f1c:
1102                 /* ZB_BW_CNTL */
1103                 track->zb_cb_clear = !!(idx_value & (1 << 5));
1104                 track->cb_dirty = true;
1105                 track->zb_dirty = true;
1106                 if (p->rdev->hyperz_filp != p->filp) {
1107                         if (idx_value & (R300_HIZ_ENABLE |
1108                                          R300_RD_COMP_ENABLE |
1109                                          R300_WR_COMP_ENABLE |
1110                                          R300_FAST_FILL_ENABLE))
1111                                 goto fail;
1112                 }
1113                 break;
1114         case 0x4e04:
1115                 /* RB3D_BLENDCNTL */
1116                 track->blend_read_enable = !!(idx_value & (1 << 2));
1117                 track->cb_dirty = true;
1118                 break;
1119         case R300_RB3D_AARESOLVE_OFFSET:
1120                 r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1121                 if (r) {
1122                         DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1123                                   idx, reg);
1124                         radeon_cs_dump_packet(p, pkt);
1125                         return r;
1126                 }
1127                 track->aa.robj = reloc->robj;
1128                 track->aa.offset = idx_value;
1129                 track->aa_dirty = true;
1130                 ib[idx] = idx_value + ((u32)reloc->gpu_offset);
1131                 break;
1132         case R300_RB3D_AARESOLVE_PITCH:
1133                 track->aa.pitch = idx_value & 0x3FFE;
1134                 track->aa_dirty = true;
1135                 break;
1136         case R300_RB3D_AARESOLVE_CTL:
1137                 track->aaresolve = idx_value & 0x1;
1138                 track->aa_dirty = true;
1139                 break;
1140         case 0x4f30: /* ZB_MASK_OFFSET */
1141         case 0x4f34: /* ZB_ZMASK_PITCH */
1142         case 0x4f44: /* ZB_HIZ_OFFSET */
1143         case 0x4f54: /* ZB_HIZ_PITCH */
1144                 if (idx_value && (p->rdev->hyperz_filp != p->filp))
1145                         goto fail;
1146                 break;
1147         case 0x4028:
1148                 if (idx_value && (p->rdev->hyperz_filp != p->filp))
1149                         goto fail;
1150                 /* GB_Z_PEQ_CONFIG */
1151                 if (p->rdev->family >= CHIP_RV350)
1152                         break;
1153                 goto fail;
1154                 break;
1155         case 0x4be8:
1156                 /* valid register only on RV530 */
1157                 if (p->rdev->family == CHIP_RV530)
1158                         break;
1159                 fallthrough;
1160                 /* fallthrough do not move */
1161         default:
1162                 goto fail;
1163         }
1164         return 0;
1165 fail:
1166         pr_err("Forbidden register 0x%04X in cs at %d (val=%08x)\n",
1167                reg, idx, idx_value);
1168         return -EINVAL;
1169 }
1170
1171 static int r300_packet3_check(struct radeon_cs_parser *p,
1172                               struct radeon_cs_packet *pkt)
1173 {
1174         struct radeon_bo_list *reloc;
1175         struct r100_cs_track *track;
1176         volatile uint32_t *ib;
1177         unsigned idx;
1178         int r;
1179
1180         ib = p->ib.ptr;
1181         idx = pkt->idx + 1;
1182         track = (struct r100_cs_track *)p->track;
1183         switch (pkt->opcode) {
1184         case PACKET3_3D_LOAD_VBPNTR:
1185                 r = r100_packet3_load_vbpntr(p, pkt, idx);
1186                 if (r)
1187                         return r;
1188                 break;
1189         case PACKET3_INDX_BUFFER:
1190                 r = radeon_cs_packet_next_reloc(p, &reloc, 0);
1191                 if (r) {
1192                         DRM_ERROR("No reloc for packet3 %d\n", pkt->opcode);
1193                         radeon_cs_dump_packet(p, pkt);
1194                         return r;
1195                 }
1196                 ib[idx+1] = radeon_get_ib_value(p, idx + 1) + ((u32)reloc->gpu_offset);
1197                 r = r100_cs_track_check_pkt3_indx_buffer(p, pkt, reloc->robj);
1198                 if (r) {
1199                         return r;
1200                 }
1201                 break;
1202         /* Draw packet */
1203         case PACKET3_3D_DRAW_IMMD:
1204                 /* Number of dwords is vtx_size * (num_vertices - 1)
1205                  * PRIM_WALK must be equal to 3 vertex data in embedded
1206                  * in cmd stream */
1207                 if (((radeon_get_ib_value(p, idx + 1) >> 4) & 0x3) != 3) {
1208                         DRM_ERROR("PRIM_WALK must be 3 for IMMD draw\n");
1209                         return -EINVAL;
1210                 }
1211                 track->vap_vf_cntl = radeon_get_ib_value(p, idx + 1);
1212                 track->immd_dwords = pkt->count - 1;
1213                 r = r100_cs_track_check(p->rdev, track);
1214                 if (r) {
1215                         return r;
1216                 }
1217                 break;
1218         case PACKET3_3D_DRAW_IMMD_2:
1219                 /* Number of dwords is vtx_size * (num_vertices - 1)
1220                  * PRIM_WALK must be equal to 3 vertex data in embedded
1221                  * in cmd stream */
1222                 if (((radeon_get_ib_value(p, idx) >> 4) & 0x3) != 3) {
1223                         DRM_ERROR("PRIM_WALK must be 3 for IMMD draw\n");
1224                         return -EINVAL;
1225                 }
1226                 track->vap_vf_cntl = radeon_get_ib_value(p, idx);
1227                 track->immd_dwords = pkt->count;
1228                 r = r100_cs_track_check(p->rdev, track);
1229                 if (r) {
1230                         return r;
1231                 }
1232                 break;
1233         case PACKET3_3D_DRAW_VBUF:
1234                 track->vap_vf_cntl = radeon_get_ib_value(p, idx + 1);
1235                 r = r100_cs_track_check(p->rdev, track);
1236                 if (r) {
1237                         return r;
1238                 }
1239                 break;
1240         case PACKET3_3D_DRAW_VBUF_2:
1241                 track->vap_vf_cntl = radeon_get_ib_value(p, idx);
1242                 r = r100_cs_track_check(p->rdev, track);
1243                 if (r) {
1244                         return r;
1245                 }
1246                 break;
1247         case PACKET3_3D_DRAW_INDX:
1248                 track->vap_vf_cntl = radeon_get_ib_value(p, idx + 1);
1249                 r = r100_cs_track_check(p->rdev, track);
1250                 if (r) {
1251                         return r;
1252                 }
1253                 break;
1254         case PACKET3_3D_DRAW_INDX_2:
1255                 track->vap_vf_cntl = radeon_get_ib_value(p, idx);
1256                 r = r100_cs_track_check(p->rdev, track);
1257                 if (r) {
1258                         return r;
1259                 }
1260                 break;
1261         case PACKET3_3D_CLEAR_HIZ:
1262         case PACKET3_3D_CLEAR_ZMASK:
1263                 if (p->rdev->hyperz_filp != p->filp)
1264                         return -EINVAL;
1265                 break;
1266         case PACKET3_3D_CLEAR_CMASK:
1267                 if (p->rdev->cmask_filp != p->filp)
1268                         return -EINVAL;
1269                 break;
1270         case PACKET3_NOP:
1271                 break;
1272         default:
1273                 DRM_ERROR("Packet3 opcode %x not supported\n", pkt->opcode);
1274                 return -EINVAL;
1275         }
1276         return 0;
1277 }
1278
1279 int r300_cs_parse(struct radeon_cs_parser *p)
1280 {
1281         struct radeon_cs_packet pkt;
1282         struct r100_cs_track *track;
1283         int r;
1284
1285         track = kzalloc(sizeof(*track), GFP_KERNEL);
1286         if (track == NULL)
1287                 return -ENOMEM;
1288         r100_cs_track_clear(p->rdev, track);
1289         p->track = track;
1290         do {
1291                 r = radeon_cs_packet_parse(p, &pkt, p->idx);
1292                 if (r) {
1293                         return r;
1294                 }
1295                 p->idx += pkt.count + 2;
1296                 switch (pkt.type) {
1297                 case RADEON_PACKET_TYPE0:
1298                         r = r100_cs_parse_packet0(p, &pkt,
1299                                                   p->rdev->config.r300.reg_safe_bm,
1300                                                   p->rdev->config.r300.reg_safe_bm_size,
1301                                                   &r300_packet0_check);
1302                         break;
1303                 case RADEON_PACKET_TYPE2:
1304                         break;
1305                 case RADEON_PACKET_TYPE3:
1306                         r = r300_packet3_check(p, &pkt);
1307                         break;
1308                 default:
1309                         DRM_ERROR("Unknown packet type %d !\n", pkt.type);
1310                         return -EINVAL;
1311                 }
1312                 if (r) {
1313                         return r;
1314                 }
1315         } while (p->idx < p->chunk_ib->length_dw);
1316         return 0;
1317 }
1318
1319 void r300_set_reg_safe(struct radeon_device *rdev)
1320 {
1321         rdev->config.r300.reg_safe_bm = r300_reg_safe_bm;
1322         rdev->config.r300.reg_safe_bm_size = ARRAY_SIZE(r300_reg_safe_bm);
1323 }
1324
1325 void r300_mc_program(struct radeon_device *rdev)
1326 {
1327         struct r100_mc_save save;
1328
1329         r100_debugfs_mc_info_init(rdev);
1330
1331         /* Stops all mc clients */
1332         r100_mc_stop(rdev, &save);
1333         if (rdev->flags & RADEON_IS_AGP) {
1334                 WREG32(R_00014C_MC_AGP_LOCATION,
1335                         S_00014C_MC_AGP_START(rdev->mc.gtt_start >> 16) |
1336                         S_00014C_MC_AGP_TOP(rdev->mc.gtt_end >> 16));
1337                 WREG32(R_000170_AGP_BASE, lower_32_bits(rdev->mc.agp_base));
1338                 WREG32(R_00015C_AGP_BASE_2,
1339                         upper_32_bits(rdev->mc.agp_base) & 0xff);
1340         } else {
1341                 WREG32(R_00014C_MC_AGP_LOCATION, 0x0FFFFFFF);
1342                 WREG32(R_000170_AGP_BASE, 0);
1343                 WREG32(R_00015C_AGP_BASE_2, 0);
1344         }
1345         /* Wait for mc idle */
1346         if (r300_mc_wait_for_idle(rdev))
1347                 DRM_INFO("Failed to wait MC idle before programming MC.\n");
1348         /* Program MC, should be a 32bits limited address space */
1349         WREG32(R_000148_MC_FB_LOCATION,
1350                 S_000148_MC_FB_START(rdev->mc.vram_start >> 16) |
1351                 S_000148_MC_FB_TOP(rdev->mc.vram_end >> 16));
1352         r100_mc_resume(rdev, &save);
1353 }
1354
1355 void r300_clock_startup(struct radeon_device *rdev)
1356 {
1357         u32 tmp;
1358
1359         if (radeon_dynclks != -1 && radeon_dynclks)
1360                 radeon_legacy_set_clock_gating(rdev, 1);
1361         /* We need to force on some of the block */
1362         tmp = RREG32_PLL(R_00000D_SCLK_CNTL);
1363         tmp |= S_00000D_FORCE_CP(1) | S_00000D_FORCE_VIP(1);
1364         if ((rdev->family == CHIP_RV350) || (rdev->family == CHIP_RV380))
1365                 tmp |= S_00000D_FORCE_VAP(1);
1366         WREG32_PLL(R_00000D_SCLK_CNTL, tmp);
1367 }
1368
1369 static int r300_startup(struct radeon_device *rdev)
1370 {
1371         int r;
1372
1373         /* set common regs */
1374         r100_set_common_regs(rdev);
1375         /* program mc */
1376         r300_mc_program(rdev);
1377         /* Resume clock */
1378         r300_clock_startup(rdev);
1379         /* Initialize GPU configuration (# pipes, ...) */
1380         r300_gpu_init(rdev);
1381         /* Initialize GART (initialize after TTM so we can allocate
1382          * memory through TTM but finalize after TTM) */
1383         if (rdev->flags & RADEON_IS_PCIE) {
1384                 r = rv370_pcie_gart_enable(rdev);
1385                 if (r)
1386                         return r;
1387         }
1388
1389         if (rdev->family == CHIP_R300 ||
1390             rdev->family == CHIP_R350 ||
1391             rdev->family == CHIP_RV350)
1392                 r100_enable_bm(rdev);
1393
1394         if (rdev->flags & RADEON_IS_PCI) {
1395                 r = r100_pci_gart_enable(rdev);
1396                 if (r)
1397                         return r;
1398         }
1399
1400         /* allocate wb buffer */
1401         r = radeon_wb_init(rdev);
1402         if (r)
1403                 return r;
1404
1405         r = radeon_fence_driver_start_ring(rdev, RADEON_RING_TYPE_GFX_INDEX);
1406         if (r) {
1407                 dev_err(rdev->dev, "failed initializing CP fences (%d).\n", r);
1408                 return r;
1409         }
1410
1411         /* Enable IRQ */
1412         if (!rdev->irq.installed) {
1413                 r = radeon_irq_kms_init(rdev);
1414                 if (r)
1415                         return r;
1416         }
1417
1418         r100_irq_set(rdev);
1419         rdev->config.r300.hdp_cntl = RREG32(RADEON_HOST_PATH_CNTL);
1420         /* 1M ring buffer */
1421         r = r100_cp_init(rdev, 1024 * 1024);
1422         if (r) {
1423                 dev_err(rdev->dev, "failed initializing CP (%d).\n", r);
1424                 return r;
1425         }
1426
1427         r = radeon_ib_pool_init(rdev);
1428         if (r) {
1429                 dev_err(rdev->dev, "IB initialization failed (%d).\n", r);
1430                 return r;
1431         }
1432
1433         return 0;
1434 }
1435
1436 int r300_resume(struct radeon_device *rdev)
1437 {
1438         int r;
1439
1440         /* Make sur GART are not working */
1441         if (rdev->flags & RADEON_IS_PCIE)
1442                 rv370_pcie_gart_disable(rdev);
1443         if (rdev->flags & RADEON_IS_PCI)
1444                 r100_pci_gart_disable(rdev);
1445         /* Resume clock before doing reset */
1446         r300_clock_startup(rdev);
1447         /* Reset gpu before posting otherwise ATOM will enter infinite loop */
1448         if (radeon_asic_reset(rdev)) {
1449                 dev_warn(rdev->dev, "GPU reset failed ! (0xE40=0x%08X, 0x7C0=0x%08X)\n",
1450                         RREG32(R_000E40_RBBM_STATUS),
1451                         RREG32(R_0007C0_CP_STAT));
1452         }
1453         /* post */
1454         radeon_combios_asic_init(rdev->ddev);
1455         /* Resume clock after posting */
1456         r300_clock_startup(rdev);
1457         /* Initialize surface registers */
1458         radeon_surface_init(rdev);
1459
1460         rdev->accel_working = true;
1461         r = r300_startup(rdev);
1462         if (r) {
1463                 rdev->accel_working = false;
1464         }
1465         return r;
1466 }
1467
1468 int r300_suspend(struct radeon_device *rdev)
1469 {
1470         radeon_pm_suspend(rdev);
1471         r100_cp_disable(rdev);
1472         radeon_wb_disable(rdev);
1473         r100_irq_disable(rdev);
1474         if (rdev->flags & RADEON_IS_PCIE)
1475                 rv370_pcie_gart_disable(rdev);
1476         if (rdev->flags & RADEON_IS_PCI)
1477                 r100_pci_gart_disable(rdev);
1478         return 0;
1479 }
1480
1481 void r300_fini(struct radeon_device *rdev)
1482 {
1483         radeon_pm_fini(rdev);
1484         r100_cp_fini(rdev);
1485         radeon_wb_fini(rdev);
1486         radeon_ib_pool_fini(rdev);
1487         radeon_gem_fini(rdev);
1488         if (rdev->flags & RADEON_IS_PCIE)
1489                 rv370_pcie_gart_fini(rdev);
1490         if (rdev->flags & RADEON_IS_PCI)
1491                 r100_pci_gart_fini(rdev);
1492         radeon_agp_fini(rdev);
1493         radeon_irq_kms_fini(rdev);
1494         radeon_fence_driver_fini(rdev);
1495         radeon_bo_fini(rdev);
1496         radeon_atombios_fini(rdev);
1497         kfree(rdev->bios);
1498         rdev->bios = NULL;
1499 }
1500
1501 int r300_init(struct radeon_device *rdev)
1502 {
1503         int r;
1504
1505         /* Disable VGA */
1506         r100_vga_render_disable(rdev);
1507         /* Initialize scratch registers */
1508         radeon_scratch_init(rdev);
1509         /* Initialize surface registers */
1510         radeon_surface_init(rdev);
1511         /* TODO: disable VGA need to use VGA request */
1512         /* restore some register to sane defaults */
1513         r100_restore_sanity(rdev);
1514         /* BIOS*/
1515         if (!radeon_get_bios(rdev)) {
1516                 if (ASIC_IS_AVIVO(rdev))
1517                         return -EINVAL;
1518         }
1519         if (rdev->is_atom_bios) {
1520                 dev_err(rdev->dev, "Expecting combios for RS400/RS480 GPU\n");
1521                 return -EINVAL;
1522         } else {
1523                 r = radeon_combios_init(rdev);
1524                 if (r)
1525                         return r;
1526         }
1527         /* Reset gpu before posting otherwise ATOM will enter infinite loop */
1528         if (radeon_asic_reset(rdev)) {
1529                 dev_warn(rdev->dev,
1530                         "GPU reset failed ! (0xE40=0x%08X, 0x7C0=0x%08X)\n",
1531                         RREG32(R_000E40_RBBM_STATUS),
1532                         RREG32(R_0007C0_CP_STAT));
1533         }
1534         /* check if cards are posted or not */
1535         if (radeon_boot_test_post_card(rdev) == false)
1536                 return -EINVAL;
1537         /* Set asic errata */
1538         r300_errata(rdev);
1539         /* Initialize clocks */
1540         radeon_get_clock_info(rdev->ddev);
1541         /* initialize AGP */
1542         if (rdev->flags & RADEON_IS_AGP) {
1543                 r = radeon_agp_init(rdev);
1544                 if (r) {
1545                         radeon_agp_disable(rdev);
1546                 }
1547         }
1548         /* initialize memory controller */
1549         r300_mc_init(rdev);
1550         /* Fence driver */
1551         radeon_fence_driver_init(rdev);
1552         /* Memory manager */
1553         r = radeon_bo_init(rdev);
1554         if (r)
1555                 return r;
1556         if (rdev->flags & RADEON_IS_PCIE) {
1557                 r = rv370_pcie_gart_init(rdev);
1558                 if (r)
1559                         return r;
1560         }
1561         if (rdev->flags & RADEON_IS_PCI) {
1562                 r = r100_pci_gart_init(rdev);
1563                 if (r)
1564                         return r;
1565         }
1566         r300_set_reg_safe(rdev);
1567
1568         /* Initialize power management */
1569         radeon_pm_init(rdev);
1570
1571         rdev->accel_working = true;
1572         r = r300_startup(rdev);
1573         if (r) {
1574                 /* Something went wrong with the accel init, so stop accel */
1575                 dev_err(rdev->dev, "Disabling GPU acceleration\n");
1576                 r100_cp_fini(rdev);
1577                 radeon_wb_fini(rdev);
1578                 radeon_ib_pool_fini(rdev);
1579                 radeon_irq_kms_fini(rdev);
1580                 if (rdev->flags & RADEON_IS_PCIE)
1581                         rv370_pcie_gart_fini(rdev);
1582                 if (rdev->flags & RADEON_IS_PCI)
1583                         r100_pci_gart_fini(rdev);
1584                 radeon_agp_fini(rdev);
1585                 rdev->accel_working = false;
1586         }
1587         return 0;
1588 }