drivers/gpu/drm/radeon/r100.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_drm.h"
  32 #include "radeon_microcode.h"
  33 #include "radeon_reg.h"
  34 #include "radeon.h"
  35
  36 /* This files gather functions specifics to:
  37  * r100,rv100,rs100,rv200,rs200,r200,rv250,rs300,rv280
  38  *
  39  * Some of these functions might be used by newer ASICs.
  40  */
  41 void r100_hdp_reset(struct radeon_device *rdev);
  42 void r100_gpu_init(struct radeon_device *rdev);
  43 int r100_gui_wait_for_idle(struct radeon_device *rdev);
  44 int r100_mc_wait_for_idle(struct radeon_device *rdev);
  45 void r100_gpu_wait_for_vsync(struct radeon_device *rdev);
  46 void r100_gpu_wait_for_vsync2(struct radeon_device *rdev);
  47 int r100_debugfs_mc_info_init(struct radeon_device *rdev);
  48
  49
  50 /*
  51  * PCI GART
  52  */
  53 void r100_pci_gart_tlb_flush(struct radeon_device *rdev)
  54 {
  55         /* TODO: can we do somethings here ? */
  56         /* It seems hw only cache one entry so we should discard this
  57          * entry otherwise if first GPU GART read hit this entry it
  58          * could end up in wrong address. */
  59 }
  60
  61 int r100_pci_gart_enable(struct radeon_device *rdev)
  62 {
  63         uint32_t tmp;
  64         int r;
  65
  66         /* Initialize common gart structure */
  67         r = radeon_gart_init(rdev);
  68         if (r) {
  69                 return r;
  70         }
  71         if (rdev->gart.table.ram.ptr == NULL) {
  72                 rdev->gart.table_size = rdev->gart.num_gpu_pages * 4;
  73                 r = radeon_gart_table_ram_alloc(rdev);
  74                 if (r) {
  75                         return r;
  76                 }
  77         }
  78         /* discard memory request outside of configured range */
  79         tmp = RREG32(RADEON_AIC_CNTL) | RADEON_DIS_OUT_OF_PCI_GART_ACCESS;
  80         WREG32(RADEON_AIC_CNTL, tmp);
  81         /* set address range for PCI address translate */
  82         WREG32(RADEON_AIC_LO_ADDR, rdev->mc.gtt_location);
  83         tmp = rdev->mc.gtt_location + rdev->mc.gtt_size - 1;
  84         WREG32(RADEON_AIC_HI_ADDR, tmp);
  85         /* Enable bus mastering */
  86         tmp = RREG32(RADEON_BUS_CNTL) & ~RADEON_BUS_MASTER_DIS;
  87         WREG32(RADEON_BUS_CNTL, tmp);
  88         /* set PCI GART page-table base address */
  89         WREG32(RADEON_AIC_PT_BASE, rdev->gart.table_addr);
  90         tmp = RREG32(RADEON_AIC_CNTL) | RADEON_PCIGART_TRANSLATE_EN;
  91         WREG32(RADEON_AIC_CNTL, tmp);
  92         r100_pci_gart_tlb_flush(rdev);
  93         rdev->gart.ready = true;
  94         return 0;
  95 }
  96
  97 void r100_pci_gart_disable(struct radeon_device *rdev)
  98 {
  99         uint32_t tmp;
 100
 101         /* discard memory request outside of configured range */
 102         tmp = RREG32(RADEON_AIC_CNTL) | RADEON_DIS_OUT_OF_PCI_GART_ACCESS;
 103         WREG32(RADEON_AIC_CNTL, tmp & ~RADEON_PCIGART_TRANSLATE_EN);
 104         WREG32(RADEON_AIC_LO_ADDR, 0);
 105         WREG32(RADEON_AIC_HI_ADDR, 0);
 106 }
 107
 108 int r100_pci_gart_set_page(struct radeon_device *rdev, int i, uint64_t addr)
 109 {
 110         if (i < 0 || i > rdev->gart.num_gpu_pages) {
 111                 return -EINVAL;
 112         }
 113         rdev->gart.table.ram.ptr[i] = cpu_to_le32(lower_32_bits(addr));
 114         return 0;
 115 }
 116
 117 int r100_gart_enable(struct radeon_device *rdev)
 118 {
 119         if (rdev->flags & RADEON_IS_AGP) {
 120                 r100_pci_gart_disable(rdev);
 121                 return 0;
 122         }
 123         return r100_pci_gart_enable(rdev);
 124 }
 125
 126
 127 /*
 128  * MC
 129  */
 130 void r100_mc_disable_clients(struct radeon_device *rdev)
 131 {
 132         uint32_t ov0_scale_cntl, crtc_ext_cntl, crtc_gen_cntl, crtc2_gen_cntl;
 133
 134         /* FIXME: is this function correct for rs100,rs200,rs300 ? */
 135         if (r100_gui_wait_for_idle(rdev)) {
 136                 printk(KERN_WARNING "Failed to wait GUI idle while "
 137                        "programming pipes. Bad things might happen.\n");
 138         }
 139
 140         /* stop display and memory access */
 141         ov0_scale_cntl = RREG32(RADEON_OV0_SCALE_CNTL);
 142         WREG32(RADEON_OV0_SCALE_CNTL, ov0_scale_cntl & ~RADEON_SCALER_ENABLE);
 143         crtc_ext_cntl = RREG32(RADEON_CRTC_EXT_CNTL);
 144         WREG32(RADEON_CRTC_EXT_CNTL, crtc_ext_cntl | RADEON_CRTC_DISPLAY_DIS);
 145         crtc_gen_cntl = RREG32(RADEON_CRTC_GEN_CNTL);
 146
 147         r100_gpu_wait_for_vsync(rdev);
 148
 149         WREG32(RADEON_CRTC_GEN_CNTL,
 150                (crtc_gen_cntl & ~(RADEON_CRTC_CUR_EN | RADEON_CRTC_ICON_EN)) |
 151                RADEON_CRTC_DISP_REQ_EN_B | RADEON_CRTC_EXT_DISP_EN);
 152
 153         if (!(rdev->flags & RADEON_SINGLE_CRTC)) {
 154                 crtc2_gen_cntl = RREG32(RADEON_CRTC2_GEN_CNTL);
 155
 156                 r100_gpu_wait_for_vsync2(rdev);
 157                 WREG32(RADEON_CRTC2_GEN_CNTL,
 158                        (crtc2_gen_cntl &
 159                         ~(RADEON_CRTC2_CUR_EN | RADEON_CRTC2_ICON_EN)) |
 160                        RADEON_CRTC2_DISP_REQ_EN_B);
 161         }
 162
 163         udelay(500);
 164 }
 165
 166 void r100_mc_setup(struct radeon_device *rdev)
 167 {
 168         uint32_t tmp;
 169         int r;
 170
 171         r = r100_debugfs_mc_info_init(rdev);
 172         if (r) {
 173                 DRM_ERROR("Failed to register debugfs file for R100 MC !\n");
 174         }
 175         /* Write VRAM size in case we are limiting it */
 176         WREG32(RADEON_CONFIG_MEMSIZE, rdev->mc.real_vram_size);
 177         /* Novell bug 204882 for RN50/M6/M7 with 8/16/32MB VRAM,
 178          * if the aperture is 64MB but we have 32MB VRAM
 179          * we report only 32MB VRAM but we have to set MC_FB_LOCATION
 180          * to 64MB, otherwise the gpu accidentially dies */
 181         tmp = rdev->mc.vram_location + rdev->mc.mc_vram_size - 1;
 182         tmp = REG_SET(RADEON_MC_FB_TOP, tmp >> 16);
 183         tmp |= REG_SET(RADEON_MC_FB_START, rdev->mc.vram_location >> 16);
 184         WREG32(RADEON_MC_FB_LOCATION, tmp);
 185
 186         /* Enable bus mastering */
 187         tmp = RREG32(RADEON_BUS_CNTL) & ~RADEON_BUS_MASTER_DIS;
 188         WREG32(RADEON_BUS_CNTL, tmp);
 189
 190         if (rdev->flags & RADEON_IS_AGP) {
 191                 tmp = rdev->mc.gtt_location + rdev->mc.gtt_size - 1;
 192                 tmp = REG_SET(RADEON_MC_AGP_TOP, tmp >> 16);
 193                 tmp |= REG_SET(RADEON_MC_AGP_START, rdev->mc.gtt_location >> 16);
 194                 WREG32(RADEON_MC_AGP_LOCATION, tmp);
 195                 WREG32(RADEON_AGP_BASE, rdev->mc.agp_base);
 196         } else {
 197                 WREG32(RADEON_MC_AGP_LOCATION, 0x0FFFFFFF);
 198                 WREG32(RADEON_AGP_BASE, 0);
 199         }
 200
 201         tmp = RREG32(RADEON_HOST_PATH_CNTL) & RADEON_HDP_APER_CNTL;
 202         tmp |= (7 << 28);
 203         WREG32(RADEON_HOST_PATH_CNTL, tmp | RADEON_HDP_SOFT_RESET | RADEON_HDP_READ_BUFFER_INVALIDATE);
 204         (void)RREG32(RADEON_HOST_PATH_CNTL);
 205         WREG32(RADEON_HOST_PATH_CNTL, tmp);
 206         (void)RREG32(RADEON_HOST_PATH_CNTL);
 207 }
 208
 209 int r100_mc_init(struct radeon_device *rdev)
 210 {
 211         int r;
 212
 213         if (r100_debugfs_rbbm_init(rdev)) {
 214                 DRM_ERROR("Failed to register debugfs file for RBBM !\n");
 215         }
 216
 217         r100_gpu_init(rdev);
 218         /* Disable gart which also disable out of gart access */
 219         r100_pci_gart_disable(rdev);
 220
 221         /* Setup GPU memory space */
 222         rdev->mc.gtt_location = 0xFFFFFFFFUL;
 223         if (rdev->flags & RADEON_IS_AGP) {
 224                 r = radeon_agp_init(rdev);
 225                 if (r) {
 226                         printk(KERN_WARNING "[drm] Disabling AGP\n");
 227                         rdev->flags &= ~RADEON_IS_AGP;
 228                         rdev->mc.gtt_size = radeon_gart_size * 1024 * 1024;
 229                 } else {
 230                         rdev->mc.gtt_location = rdev->mc.agp_base;
 231                 }
 232         }
 233         r = radeon_mc_setup(rdev);
 234         if (r) {
 235                 return r;
 236         }
 237
 238         r100_mc_disable_clients(rdev);
 239         if (r100_mc_wait_for_idle(rdev)) {
 240                 printk(KERN_WARNING "Failed to wait MC idle while "
 241                        "programming pipes. Bad things might happen.\n");
 242         }
 243
 244         r100_mc_setup(rdev);
 245         return 0;
 246 }
 247
 248 void r100_mc_fini(struct radeon_device *rdev)
 249 {
 250         r100_pci_gart_disable(rdev);
 251         radeon_gart_table_ram_free(rdev);
 252         radeon_gart_fini(rdev);
 253 }
 254
 255
 256 /*
 257  * Fence emission
 258  */
 259 void r100_fence_ring_emit(struct radeon_device *rdev,
 260                           struct radeon_fence *fence)
 261 {
 262         /* Who ever call radeon_fence_emit should call ring_lock and ask
 263          * for enough space (today caller are ib schedule and buffer move) */
 264         /* Wait until IDLE & CLEAN */
 265         radeon_ring_write(rdev, PACKET0(0x1720, 0));
 266         radeon_ring_write(rdev, (1 << 16) | (1 << 17));
 267         /* Emit fence sequence & fire IRQ */
 268         radeon_ring_write(rdev, PACKET0(rdev->fence_drv.scratch_reg, 0));
 269         radeon_ring_write(rdev, fence->seq);
 270         radeon_ring_write(rdev, PACKET0(RADEON_GEN_INT_STATUS, 0));
 271         radeon_ring_write(rdev, RADEON_SW_INT_FIRE);
 272 }
 273
 274
 275 /*
 276  * Writeback
 277  */
 278 int r100_wb_init(struct radeon_device *rdev)
 279 {
 280         int r;
 281
 282         if (rdev->wb.wb_obj == NULL) {
 283                 r = radeon_object_create(rdev, NULL, 4096,
 284                                          true,
 285                                          RADEON_GEM_DOMAIN_GTT,
 286                                          false, &rdev->wb.wb_obj);
 287                 if (r) {
 288                         DRM_ERROR("radeon: failed to create WB buffer (%d).\n", r);
 289                         return r;
 290                 }
 291                 r = radeon_object_pin(rdev->wb.wb_obj,
 292                                       RADEON_GEM_DOMAIN_GTT,
 293                                       &rdev->wb.gpu_addr);
 294                 if (r) {
 295                         DRM_ERROR("radeon: failed to pin WB buffer (%d).\n", r);
 296                         return r;
 297                 }
 298                 r = radeon_object_kmap(rdev->wb.wb_obj, (void **)&rdev->wb.wb);
 299                 if (r) {
 300                         DRM_ERROR("radeon: failed to map WB buffer (%d).\n", r);
 301                         return r;
 302                 }
 303         }
 304         WREG32(0x774, rdev->wb.gpu_addr);
 305         WREG32(0x70C, rdev->wb.gpu_addr + 1024);
 306         WREG32(0x770, 0xff);
 307         return 0;
 308 }
 309
 310 void r100_wb_fini(struct radeon_device *rdev)
 311 {
 312         if (rdev->wb.wb_obj) {
 313                 radeon_object_kunmap(rdev->wb.wb_obj);
 314                 radeon_object_unpin(rdev->wb.wb_obj);
 315                 radeon_object_unref(&rdev->wb.wb_obj);
 316                 rdev->wb.wb = NULL;
 317                 rdev->wb.wb_obj = NULL;
 318         }
 319 }
 320
 321 int r100_copy_blit(struct radeon_device *rdev,
 322                    uint64_t src_offset,
 323                    uint64_t dst_offset,
 324                    unsigned num_pages,
 325                    struct radeon_fence *fence)
 326 {
 327         uint32_t cur_pages;
 328         uint32_t stride_bytes = PAGE_SIZE;
 329         uint32_t pitch;
 330         uint32_t stride_pixels;
 331         unsigned ndw;
 332         int num_loops;
 333         int r = 0;
 334
 335         /* radeon limited to 16k stride */
 336         stride_bytes &= 0x3fff;
 337         /* radeon pitch is /64 */
 338         pitch = stride_bytes / 64;
 339         stride_pixels = stride_bytes / 4;
 340         num_loops = DIV_ROUND_UP(num_pages, 8191);
 341
 342         /* Ask for enough room for blit + flush + fence */
 343         ndw = 64 + (10 * num_loops);
 344         r = radeon_ring_lock(rdev, ndw);
 345         if (r) {
 346                 DRM_ERROR("radeon: moving bo (%d) asking for %u dw.\n", r, ndw);
 347                 return -EINVAL;
 348         }
 349         while (num_pages > 0) {
 350                 cur_pages = num_pages;
 351                 if (cur_pages > 8191) {
 352                         cur_pages = 8191;
 353                 }
 354                 num_pages -= cur_pages;
 355
 356                 /* pages are in Y direction - height
 357                    page width in X direction - width */
 358                 radeon_ring_write(rdev, PACKET3(PACKET3_BITBLT_MULTI, 8));
 359                 radeon_ring_write(rdev,
 360                                   RADEON_GMC_SRC_PITCH_OFFSET_CNTL |
 361                                   RADEON_GMC_DST_PITCH_OFFSET_CNTL |
 362                                   RADEON_GMC_SRC_CLIPPING |
 363                                   RADEON_GMC_DST_CLIPPING |
 364                                   RADEON_GMC_BRUSH_NONE |
 365                                   (RADEON_COLOR_FORMAT_ARGB8888 << 8) |
 366                                   RADEON_GMC_SRC_DATATYPE_COLOR |
 367                                   RADEON_ROP3_S |
 368                                   RADEON_DP_SRC_SOURCE_MEMORY |
 369                                   RADEON_GMC_CLR_CMP_CNTL_DIS |
 370                                   RADEON_GMC_WR_MSK_DIS);
 371                 radeon_ring_write(rdev, (pitch << 22) | (src_offset >> 10));
 372                 radeon_ring_write(rdev, (pitch << 22) | (dst_offset >> 10));
 373                 radeon_ring_write(rdev, (0x1fff) | (0x1fff << 16));
 374                 radeon_ring_write(rdev, 0);
 375                 radeon_ring_write(rdev, (0x1fff) | (0x1fff << 16));
 376                 radeon_ring_write(rdev, num_pages);
 377                 radeon_ring_write(rdev, num_pages);
 378                 radeon_ring_write(rdev, cur_pages | (stride_pixels << 16));
 379         }
 380         radeon_ring_write(rdev, PACKET0(RADEON_DSTCACHE_CTLSTAT, 0));
 381         radeon_ring_write(rdev, RADEON_RB2D_DC_FLUSH_ALL);
 382         radeon_ring_write(rdev, PACKET0(RADEON_WAIT_UNTIL, 0));
 383         radeon_ring_write(rdev,
 384                           RADEON_WAIT_2D_IDLECLEAN |
 385                           RADEON_WAIT_HOST_IDLECLEAN |
 386                           RADEON_WAIT_DMA_GUI_IDLE);
 387         if (fence) {
 388                 r = radeon_fence_emit(rdev, fence);
 389         }
 390         radeon_ring_unlock_commit(rdev);
 391         return r;
 392 }
 393
 394
 395 /*
 396  * CP
 397  */
 398 void r100_ring_start(struct radeon_device *rdev)
 399 {
 400         int r;
 401
 402         r = radeon_ring_lock(rdev, 2);
 403         if (r) {
 404                 return;
 405         }
 406         radeon_ring_write(rdev, PACKET0(RADEON_ISYNC_CNTL, 0));
 407         radeon_ring_write(rdev,
 408                           RADEON_ISYNC_ANY2D_IDLE3D |
 409                           RADEON_ISYNC_ANY3D_IDLE2D |
 410                           RADEON_ISYNC_WAIT_IDLEGUI |
 411                           RADEON_ISYNC_CPSCRATCH_IDLEGUI);
 412         radeon_ring_unlock_commit(rdev);
 413 }
 414
 415 static void r100_cp_load_microcode(struct radeon_device *rdev)
 416 {
 417         int i;
 418
 419         if (r100_gui_wait_for_idle(rdev)) {
 420                 printk(KERN_WARNING "Failed to wait GUI idle while "
 421                        "programming pipes. Bad things might happen.\n");
 422         }
 423
 424         WREG32(RADEON_CP_ME_RAM_ADDR, 0);
 425         if ((rdev->family == CHIP_R100) || (rdev->family == CHIP_RV100) ||
 426             (rdev->family == CHIP_RV200) || (rdev->family == CHIP_RS100) ||
 427             (rdev->family == CHIP_RS200)) {
 428                 DRM_INFO("Loading R100 Microcode\n");
 429                 for (i = 0; i < 256; i++) {
 430                         WREG32(RADEON_CP_ME_RAM_DATAH, R100_cp_microcode[i][1]);
 431                         WREG32(RADEON_CP_ME_RAM_DATAL, R100_cp_microcode[i][0]);
 432                 }
 433         } else if ((rdev->family == CHIP_R200) ||
 434                    (rdev->family == CHIP_RV250) ||
 435                    (rdev->family == CHIP_RV280) ||
 436                    (rdev->family == CHIP_RS300)) {
 437                 DRM_INFO("Loading R200 Microcode\n");
 438                 for (i = 0; i < 256; i++) {
 439                         WREG32(RADEON_CP_ME_RAM_DATAH, R200_cp_microcode[i][1]);
 440                         WREG32(RADEON_CP_ME_RAM_DATAL, R200_cp_microcode[i][0]);
 441                 }
 442         } else if ((rdev->family == CHIP_R300) ||
 443                    (rdev->family == CHIP_R350) ||
 444                    (rdev->family == CHIP_RV350) ||
 445                    (rdev->family == CHIP_RV380) ||
 446                    (rdev->family == CHIP_RS400) ||
 447                    (rdev->family == CHIP_RS480)) {
 448                 DRM_INFO("Loading R300 Microcode\n");
 449                 for (i = 0; i < 256; i++) {
 450                         WREG32(RADEON_CP_ME_RAM_DATAH, R300_cp_microcode[i][1]);
 451                         WREG32(RADEON_CP_ME_RAM_DATAL, R300_cp_microcode[i][0]);
 452                 }
 453         } else if ((rdev->family == CHIP_R420) ||
 454                    (rdev->family == CHIP_R423) ||
 455                    (rdev->family == CHIP_RV410)) {
 456                 DRM_INFO("Loading R400 Microcode\n");
 457                 for (i = 0; i < 256; i++) {
 458                         WREG32(RADEON_CP_ME_RAM_DATAH, R420_cp_microcode[i][1]);
 459                         WREG32(RADEON_CP_ME_RAM_DATAL, R420_cp_microcode[i][0]);
 460                 }
 461         } else if ((rdev->family == CHIP_RS690) ||
 462                    (rdev->family == CHIP_RS740)) {
 463                 DRM_INFO("Loading RS690/RS740 Microcode\n");
 464                 for (i = 0; i < 256; i++) {
 465                         WREG32(RADEON_CP_ME_RAM_DATAH, RS690_cp_microcode[i][1]);
 466                         WREG32(RADEON_CP_ME_RAM_DATAL, RS690_cp_microcode[i][0]);
 467                 }
 468         } else if (rdev->family == CHIP_RS600) {
 469                 DRM_INFO("Loading RS600 Microcode\n");
 470                 for (i = 0; i < 256; i++) {
 471                         WREG32(RADEON_CP_ME_RAM_DATAH, RS600_cp_microcode[i][1]);
 472                         WREG32(RADEON_CP_ME_RAM_DATAL, RS600_cp_microcode[i][0]);
 473                 }
 474         } else if ((rdev->family == CHIP_RV515) ||
 475                    (rdev->family == CHIP_R520) ||
 476                    (rdev->family == CHIP_RV530) ||
 477                    (rdev->family == CHIP_R580) ||
 478                    (rdev->family == CHIP_RV560) ||
 479                    (rdev->family == CHIP_RV570)) {
 480                 DRM_INFO("Loading R500 Microcode\n");
 481                 for (i = 0; i < 256; i++) {
 482                         WREG32(RADEON_CP_ME_RAM_DATAH, R520_cp_microcode[i][1]);
 483                         WREG32(RADEON_CP_ME_RAM_DATAL, R520_cp_microcode[i][0]);
 484                 }
 485         }
 486 }
 487
 488 int r100_cp_init(struct radeon_device *rdev, unsigned ring_size)
 489 {
 490         unsigned rb_bufsz;
 491         unsigned rb_blksz;
 492         unsigned max_fetch;
 493         unsigned pre_write_timer;
 494         unsigned pre_write_limit;
 495         unsigned indirect2_start;
 496         unsigned indirect1_start;
 497         uint32_t tmp;
 498         int r;
 499
 500         if (r100_debugfs_cp_init(rdev)) {
 501                 DRM_ERROR("Failed to register debugfs file for CP !\n");
 502         }
 503         /* Reset CP */
 504         tmp = RREG32(RADEON_CP_CSQ_STAT);
 505         if ((tmp & (1 << 31))) {
 506                 DRM_INFO("radeon: cp busy (0x%08X) resetting\n", tmp);
 507                 WREG32(RADEON_CP_CSQ_MODE, 0);
 508                 WREG32(RADEON_CP_CSQ_CNTL, 0);
 509                 WREG32(RADEON_RBBM_SOFT_RESET, RADEON_SOFT_RESET_CP);
 510                 tmp = RREG32(RADEON_RBBM_SOFT_RESET);
 511                 mdelay(2);
 512                 WREG32(RADEON_RBBM_SOFT_RESET, 0);
 513                 tmp = RREG32(RADEON_RBBM_SOFT_RESET);
 514                 mdelay(2);
 515                 tmp = RREG32(RADEON_CP_CSQ_STAT);
 516                 if ((tmp & (1 << 31))) {
 517                         DRM_INFO("radeon: cp reset failed (0x%08X)\n", tmp);
 518                 }
 519         } else {
 520                 DRM_INFO("radeon: cp idle (0x%08X)\n", tmp);
 521         }
 522         /* Align ring size */
 523         rb_bufsz = drm_order(ring_size / 8);
 524         ring_size = (1 << (rb_bufsz + 1)) * 4;
 525         r100_cp_load_microcode(rdev);
 526         r = radeon_ring_init(rdev, ring_size);
 527         if (r) {
 528                 return r;
 529         }
 530         /* Each time the cp read 1024 bytes (16 dword/quadword) update
 531          * the rptr copy in system ram */
 532         rb_blksz = 9;
 533         /* cp will read 128bytes at a time (4 dwords) */
 534         max_fetch = 1;
 535         rdev->cp.align_mask = 16 - 1;
 536         /* Write to CP_RB_WPTR will be delayed for pre_write_timer clocks */
 537         pre_write_timer = 64;
 538         /* Force CP_RB_WPTR write if written more than one time before the
 539          * delay expire
 540          */
 541         pre_write_limit = 0;
 542         /* Setup the cp cache like this (cache size is 96 dwords) :
 543          *      RING            0  to 15
 544          *      INDIRECT1       16 to 79
 545          *      INDIRECT2       80 to 95
 546          * So ring cache size is 16dwords (> (2 * max_fetch = 2 * 4dwords))
 547          *    indirect1 cache size is 64dwords (> (2 * max_fetch = 2 * 4dwords))
 548          *    indirect2 cache size is 16dwords (> (2 * max_fetch = 2 * 4dwords))
 549          * Idea being that most of the gpu cmd will be through indirect1 buffer
 550          * so it gets the bigger cache.
 551          */
 552         indirect2_start = 80;
 553         indirect1_start = 16;
 554         /* cp setup */
 555         WREG32(0x718, pre_write_timer | (pre_write_limit << 28));
 556         WREG32(RADEON_CP_RB_CNTL,
 557 #ifdef __BIG_ENDIAN
 558                RADEON_BUF_SWAP_32BIT |
 559 #endif
 560                REG_SET(RADEON_RB_BUFSZ, rb_bufsz) |
 561                REG_SET(RADEON_RB_BLKSZ, rb_blksz) |
 562                REG_SET(RADEON_MAX_FETCH, max_fetch) |
 563                RADEON_RB_NO_UPDATE);
 564         /* Set ring address */
 565         DRM_INFO("radeon: ring at 0x%016lX\n", (unsigned long)rdev->cp.gpu_addr);
 566         WREG32(RADEON_CP_RB_BASE, rdev->cp.gpu_addr);
 567         /* Force read & write ptr to 0 */
 568         tmp = RREG32(RADEON_CP_RB_CNTL);
 569         WREG32(RADEON_CP_RB_CNTL, tmp | RADEON_RB_RPTR_WR_ENA);
 570         WREG32(RADEON_CP_RB_RPTR_WR, 0);
 571         WREG32(RADEON_CP_RB_WPTR, 0);
 572         WREG32(RADEON_CP_RB_CNTL, tmp);
 573         udelay(10);
 574         rdev->cp.rptr = RREG32(RADEON_CP_RB_RPTR);
 575         rdev->cp.wptr = RREG32(RADEON_CP_RB_WPTR);
 576         /* Set cp mode to bus mastering & enable cp*/
 577         WREG32(RADEON_CP_CSQ_MODE,
 578                REG_SET(RADEON_INDIRECT2_START, indirect2_start) |
 579                REG_SET(RADEON_INDIRECT1_START, indirect1_start));
 580         WREG32(0x718, 0);
 581         WREG32(0x744, 0x00004D4D);
 582         WREG32(RADEON_CP_CSQ_CNTL, RADEON_CSQ_PRIBM_INDBM);
 583         radeon_ring_start(rdev);
 584         r = radeon_ring_test(rdev);
 585         if (r) {
 586                 DRM_ERROR("radeon: cp isn't working (%d).\n", r);
 587                 return r;
 588         }
 589         rdev->cp.ready = true;
 590         return 0;
 591 }
 592
 593 void r100_cp_fini(struct radeon_device *rdev)
 594 {
 595         /* Disable ring */
 596         rdev->cp.ready = false;
 597         WREG32(RADEON_CP_CSQ_CNTL, 0);
 598         radeon_ring_fini(rdev);
 599         DRM_INFO("radeon: cp finalized\n");
 600 }
 601
 602 void r100_cp_disable(struct radeon_device *rdev)
 603 {
 604         /* Disable ring */
 605         rdev->cp.ready = false;
 606         WREG32(RADEON_CP_CSQ_MODE, 0);
 607         WREG32(RADEON_CP_CSQ_CNTL, 0);
 608         if (r100_gui_wait_for_idle(rdev)) {
 609                 printk(KERN_WARNING "Failed to wait GUI idle while "
 610                        "programming pipes. Bad things might happen.\n");
 611         }
 612 }
 613
 614 int r100_cp_reset(struct radeon_device *rdev)
 615 {
 616         uint32_t tmp;
 617         bool reinit_cp;
 618         int i;
 619
 620         reinit_cp = rdev->cp.ready;
 621         rdev->cp.ready = false;
 622         WREG32(RADEON_CP_CSQ_MODE, 0);
 623         WREG32(RADEON_CP_CSQ_CNTL, 0);
 624         WREG32(RADEON_RBBM_SOFT_RESET, RADEON_SOFT_RESET_CP);
 625         (void)RREG32(RADEON_RBBM_SOFT_RESET);
 626         udelay(200);
 627         WREG32(RADEON_RBBM_SOFT_RESET, 0);
 628         /* Wait to prevent race in RBBM_STATUS */
 629         mdelay(1);
 630         for (i = 0; i < rdev->usec_timeout; i++) {
 631                 tmp = RREG32(RADEON_RBBM_STATUS);
 632                 if (!(tmp & (1 << 16))) {
 633                         DRM_INFO("CP reset succeed (RBBM_STATUS=0x%08X)\n",
 634                                  tmp);
 635                         if (reinit_cp) {
 636                                 return r100_cp_init(rdev, rdev->cp.ring_size);
 637                         }
 638                         return 0;
 639                 }
 640                 DRM_UDELAY(1);
 641         }
 642         tmp = RREG32(RADEON_RBBM_STATUS);
 643         DRM_ERROR("Failed to reset CP (RBBM_STATUS=0x%08X)!\n", tmp);
 644         return -1;
 645 }
 646
 647
 648 /*
 649  * CS functions
 650  */
 651 int r100_cs_parse_packet0(struct radeon_cs_parser *p,
 652                           struct radeon_cs_packet *pkt,
 653                           const unsigned *auth, unsigned n,
 654                           radeon_packet0_check_t check)
 655 {
 656         unsigned reg;
 657         unsigned i, j, m;
 658         unsigned idx;
 659         int r;
 660
 661         idx = pkt->idx + 1;
 662         reg = pkt->reg;
 663         /* Check that register fall into register range
 664          * determined by the number of entry (n) in the
 665          * safe register bitmap.
 666          */
 667         if (pkt->one_reg_wr) {
 668                 if ((reg >> 7) > n) {
 669                         return -EINVAL;
 670                 }
 671         } else {
 672                 if (((reg + (pkt->count << 2)) >> 7) > n) {
 673                         return -EINVAL;
 674                 }
 675         }
 676         for (i = 0; i <= pkt->count; i++, idx++) {
 677                 j = (reg >> 7);
 678                 m = 1 << ((reg >> 2) & 31);
 679                 if (auth[j] & m) {
 680                         r = check(p, pkt, idx, reg);
 681                         if (r) {
 682                                 return r;
 683                         }
 684                 }
 685                 if (pkt->one_reg_wr) {
 686                         if (!(auth[j] & m)) {
 687                                 break;
 688                         }
 689                 } else {
 690                         reg += 4;
 691                 }
 692         }
 693         return 0;
 694 }
 695
 696 void r100_cs_dump_packet(struct radeon_cs_parser *p,
 697                          struct radeon_cs_packet *pkt)
 698 {
 699         struct radeon_cs_chunk *ib_chunk;
 700         volatile uint32_t *ib;
 701         unsigned i;
 702         unsigned idx;
 703
 704         ib = p->ib->ptr;
 705         ib_chunk = &p->chunks[p->chunk_ib_idx];
 706         idx = pkt->idx;
 707         for (i = 0; i <= (pkt->count + 1); i++, idx++) {
 708                 DRM_INFO("ib[%d]=0x%08X\n", idx, ib[idx]);
 709         }
 710 }
 711
 712 /**
 713  * r100_cs_packet_parse() - parse cp packet and point ib index to next packet
 714  * @parser:     parser structure holding parsing context.
 715  * @pkt:        where to store packet informations
 716  *
 717  * Assume that chunk_ib_index is properly set. Will return -EINVAL
 718  * if packet is bigger than remaining ib size. or if packets is unknown.
 719  **/
 720 int r100_cs_packet_parse(struct radeon_cs_parser *p,
 721                          struct radeon_cs_packet *pkt,
 722                          unsigned idx)
 723 {
 724         struct radeon_cs_chunk *ib_chunk = &p->chunks[p->chunk_ib_idx];
 725         uint32_t header = ib_chunk->kdata[idx];
 726
 727         if (idx >= ib_chunk->length_dw) {
 728                 DRM_ERROR("Can not parse packet at %d after CS end %d !\n",
 729                           idx, ib_chunk->length_dw);
 730                 return -EINVAL;
 731         }
 732         pkt->idx = idx;
 733         pkt->type = CP_PACKET_GET_TYPE(header);
 734         pkt->count = CP_PACKET_GET_COUNT(header);
 735         switch (pkt->type) {
 736         case PACKET_TYPE0:
 737                 pkt->reg = CP_PACKET0_GET_REG(header);
 738                 pkt->one_reg_wr = CP_PACKET0_GET_ONE_REG_WR(header);
 739                 break;
 740         case PACKET_TYPE3:
 741                 pkt->opcode = CP_PACKET3_GET_OPCODE(header);
 742                 break;
 743         case PACKET_TYPE2:
 744                 pkt->count = -1;
 745                 break;
 746         default:
 747                 DRM_ERROR("Unknown packet type %d at %d !\n", pkt->type, idx);
 748                 return -EINVAL;
 749         }
 750         if ((pkt->count + 1 + pkt->idx) >= ib_chunk->length_dw) {
 751                 DRM_ERROR("Packet (%d:%d:%d) end after CS buffer (%d) !\n",
 752                           pkt->idx, pkt->type, pkt->count, ib_chunk->length_dw);
 753                 return -EINVAL;
 754         }
 755         return 0;
 756 }
 757
 758 /**
 759  * r100_cs_packet_next_vline() - parse userspace VLINE packet
 760  * @parser:             parser structure holding parsing context.
 761  *
 762  * Userspace sends a special sequence for VLINE waits.
 763  * PACKET0 - VLINE_START_END + value
 764  * PACKET0 - WAIT_UNTIL +_value
 765  * RELOC (P3) - crtc_id in reloc.
 766  *
 767  * This function parses this and relocates the VLINE START END
 768  * and WAIT UNTIL packets to the correct crtc.
 769  * It also detects a switched off crtc and nulls out the
 770  * wait in that case.
 771  */
 772 int r100_cs_packet_parse_vline(struct radeon_cs_parser *p)
 773 {
 774         struct radeon_cs_chunk *ib_chunk;
 775         struct drm_mode_object *obj;
 776         struct drm_crtc *crtc;
 777         struct radeon_crtc *radeon_crtc;
 778         struct radeon_cs_packet p3reloc, waitreloc;
 779         int crtc_id;
 780         int r;
 781         uint32_t header, h_idx, reg;
 782
 783         ib_chunk = &p->chunks[p->chunk_ib_idx];
 784
 785         /* parse the wait until */
 786         r = r100_cs_packet_parse(p, &waitreloc, p->idx);
 787         if (r)
 788                 return r;
 789
 790         /* check its a wait until and only 1 count */
 791         if (waitreloc.reg != RADEON_WAIT_UNTIL ||
 792             waitreloc.count != 0) {
 793                 DRM_ERROR("vline wait had illegal wait until segment\n");
 794                 r = -EINVAL;
 795                 return r;
 796         }
 797
 798         if (ib_chunk->kdata[waitreloc.idx + 1] != RADEON_WAIT_CRTC_VLINE) {
 799                 DRM_ERROR("vline wait had illegal wait until\n");
 800                 r = -EINVAL;
 801                 return r;
 802         }
 803
 804         /* jump over the NOP */
 805         r = r100_cs_packet_parse(p, &p3reloc, p->idx);
 806         if (r)
 807                 return r;
 808
 809         h_idx = p->idx - 2;
 810         p->idx += waitreloc.count;
 811         p->idx += p3reloc.count;
 812
 813         header = ib_chunk->kdata[h_idx];
 814         crtc_id = ib_chunk->kdata[h_idx + 5];
 815         reg = ib_chunk->kdata[h_idx] >> 2;
 816         mutex_lock(&p->rdev->ddev->mode_config.mutex);
 817         obj = drm_mode_object_find(p->rdev->ddev, crtc_id, DRM_MODE_OBJECT_CRTC);
 818         if (!obj) {
 819                 DRM_ERROR("cannot find crtc %d\n", crtc_id);
 820                 r = -EINVAL;
 821                 goto out;
 822         }
 823         crtc = obj_to_crtc(obj);
 824         radeon_crtc = to_radeon_crtc(crtc);
 825         crtc_id = radeon_crtc->crtc_id;
 826
 827         if (!crtc->enabled) {
 828                 /* if the CRTC isn't enabled - we need to nop out the wait until */
 829                 ib_chunk->kdata[h_idx + 2] = PACKET2(0);
 830                 ib_chunk->kdata[h_idx + 3] = PACKET2(0);
 831         } else if (crtc_id == 1) {
 832                 switch (reg) {
 833                 case AVIVO_D1MODE_VLINE_START_END:
 834                         header &= R300_CP_PACKET0_REG_MASK;
 835                         header |= AVIVO_D2MODE_VLINE_START_END >> 2;
 836                         break;
 837                 case RADEON_CRTC_GUI_TRIG_VLINE:
 838                         header &= R300_CP_PACKET0_REG_MASK;
 839                         header |= RADEON_CRTC2_GUI_TRIG_VLINE >> 2;
 840                         break;
 841                 default:
 842                         DRM_ERROR("unknown crtc reloc\n");
 843                         r = -EINVAL;
 844                         goto out;
 845                 }
 846                 ib_chunk->kdata[h_idx] = header;
 847                 ib_chunk->kdata[h_idx + 3] |= RADEON_ENG_DISPLAY_SELECT_CRTC1;
 848         }
 849 out:
 850         mutex_unlock(&p->rdev->ddev->mode_config.mutex);
 851         return r;
 852 }
 853
 854 /**
 855  * r100_cs_packet_next_reloc() - parse next packet which should be reloc packet3
 856  * @parser:             parser structure holding parsing context.
 857  * @data:               pointer to relocation data
 858  * @offset_start:       starting offset
 859  * @offset_mask:        offset mask (to align start offset on)
 860  * @reloc:              reloc informations
 861  *
 862  * Check next packet is relocation packet3, do bo validation and compute
 863  * GPU offset using the provided start.
 864  **/
 865 int r100_cs_packet_next_reloc(struct radeon_cs_parser *p,
 866                               struct radeon_cs_reloc **cs_reloc)
 867 {
 868         struct radeon_cs_chunk *ib_chunk;
 869         struct radeon_cs_chunk *relocs_chunk;
 870         struct radeon_cs_packet p3reloc;
 871         unsigned idx;
 872         int r;
 873
 874         if (p->chunk_relocs_idx == -1) {
 875                 DRM_ERROR("No relocation chunk !\n");
 876                 return -EINVAL;
 877         }
 878         *cs_reloc = NULL;
 879         ib_chunk = &p->chunks[p->chunk_ib_idx];
 880         relocs_chunk = &p->chunks[p->chunk_relocs_idx];
 881         r = r100_cs_packet_parse(p, &p3reloc, p->idx);
 882         if (r) {
 883                 return r;
 884         }
 885         p->idx += p3reloc.count + 2;
 886         if (p3reloc.type != PACKET_TYPE3 || p3reloc.opcode != PACKET3_NOP) {
 887                 DRM_ERROR("No packet3 for relocation for packet at %d.\n",
 888                           p3reloc.idx);
 889                 r100_cs_dump_packet(p, &p3reloc);
 890                 return -EINVAL;
 891         }
 892         idx = ib_chunk->kdata[p3reloc.idx + 1];
 893         if (idx >= relocs_chunk->length_dw) {
 894                 DRM_ERROR("Relocs at %d after relocations chunk end %d !\n",
 895                           idx, relocs_chunk->length_dw);
 896                 r100_cs_dump_packet(p, &p3reloc);
 897                 return -EINVAL;
 898         }
 899         /* FIXME: we assume reloc size is 4 dwords */
 900         *cs_reloc = p->relocs_ptr[(idx / 4)];
 901         return 0;
 902 }
 903
 904 static int r100_packet0_check(struct radeon_cs_parser *p,
 905                               struct radeon_cs_packet *pkt)
 906 {
 907         struct radeon_cs_chunk *ib_chunk;
 908         struct radeon_cs_reloc *reloc;
 909         volatile uint32_t *ib;
 910         uint32_t tmp;
 911         unsigned reg;
 912         unsigned i;
 913         unsigned idx;
 914         bool onereg;
 915         int r;
 916         u32 tile_flags = 0;
 917
 918         ib = p->ib->ptr;
 919         ib_chunk = &p->chunks[p->chunk_ib_idx];
 920         idx = pkt->idx + 1;
 921         reg = pkt->reg;
 922         onereg = false;
 923         if (CP_PACKET0_GET_ONE_REG_WR(ib_chunk->kdata[pkt->idx])) {
 924                 onereg = true;
 925         }
 926         for (i = 0; i <= pkt->count; i++, idx++, reg += 4) {
 927                 switch (reg) {
 928                 case RADEON_CRTC_GUI_TRIG_VLINE:
 929                         r = r100_cs_packet_parse_vline(p);
 930                         if (r) {
 931                                 DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
 932                                                 idx, reg);
 933                                 r100_cs_dump_packet(p, pkt);
 934                                 return r;
 935                         }
 936                         break;
 937                 /* FIXME: only allow PACKET3 blit? easier to check for out of
 938                  * range access */
 939                 case RADEON_DST_PITCH_OFFSET:
 940                 case RADEON_SRC_PITCH_OFFSET:
 941                         r = r100_cs_packet_next_reloc(p, &reloc);
 942                         if (r) {
 943                                 DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
 944                                           idx, reg);
 945                                 r100_cs_dump_packet(p, pkt);
 946                                 return r;
 947                         }
 948                         tmp = ib_chunk->kdata[idx] & 0x003fffff;
 949                         tmp += (((u32)reloc->lobj.gpu_offset) >> 10);
 950
 951                         if (reloc->lobj.tiling_flags & RADEON_TILING_MACRO)
 952                                 tile_flags |= RADEON_DST_TILE_MACRO;
 953                         if (reloc->lobj.tiling_flags & RADEON_TILING_MICRO) {
 954                                 if (reg == RADEON_SRC_PITCH_OFFSET) {
 955                                         DRM_ERROR("Cannot src blit from microtiled surface\n");
 956                                         r100_cs_dump_packet(p, pkt);
 957                                         return -EINVAL;
 958                                 }
 959                                 tile_flags |= RADEON_DST_TILE_MICRO;
 960                         }
 961
 962                         tmp |= tile_flags;
 963                         ib[idx] = (ib_chunk->kdata[idx] & 0x3fc00000) | tmp;
 964                         break;
 965                 case RADEON_RB3D_DEPTHOFFSET:
 966                 case RADEON_RB3D_COLOROFFSET:
 967                 case R300_RB3D_COLOROFFSET0:
 968                 case R300_ZB_DEPTHOFFSET:
 969                 case R200_PP_TXOFFSET_0:
 970                 case R200_PP_TXOFFSET_1:
 971                 case R200_PP_TXOFFSET_2:
 972                 case R200_PP_TXOFFSET_3:
 973                 case R200_PP_TXOFFSET_4:
 974                 case R200_PP_TXOFFSET_5:
 975                 case RADEON_PP_TXOFFSET_0:
 976                 case RADEON_PP_TXOFFSET_1:
 977                 case RADEON_PP_TXOFFSET_2:
 978                 case R300_TX_OFFSET_0:
 979                 case R300_TX_OFFSET_0+4:
 980                 case R300_TX_OFFSET_0+8:
 981                 case R300_TX_OFFSET_0+12:
 982                 case R300_TX_OFFSET_0+16:
 983                 case R300_TX_OFFSET_0+20:
 984                 case R300_TX_OFFSET_0+24:
 985                 case R300_TX_OFFSET_0+28:
 986                 case R300_TX_OFFSET_0+32:
 987                 case R300_TX_OFFSET_0+36:
 988                 case R300_TX_OFFSET_0+40:
 989                 case R300_TX_OFFSET_0+44:
 990                 case R300_TX_OFFSET_0+48:
 991                 case R300_TX_OFFSET_0+52:
 992                 case R300_TX_OFFSET_0+56:
 993                 case R300_TX_OFFSET_0+60:
 994                         /* rn50 has no 3D engine so fail on any 3d setup */
 995                         if (ASIC_IS_RN50(p->rdev)) {
 996                                 DRM_ERROR("attempt to use RN50 3D engine failed\n");
 997                                 return -EINVAL;
 998                         }
 999                         r = r100_cs_packet_next_reloc(p, &reloc);
1000                         if (r) {
1001                                 DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1002                                           idx, reg);
1003                                 r100_cs_dump_packet(p, pkt);
1004                                 return r;
1005                         }
1006                         ib[idx] = ib_chunk->kdata[idx] + ((u32)reloc->lobj.gpu_offset);
1007                         break;
1008                 case R300_RB3D_COLORPITCH0:
1009                 case RADEON_RB3D_COLORPITCH:
1010                         r = r100_cs_packet_next_reloc(p, &reloc);
1011                         if (r) {
1012                                 DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1013                                           idx, reg);
1014                                 r100_cs_dump_packet(p, pkt);
1015                                 return r;
1016                         }
1017
1018                         if (reloc->lobj.tiling_flags & RADEON_TILING_MACRO)
1019                                 tile_flags |= RADEON_COLOR_TILE_ENABLE;
1020                         if (reloc->lobj.tiling_flags & RADEON_TILING_MICRO)
1021                                 tile_flags |= RADEON_COLOR_MICROTILE_ENABLE;
1022
1023                         tmp = ib_chunk->kdata[idx] & ~(0x7 << 16);
1024                         tmp |= tile_flags;
1025                         ib[idx] = tmp;
1026                         break;
1027                 default:
1028                         /* FIXME: we don't want to allow anyothers packet */
1029                         break;
1030                 }
1031                 if (onereg) {
1032                         /* FIXME: forbid onereg write to register on relocate */
1033                         break;
1034                 }
1035         }
1036         return 0;
1037 }
1038
1039 int r100_cs_track_check_pkt3_indx_buffer(struct radeon_cs_parser *p,
1040                                          struct radeon_cs_packet *pkt,
1041                                          struct radeon_object *robj)
1042 {
1043         struct radeon_cs_chunk *ib_chunk;
1044         unsigned idx;
1045
1046         ib_chunk = &p->chunks[p->chunk_ib_idx];
1047         idx = pkt->idx + 1;
1048         if ((ib_chunk->kdata[idx+2] + 1) > radeon_object_size(robj)) {
1049                 DRM_ERROR("[drm] Buffer too small for PACKET3 INDX_BUFFER "
1050                           "(need %u have %lu) !\n",
1051                           ib_chunk->kdata[idx+2] + 1,
1052                           radeon_object_size(robj));
1053                 return -EINVAL;
1054         }
1055         return 0;
1056 }
1057
1058 static int r100_packet3_check(struct radeon_cs_parser *p,
1059                               struct radeon_cs_packet *pkt)
1060 {
1061         struct radeon_cs_chunk *ib_chunk;
1062         struct radeon_cs_reloc *reloc;
1063         unsigned idx;
1064         unsigned i, c;
1065         volatile uint32_t *ib;
1066         int r;
1067
1068         ib = p->ib->ptr;
1069         ib_chunk = &p->chunks[p->chunk_ib_idx];
1070         idx = pkt->idx + 1;
1071         switch (pkt->opcode) {
1072         case PACKET3_3D_LOAD_VBPNTR:
1073                 c = ib_chunk->kdata[idx++];
1074                 for (i = 0; i < (c - 1); i += 2, idx += 3) {
1075                         r = r100_cs_packet_next_reloc(p, &reloc);
1076                         if (r) {
1077                                 DRM_ERROR("No reloc for packet3 %d\n",
1078                                           pkt->opcode);
1079                                 r100_cs_dump_packet(p, pkt);
1080                                 return r;
1081                         }
1082                         ib[idx+1] = ib_chunk->kdata[idx+1] + ((u32)reloc->lobj.gpu_offset);
1083                         r = r100_cs_packet_next_reloc(p, &reloc);
1084                         if (r) {
1085                                 DRM_ERROR("No reloc for packet3 %d\n",
1086                                           pkt->opcode);
1087                                 r100_cs_dump_packet(p, pkt);
1088                                 return r;
1089                         }
1090                         ib[idx+2] = ib_chunk->kdata[idx+2] + ((u32)reloc->lobj.gpu_offset);
1091                 }
1092                 if (c & 1) {
1093                         r = r100_cs_packet_next_reloc(p, &reloc);
1094                         if (r) {
1095                                 DRM_ERROR("No reloc for packet3 %d\n",
1096                                           pkt->opcode);
1097                                 r100_cs_dump_packet(p, pkt);
1098                                 return r;
1099                         }
1100                         ib[idx+1] = ib_chunk->kdata[idx+1] + ((u32)reloc->lobj.gpu_offset);
1101                 }
1102                 break;
1103         case PACKET3_INDX_BUFFER:
1104                 r = r100_cs_packet_next_reloc(p, &reloc);
1105                 if (r) {
1106                         DRM_ERROR("No reloc for packet3 %d\n", pkt->opcode);
1107                         r100_cs_dump_packet(p, pkt);
1108                         return r;
1109                 }
1110                 ib[idx+1] = ib_chunk->kdata[idx+1] + ((u32)reloc->lobj.gpu_offset);
1111                 r = r100_cs_track_check_pkt3_indx_buffer(p, pkt, reloc->robj);
1112                 if (r) {
1113                         return r;
1114                 }
1115                 break;
1116         case 0x23:
1117                 /* FIXME: cleanup */
1118                 /* 3D_RNDR_GEN_INDX_PRIM on r100/r200 */
1119                 r = r100_cs_packet_next_reloc(p, &reloc);
1120                 if (r) {
1121                         DRM_ERROR("No reloc for packet3 %d\n", pkt->opcode);
1122                         r100_cs_dump_packet(p, pkt);
1123                         return r;
1124                 }
1125                 ib[idx] = ib_chunk->kdata[idx] + ((u32)reloc->lobj.gpu_offset);
1126                 break;
1127         case PACKET3_3D_DRAW_IMMD:
1128                 /* triggers drawing using in-packet vertex data */
1129         case PACKET3_3D_DRAW_IMMD_2:
1130                 /* triggers drawing using in-packet vertex data */
1131         case PACKET3_3D_DRAW_VBUF_2:
1132                 /* triggers drawing of vertex buffers setup elsewhere */
1133         case PACKET3_3D_DRAW_INDX_2:
1134                 /* triggers drawing using indices to vertex buffer */
1135         case PACKET3_3D_DRAW_VBUF:
1136                 /* triggers drawing of vertex buffers setup elsewhere */
1137         case PACKET3_3D_DRAW_INDX:
1138                 /* triggers drawing using indices to vertex buffer */
1139         case PACKET3_NOP:
1140                 break;
1141         default:
1142                 DRM_ERROR("Packet3 opcode %x not supported\n", pkt->opcode);
1143                 return -EINVAL;
1144         }
1145         return 0;
1146 }
1147
1148 int r100_cs_parse(struct radeon_cs_parser *p)
1149 {
1150         struct radeon_cs_packet pkt;
1151         int r;
1152
1153         do {
1154                 r = r100_cs_packet_parse(p, &pkt, p->idx);
1155                 if (r) {
1156                         return r;
1157                 }
1158                 p->idx += pkt.count + 2;
1159                 switch (pkt.type) {
1160                         case PACKET_TYPE0:
1161                                 r = r100_packet0_check(p, &pkt);
1162                                 break;
1163                         case PACKET_TYPE2:
1164                                 break;
1165                         case PACKET_TYPE3:
1166                                 r = r100_packet3_check(p, &pkt);
1167                                 break;
1168                         default:
1169                                 DRM_ERROR("Unknown packet type %d !\n",
1170                                           pkt.type);
1171                                 return -EINVAL;
1172                 }
1173                 if (r) {
1174                         return r;
1175                 }
1176         } while (p->idx < p->chunks[p->chunk_ib_idx].length_dw);
1177         return 0;
1178 }
1179
1180
1181 /*
1182  * Global GPU functions
1183  */
1184 void r100_errata(struct radeon_device *rdev)
1185 {
1186         rdev->pll_errata = 0;
1187
1188         if (rdev->family == CHIP_RV200 || rdev->family == CHIP_RS200) {
1189                 rdev->pll_errata |= CHIP_ERRATA_PLL_DUMMYREADS;
1190         }
1191
1192         if (rdev->family == CHIP_RV100 ||
1193             rdev->family == CHIP_RS100 ||
1194             rdev->family == CHIP_RS200) {
1195                 rdev->pll_errata |= CHIP_ERRATA_PLL_DELAY;
1196         }
1197 }
1198
1199 /* Wait for vertical sync on primary CRTC */
1200 void r100_gpu_wait_for_vsync(struct radeon_device *rdev)
1201 {
1202         uint32_t crtc_gen_cntl, tmp;
1203         int i;
1204
1205         crtc_gen_cntl = RREG32(RADEON_CRTC_GEN_CNTL);
1206         if ((crtc_gen_cntl & RADEON_CRTC_DISP_REQ_EN_B) ||
1207             !(crtc_gen_cntl & RADEON_CRTC_EN)) {
1208                 return;
1209         }
1210         /* Clear the CRTC_VBLANK_SAVE bit */
1211         WREG32(RADEON_CRTC_STATUS, RADEON_CRTC_VBLANK_SAVE_CLEAR);
1212         for (i = 0; i < rdev->usec_timeout; i++) {
1213                 tmp = RREG32(RADEON_CRTC_STATUS);
1214                 if (tmp & RADEON_CRTC_VBLANK_SAVE) {
1215                         return;
1216                 }
1217                 DRM_UDELAY(1);
1218         }
1219 }
1220
1221 /* Wait for vertical sync on secondary CRTC */
1222 void r100_gpu_wait_for_vsync2(struct radeon_device *rdev)
1223 {
1224         uint32_t crtc2_gen_cntl, tmp;
1225         int i;
1226
1227         crtc2_gen_cntl = RREG32(RADEON_CRTC2_GEN_CNTL);
1228         if ((crtc2_gen_cntl & RADEON_CRTC2_DISP_REQ_EN_B) ||
1229             !(crtc2_gen_cntl & RADEON_CRTC2_EN))
1230                 return;
1231
1232         /* Clear the CRTC_VBLANK_SAVE bit */
1233         WREG32(RADEON_CRTC2_STATUS, RADEON_CRTC2_VBLANK_SAVE_CLEAR);
1234         for (i = 0; i < rdev->usec_timeout; i++) {
1235                 tmp = RREG32(RADEON_CRTC2_STATUS);
1236                 if (tmp & RADEON_CRTC2_VBLANK_SAVE) {
1237                         return;
1238                 }
1239                 DRM_UDELAY(1);
1240         }
1241 }
1242
1243 int r100_rbbm_fifo_wait_for_entry(struct radeon_device *rdev, unsigned n)
1244 {
1245         unsigned i;
1246         uint32_t tmp;
1247
1248         for (i = 0; i < rdev->usec_timeout; i++) {
1249                 tmp = RREG32(RADEON_RBBM_STATUS) & RADEON_RBBM_FIFOCNT_MASK;
1250                 if (tmp >= n) {
1251                         return 0;
1252                 }
1253                 DRM_UDELAY(1);
1254         }
1255         return -1;
1256 }
1257
1258 int r100_gui_wait_for_idle(struct radeon_device *rdev)
1259 {
1260         unsigned i;
1261         uint32_t tmp;
1262
1263         if (r100_rbbm_fifo_wait_for_entry(rdev, 64)) {
1264                 printk(KERN_WARNING "radeon: wait for empty RBBM fifo failed !"
1265                        " Bad things might happen.\n");
1266         }
1267         for (i = 0; i < rdev->usec_timeout; i++) {
1268                 tmp = RREG32(RADEON_RBBM_STATUS);
1269                 if (!(tmp & (1 << 31))) {
1270                         return 0;
1271                 }
1272                 DRM_UDELAY(1);
1273         }
1274         return -1;
1275 }
1276
1277 int r100_mc_wait_for_idle(struct radeon_device *rdev)
1278 {
1279         unsigned i;
1280         uint32_t tmp;
1281
1282         for (i = 0; i < rdev->usec_timeout; i++) {
1283                 /* read MC_STATUS */
1284                 tmp = RREG32(0x0150);
1285                 if (tmp & (1 << 2)) {
1286                         return 0;
1287                 }
1288                 DRM_UDELAY(1);
1289         }
1290         return -1;
1291 }
1292
1293 void r100_gpu_init(struct radeon_device *rdev)
1294 {
1295         /* TODO: anythings to do here ? pipes ? */
1296         r100_hdp_reset(rdev);
1297 }
1298
1299 void r100_hdp_reset(struct radeon_device *rdev)
1300 {
1301         uint32_t tmp;
1302
1303         tmp = RREG32(RADEON_HOST_PATH_CNTL) & RADEON_HDP_APER_CNTL;
1304         tmp |= (7 << 28);
1305         WREG32(RADEON_HOST_PATH_CNTL, tmp | RADEON_HDP_SOFT_RESET | RADEON_HDP_READ_BUFFER_INVALIDATE);
1306         (void)RREG32(RADEON_HOST_PATH_CNTL);
1307         udelay(200);
1308         WREG32(RADEON_RBBM_SOFT_RESET, 0);
1309         WREG32(RADEON_HOST_PATH_CNTL, tmp);
1310         (void)RREG32(RADEON_HOST_PATH_CNTL);
1311 }
1312
1313 int r100_rb2d_reset(struct radeon_device *rdev)
1314 {
1315         uint32_t tmp;
1316         int i;
1317
1318         WREG32(RADEON_RBBM_SOFT_RESET, RADEON_SOFT_RESET_E2);
1319         (void)RREG32(RADEON_RBBM_SOFT_RESET);
1320         udelay(200);
1321         WREG32(RADEON_RBBM_SOFT_RESET, 0);
1322         /* Wait to prevent race in RBBM_STATUS */
1323         mdelay(1);
1324         for (i = 0; i < rdev->usec_timeout; i++) {
1325                 tmp = RREG32(RADEON_RBBM_STATUS);
1326                 if (!(tmp & (1 << 26))) {
1327                         DRM_INFO("RB2D reset succeed (RBBM_STATUS=0x%08X)\n",
1328                                  tmp);
1329                         return 0;
1330                 }
1331                 DRM_UDELAY(1);
1332         }
1333         tmp = RREG32(RADEON_RBBM_STATUS);
1334         DRM_ERROR("Failed to reset RB2D (RBBM_STATUS=0x%08X)!\n", tmp);
1335         return -1;
1336 }
1337
1338 int r100_gpu_reset(struct radeon_device *rdev)
1339 {
1340         uint32_t status;
1341
1342         /* reset order likely matter */
1343         status = RREG32(RADEON_RBBM_STATUS);
1344         /* reset HDP */
1345         r100_hdp_reset(rdev);
1346         /* reset rb2d */
1347         if (status & ((1 << 17) | (1 << 18) | (1 << 27))) {
1348                 r100_rb2d_reset(rdev);
1349         }
1350         /* TODO: reset 3D engine */
1351         /* reset CP */
1352         status = RREG32(RADEON_RBBM_STATUS);
1353         if (status & (1 << 16)) {
1354                 r100_cp_reset(rdev);
1355         }
1356         /* Check if GPU is idle */
1357         status = RREG32(RADEON_RBBM_STATUS);
1358         if (status & (1 << 31)) {
1359                 DRM_ERROR("Failed to reset GPU (RBBM_STATUS=0x%08X)\n", status);
1360                 return -1;
1361         }
1362         DRM_INFO("GPU reset succeed (RBBM_STATUS=0x%08X)\n", status);
1363         return 0;
1364 }
1365
1366
1367 /*
1368  * VRAM info
1369  */
1370 static void r100_vram_get_type(struct radeon_device *rdev)
1371 {
1372         uint32_t tmp;
1373
1374         rdev->mc.vram_is_ddr = false;
1375         if (rdev->flags & RADEON_IS_IGP)
1376                 rdev->mc.vram_is_ddr = true;
1377         else if (RREG32(RADEON_MEM_SDRAM_MODE_REG) & RADEON_MEM_CFG_TYPE_DDR)
1378                 rdev->mc.vram_is_ddr = true;
1379         if ((rdev->family == CHIP_RV100) ||
1380             (rdev->family == CHIP_RS100) ||
1381             (rdev->family == CHIP_RS200)) {
1382                 tmp = RREG32(RADEON_MEM_CNTL);
1383                 if (tmp & RV100_HALF_MODE) {
1384                         rdev->mc.vram_width = 32;
1385                 } else {
1386                         rdev->mc.vram_width = 64;
1387                 }
1388                 if (rdev->flags & RADEON_SINGLE_CRTC) {
1389                         rdev->mc.vram_width /= 4;
1390                         rdev->mc.vram_is_ddr = true;
1391                 }
1392         } else if (rdev->family <= CHIP_RV280) {
1393                 tmp = RREG32(RADEON_MEM_CNTL);
1394                 if (tmp & RADEON_MEM_NUM_CHANNELS_MASK) {
1395                         rdev->mc.vram_width = 128;
1396                 } else {
1397                         rdev->mc.vram_width = 64;
1398                 }
1399         } else {
1400                 /* newer IGPs */
1401                 rdev->mc.vram_width = 128;
1402         }
1403 }
1404
1405 static u32 r100_get_accessible_vram(struct radeon_device *rdev)
1406 {
1407         u32 aper_size;
1408         u8 byte;
1409
1410         aper_size = RREG32(RADEON_CONFIG_APER_SIZE);
1411
1412         /* Set HDP_APER_CNTL only on cards that are known not to be broken,
1413          * that is has the 2nd generation multifunction PCI interface
1414          */
1415         if (rdev->family == CHIP_RV280 ||
1416             rdev->family >= CHIP_RV350) {
1417                 WREG32_P(RADEON_HOST_PATH_CNTL, RADEON_HDP_APER_CNTL,
1418                        ~RADEON_HDP_APER_CNTL);
1419                 DRM_INFO("Generation 2 PCI interface, using max accessible memory\n");
1420                 return aper_size * 2;
1421         }
1422
1423         /* Older cards have all sorts of funny issues to deal with. First
1424          * check if it's a multifunction card by reading the PCI config
1425          * header type... Limit those to one aperture size
1426          */
1427         pci_read_config_byte(rdev->pdev, 0xe, &byte);
1428         if (byte & 0x80) {
1429                 DRM_INFO("Generation 1 PCI interface in multifunction mode\n");
1430                 DRM_INFO("Limiting VRAM to one aperture\n");
1431                 return aper_size;
1432         }
1433
1434         /* Single function older card. We read HDP_APER_CNTL to see how the BIOS
1435          * have set it up. We don't write this as it's broken on some ASICs but
1436          * we expect the BIOS to have done the right thing (might be too optimistic...)
1437          */
1438         if (RREG32(RADEON_HOST_PATH_CNTL) & RADEON_HDP_APER_CNTL)
1439                 return aper_size * 2;
1440         return aper_size;
1441 }
1442
1443 void r100_vram_init_sizes(struct radeon_device *rdev)
1444 {
1445         u64 config_aper_size;
1446         u32 accessible;
1447
1448         config_aper_size = RREG32(RADEON_CONFIG_APER_SIZE);
1449
1450         if (rdev->flags & RADEON_IS_IGP) {
1451                 uint32_t tom;
1452                 /* read NB_TOM to get the amount of ram stolen for the GPU */
1453                 tom = RREG32(RADEON_NB_TOM);
1454                 rdev->mc.real_vram_size = (((tom >> 16) - (tom & 0xffff) + 1) << 16);
1455                 /* for IGPs we need to keep VRAM where it was put by the BIOS */
1456                 rdev->mc.vram_location = (tom & 0xffff) << 16;
1457                 WREG32(RADEON_CONFIG_MEMSIZE, rdev->mc.real_vram_size);
1458                 rdev->mc.mc_vram_size = rdev->mc.real_vram_size;
1459         } else {
1460                 rdev->mc.real_vram_size = RREG32(RADEON_CONFIG_MEMSIZE);
1461                 /* Some production boards of m6 will report 0
1462                  * if it's 8 MB
1463                  */
1464                 if (rdev->mc.real_vram_size == 0) {
1465                         rdev->mc.real_vram_size = 8192 * 1024;
1466                         WREG32(RADEON_CONFIG_MEMSIZE, rdev->mc.real_vram_size);
1467                 }
1468                 /* let driver place VRAM */
1469                 rdev->mc.vram_location = 0xFFFFFFFFUL;
1470                  /* Fix for RN50, M6, M7 with 8/16/32(??) MBs of VRAM -
1471                   * Novell bug 204882 + along with lots of ubuntu ones */
1472                 if (config_aper_size > rdev->mc.real_vram_size)
1473                         rdev->mc.mc_vram_size = config_aper_size;
1474                 else
1475                         rdev->mc.mc_vram_size = rdev->mc.real_vram_size;
1476         }
1477
1478         /* work out accessible VRAM */
1479         accessible = r100_get_accessible_vram(rdev);
1480
1481         rdev->mc.aper_base = drm_get_resource_start(rdev->ddev, 0);
1482         rdev->mc.aper_size = drm_get_resource_len(rdev->ddev, 0);
1483
1484         if (accessible > rdev->mc.aper_size)
1485                 accessible = rdev->mc.aper_size;
1486
1487         if (rdev->mc.mc_vram_size > rdev->mc.aper_size)
1488                 rdev->mc.mc_vram_size = rdev->mc.aper_size;
1489
1490         if (rdev->mc.real_vram_size > rdev->mc.aper_size)
1491                 rdev->mc.real_vram_size = rdev->mc.aper_size;
1492 }
1493
1494 void r100_vram_info(struct radeon_device *rdev)
1495 {
1496         r100_vram_get_type(rdev);
1497
1498         r100_vram_init_sizes(rdev);
1499 }
1500
1501
1502 /*
1503  * Indirect registers accessor
1504  */
1505 void r100_pll_errata_after_index(struct radeon_device *rdev)
1506 {
1507         if (!(rdev->pll_errata & CHIP_ERRATA_PLL_DUMMYREADS)) {
1508                 return;
1509         }
1510         (void)RREG32(RADEON_CLOCK_CNTL_DATA);
1511         (void)RREG32(RADEON_CRTC_GEN_CNTL);
1512 }
1513
1514 static void r100_pll_errata_after_data(struct radeon_device *rdev)
1515 {
1516         /* This workarounds is necessary on RV100, RS100 and RS200 chips
1517          * or the chip could hang on a subsequent access
1518          */
1519         if (rdev->pll_errata & CHIP_ERRATA_PLL_DELAY) {
1520                 udelay(5000);
1521         }
1522
1523         /* This function is required to workaround a hardware bug in some (all?)
1524          * revisions of the R300.  This workaround should be called after every
1525          * CLOCK_CNTL_INDEX register access.  If not, register reads afterward
1526          * may not be correct.
1527          */
1528         if (rdev->pll_errata & CHIP_ERRATA_R300_CG) {
1529                 uint32_t save, tmp;
1530
1531                 save = RREG32(RADEON_CLOCK_CNTL_INDEX);
1532                 tmp = save & ~(0x3f | RADEON_PLL_WR_EN);
1533                 WREG32(RADEON_CLOCK_CNTL_INDEX, tmp);
1534                 tmp = RREG32(RADEON_CLOCK_CNTL_DATA);
1535                 WREG32(RADEON_CLOCK_CNTL_INDEX, save);
1536         }
1537 }
1538
1539 uint32_t r100_pll_rreg(struct radeon_device *rdev, uint32_t reg)
1540 {
1541         uint32_t data;
1542
1543         WREG8(RADEON_CLOCK_CNTL_INDEX, reg & 0x3f);
1544         r100_pll_errata_after_index(rdev);
1545         data = RREG32(RADEON_CLOCK_CNTL_DATA);
1546         r100_pll_errata_after_data(rdev);
1547         return data;
1548 }
1549
1550 void r100_pll_wreg(struct radeon_device *rdev, uint32_t reg, uint32_t v)
1551 {
1552         WREG8(RADEON_CLOCK_CNTL_INDEX, ((reg & 0x3f) | RADEON_PLL_WR_EN));
1553         r100_pll_errata_after_index(rdev);
1554         WREG32(RADEON_CLOCK_CNTL_DATA, v);
1555         r100_pll_errata_after_data(rdev);
1556 }
1557
1558 uint32_t r100_mm_rreg(struct radeon_device *rdev, uint32_t reg)
1559 {
1560         if (reg < 0x10000)
1561                 return readl(((void __iomem *)rdev->rmmio) + reg);
1562         else {
1563                 writel(reg, ((void __iomem *)rdev->rmmio) + RADEON_MM_INDEX);
1564                 return readl(((void __iomem *)rdev->rmmio) + RADEON_MM_DATA);
1565         }
1566 }
1567
1568 void r100_mm_wreg(struct radeon_device *rdev, uint32_t reg, uint32_t v)
1569 {
1570         if (reg < 0x10000)
1571                 writel(v, ((void __iomem *)rdev->rmmio) + reg);
1572         else {
1573                 writel(reg, ((void __iomem *)rdev->rmmio) + RADEON_MM_INDEX);
1574                 writel(v, ((void __iomem *)rdev->rmmio) + RADEON_MM_DATA);
1575         }
1576 }
1577
1578 int r100_init(struct radeon_device *rdev)
1579 {
1580         return 0;
1581 }
1582
1583 /*
1584  * Debugfs info
1585  */
1586 #if defined(CONFIG_DEBUG_FS)
1587 static int r100_debugfs_rbbm_info(struct seq_file *m, void *data)
1588 {
1589         struct drm_info_node *node = (struct drm_info_node *) m->private;
1590         struct drm_device *dev = node->minor->dev;
1591         struct radeon_device *rdev = dev->dev_private;
1592         uint32_t reg, value;
1593         unsigned i;
1594
1595         seq_printf(m, "RBBM_STATUS 0x%08x\n", RREG32(RADEON_RBBM_STATUS));
1596         seq_printf(m, "RBBM_CMDFIFO_STAT 0x%08x\n", RREG32(0xE7C));
1597         seq_printf(m, "CP_STAT 0x%08x\n", RREG32(RADEON_CP_STAT));
1598         for (i = 0; i < 64; i++) {
1599                 WREG32(RADEON_RBBM_CMDFIFO_ADDR, i | 0x100);
1600                 reg = (RREG32(RADEON_RBBM_CMDFIFO_DATA) - 1) >> 2;
1601                 WREG32(RADEON_RBBM_CMDFIFO_ADDR, i);
1602                 value = RREG32(RADEON_RBBM_CMDFIFO_DATA);
1603                 seq_printf(m, "[0x%03X] 0x%04X=0x%08X\n", i, reg, value);
1604         }
1605         return 0;
1606 }
1607
1608 static int r100_debugfs_cp_ring_info(struct seq_file *m, void *data)
1609 {
1610         struct drm_info_node *node = (struct drm_info_node *) m->private;
1611         struct drm_device *dev = node->minor->dev;
1612         struct radeon_device *rdev = dev->dev_private;
1613         uint32_t rdp, wdp;
1614         unsigned count, i, j;
1615
1616         radeon_ring_free_size(rdev);
1617         rdp = RREG32(RADEON_CP_RB_RPTR);
1618         wdp = RREG32(RADEON_CP_RB_WPTR);
1619         count = (rdp + rdev->cp.ring_size - wdp) & rdev->cp.ptr_mask;
1620         seq_printf(m, "CP_STAT 0x%08x\n", RREG32(RADEON_CP_STAT));
1621         seq_printf(m, "CP_RB_WPTR 0x%08x\n", wdp);
1622         seq_printf(m, "CP_RB_RPTR 0x%08x\n", rdp);
1623         seq_printf(m, "%u free dwords in ring\n", rdev->cp.ring_free_dw);
1624         seq_printf(m, "%u dwords in ring\n", count);
1625         for (j = 0; j <= count; j++) {
1626                 i = (rdp + j) & rdev->cp.ptr_mask;
1627                 seq_printf(m, "r[%04d]=0x%08x\n", i, rdev->cp.ring[i]);
1628         }
1629         return 0;
1630 }
1631
1632
1633 static int r100_debugfs_cp_csq_fifo(struct seq_file *m, void *data)
1634 {
1635         struct drm_info_node *node = (struct drm_info_node *) m->private;
1636         struct drm_device *dev = node->minor->dev;
1637         struct radeon_device *rdev = dev->dev_private;
1638         uint32_t csq_stat, csq2_stat, tmp;
1639         unsigned r_rptr, r_wptr, ib1_rptr, ib1_wptr, ib2_rptr, ib2_wptr;
1640         unsigned i;
1641
1642         seq_printf(m, "CP_STAT 0x%08x\n", RREG32(RADEON_CP_STAT));
1643         seq_printf(m, "CP_CSQ_MODE 0x%08x\n", RREG32(RADEON_CP_CSQ_MODE));
1644         csq_stat = RREG32(RADEON_CP_CSQ_STAT);
1645         csq2_stat = RREG32(RADEON_CP_CSQ2_STAT);
1646         r_rptr = (csq_stat >> 0) & 0x3ff;
1647         r_wptr = (csq_stat >> 10) & 0x3ff;
1648         ib1_rptr = (csq_stat >> 20) & 0x3ff;
1649         ib1_wptr = (csq2_stat >> 0) & 0x3ff;
1650         ib2_rptr = (csq2_stat >> 10) & 0x3ff;
1651         ib2_wptr = (csq2_stat >> 20) & 0x3ff;
1652         seq_printf(m, "CP_CSQ_STAT 0x%08x\n", csq_stat);
1653         seq_printf(m, "CP_CSQ2_STAT 0x%08x\n", csq2_stat);
1654         seq_printf(m, "Ring rptr %u\n", r_rptr);
1655         seq_printf(m, "Ring wptr %u\n", r_wptr);
1656         seq_printf(m, "Indirect1 rptr %u\n", ib1_rptr);
1657         seq_printf(m, "Indirect1 wptr %u\n", ib1_wptr);
1658         seq_printf(m, "Indirect2 rptr %u\n", ib2_rptr);
1659         seq_printf(m, "Indirect2 wptr %u\n", ib2_wptr);
1660         /* FIXME: 0, 128, 640 depends on fifo setup see cp_init_kms
1661          * 128 = indirect1_start * 8 & 640 = indirect2_start * 8 */
1662         seq_printf(m, "Ring fifo:\n");
1663         for (i = 0; i < 256; i++) {
1664                 WREG32(RADEON_CP_CSQ_ADDR, i << 2);
1665                 tmp = RREG32(RADEON_CP_CSQ_DATA);
1666                 seq_printf(m, "rfifo[%04d]=0x%08X\n", i, tmp);
1667         }
1668         seq_printf(m, "Indirect1 fifo:\n");
1669         for (i = 256; i <= 512; i++) {
1670                 WREG32(RADEON_CP_CSQ_ADDR, i << 2);
1671                 tmp = RREG32(RADEON_CP_CSQ_DATA);
1672                 seq_printf(m, "ib1fifo[%04d]=0x%08X\n", i, tmp);
1673         }
1674         seq_printf(m, "Indirect2 fifo:\n");
1675         for (i = 640; i < ib1_wptr; i++) {
1676                 WREG32(RADEON_CP_CSQ_ADDR, i << 2);
1677                 tmp = RREG32(RADEON_CP_CSQ_DATA);
1678                 seq_printf(m, "ib2fifo[%04d]=0x%08X\n", i, tmp);
1679         }
1680         return 0;
1681 }
1682
1683 static int r100_debugfs_mc_info(struct seq_file *m, void *data)
1684 {
1685         struct drm_info_node *node = (struct drm_info_node *) m->private;
1686         struct drm_device *dev = node->minor->dev;
1687         struct radeon_device *rdev = dev->dev_private;
1688         uint32_t tmp;
1689
1690         tmp = RREG32(RADEON_CONFIG_MEMSIZE);
1691         seq_printf(m, "CONFIG_MEMSIZE 0x%08x\n", tmp);
1692         tmp = RREG32(RADEON_MC_FB_LOCATION);
1693         seq_printf(m, "MC_FB_LOCATION 0x%08x\n", tmp);
1694         tmp = RREG32(RADEON_BUS_CNTL);
1695         seq_printf(m, "BUS_CNTL 0x%08x\n", tmp);
1696         tmp = RREG32(RADEON_MC_AGP_LOCATION);
1697         seq_printf(m, "MC_AGP_LOCATION 0x%08x\n", tmp);
1698         tmp = RREG32(RADEON_AGP_BASE);
1699         seq_printf(m, "AGP_BASE 0x%08x\n", tmp);
1700         tmp = RREG32(RADEON_HOST_PATH_CNTL);
1701         seq_printf(m, "HOST_PATH_CNTL 0x%08x\n", tmp);
1702         tmp = RREG32(0x01D0);
1703         seq_printf(m, "AIC_CTRL 0x%08x\n", tmp);
1704         tmp = RREG32(RADEON_AIC_LO_ADDR);
1705         seq_printf(m, "AIC_LO_ADDR 0x%08x\n", tmp);
1706         tmp = RREG32(RADEON_AIC_HI_ADDR);
1707         seq_printf(m, "AIC_HI_ADDR 0x%08x\n", tmp);
1708         tmp = RREG32(0x01E4);
1709         seq_printf(m, "AIC_TLB_ADDR 0x%08x\n", tmp);
1710         return 0;
1711 }
1712
1713 static struct drm_info_list r100_debugfs_rbbm_list[] = {
1714         {"r100_rbbm_info", r100_debugfs_rbbm_info, 0, NULL},
1715 };
1716
1717 static struct drm_info_list r100_debugfs_cp_list[] = {
1718         {"r100_cp_ring_info", r100_debugfs_cp_ring_info, 0, NULL},
1719         {"r100_cp_csq_fifo", r100_debugfs_cp_csq_fifo, 0, NULL},
1720 };
1721
1722 static struct drm_info_list r100_debugfs_mc_info_list[] = {
1723         {"r100_mc_info", r100_debugfs_mc_info, 0, NULL},
1724 };
1725 #endif
1726
1727 int r100_debugfs_rbbm_init(struct radeon_device *rdev)
1728 {
1729 #if defined(CONFIG_DEBUG_FS)
1730         return radeon_debugfs_add_files(rdev, r100_debugfs_rbbm_list, 1);
1731 #else
1732         return 0;
1733 #endif
1734 }
1735
1736 int r100_debugfs_cp_init(struct radeon_device *rdev)
1737 {
1738 #if defined(CONFIG_DEBUG_FS)
1739         return radeon_debugfs_add_files(rdev, r100_debugfs_cp_list, 2);
1740 #else
1741         return 0;
1742 #endif
1743 }
1744
1745 int r100_debugfs_mc_info_init(struct radeon_device *rdev)
1746 {
1747 #if defined(CONFIG_DEBUG_FS)
1748         return radeon_debugfs_add_files(rdev, r100_debugfs_mc_info_list, 1);
1749 #else
1750         return 0;
1751 #endif
1752 }
1753
1754 int r100_set_surface_reg(struct radeon_device *rdev, int reg,
1755                          uint32_t tiling_flags, uint32_t pitch,
1756                          uint32_t offset, uint32_t obj_size)
1757 {
1758         int surf_index = reg * 16;
1759         int flags = 0;
1760
1761         /* r100/r200 divide by 16 */
1762         if (rdev->family < CHIP_R300)
1763                 flags = pitch / 16;
1764         else
1765                 flags = pitch / 8;
1766
1767         if (rdev->family <= CHIP_RS200) {
1768                 if ((tiling_flags & (RADEON_TILING_MACRO|RADEON_TILING_MICRO))
1769                                  == (RADEON_TILING_MACRO|RADEON_TILING_MICRO))
1770                         flags |= RADEON_SURF_TILE_COLOR_BOTH;
1771                 if (tiling_flags & RADEON_TILING_MACRO)
1772                         flags |= RADEON_SURF_TILE_COLOR_MACRO;
1773         } else if (rdev->family <= CHIP_RV280) {
1774                 if (tiling_flags & (RADEON_TILING_MACRO))
1775                         flags |= R200_SURF_TILE_COLOR_MACRO;
1776                 if (tiling_flags & RADEON_TILING_MICRO)
1777                         flags |= R200_SURF_TILE_COLOR_MICRO;
1778         } else {
1779                 if (tiling_flags & RADEON_TILING_MACRO)
1780                         flags |= R300_SURF_TILE_MACRO;
1781                 if (tiling_flags & RADEON_TILING_MICRO)
1782                         flags |= R300_SURF_TILE_MICRO;
1783         }
1784
1785         DRM_DEBUG("writing surface %d %d %x %x\n", reg, flags, offset, offset+obj_size-1);
1786         WREG32(RADEON_SURFACE0_INFO + surf_index, flags);
1787         WREG32(RADEON_SURFACE0_LOWER_BOUND + surf_index, offset);
1788         WREG32(RADEON_SURFACE0_UPPER_BOUND + surf_index, offset + obj_size - 1);
1789         return 0;
1790 }
1791
1792 void r100_clear_surface_reg(struct radeon_device *rdev, int reg)
1793 {
1794         int surf_index = reg * 16;
1795         WREG32(RADEON_SURFACE0_INFO + surf_index, 0);
1796 }
1797
1798 void r100_bandwidth_update(struct radeon_device *rdev)
1799 {
1800         fixed20_12 trcd_ff, trp_ff, tras_ff, trbs_ff, tcas_ff;
1801         fixed20_12 sclk_ff, mclk_ff, sclk_eff_ff, sclk_delay_ff;
1802         fixed20_12 peak_disp_bw, mem_bw, pix_clk, pix_clk2, temp_ff, crit_point_ff;
1803         uint32_t temp, data, mem_trcd, mem_trp, mem_tras;
1804         fixed20_12 memtcas_ff[8] = {
1805                 fixed_init(1),
1806                 fixed_init(2),
1807                 fixed_init(3),
1808                 fixed_init(0),
1809                 fixed_init_half(1),
1810                 fixed_init_half(2),
1811                 fixed_init(0),
1812         };
1813         fixed20_12 memtcas_rs480_ff[8] = {
1814                 fixed_init(0),
1815                 fixed_init(1),
1816                 fixed_init(2),
1817                 fixed_init(3),
1818                 fixed_init(0),
1819                 fixed_init_half(1),
1820                 fixed_init_half(2),
1821                 fixed_init_half(3),
1822         };
1823         fixed20_12 memtcas2_ff[8] = {
1824                 fixed_init(0),
1825                 fixed_init(1),
1826                 fixed_init(2),
1827                 fixed_init(3),
1828                 fixed_init(4),
1829                 fixed_init(5),
1830                 fixed_init(6),
1831                 fixed_init(7),
1832         };
1833         fixed20_12 memtrbs[8] = {
1834                 fixed_init(1),
1835                 fixed_init_half(1),
1836                 fixed_init(2),
1837                 fixed_init_half(2),
1838                 fixed_init(3),
1839                 fixed_init_half(3),
1840                 fixed_init(4),
1841                 fixed_init_half(4)
1842         };
1843         fixed20_12 memtrbs_r4xx[8] = {
1844                 fixed_init(4),
1845                 fixed_init(5),
1846                 fixed_init(6),
1847                 fixed_init(7),
1848                 fixed_init(8),
1849                 fixed_init(9),
1850                 fixed_init(10),
1851                 fixed_init(11)
1852         };
1853         fixed20_12 min_mem_eff;
1854         fixed20_12 mc_latency_sclk, mc_latency_mclk, k1;
1855         fixed20_12 cur_latency_mclk, cur_latency_sclk;
1856         fixed20_12 disp_latency, disp_latency_overhead, disp_drain_rate,
1857                 disp_drain_rate2, read_return_rate;
1858         fixed20_12 time_disp1_drop_priority;
1859         int c;
1860         int cur_size = 16;       /* in octawords */
1861         int critical_point = 0, critical_point2;
1862 /*      uint32_t read_return_rate, time_disp1_drop_priority; */
1863         int stop_req, max_stop_req;
1864         struct drm_display_mode *mode1 = NULL;
1865         struct drm_display_mode *mode2 = NULL;
1866         uint32_t pixel_bytes1 = 0;
1867         uint32_t pixel_bytes2 = 0;
1868
1869         if (rdev->mode_info.crtcs[0]->base.enabled) {
1870                 mode1 = &rdev->mode_info.crtcs[0]->base.mode;
1871                 pixel_bytes1 = rdev->mode_info.crtcs[0]->base.fb->bits_per_pixel / 8;
1872         }
1873         if (rdev->mode_info.crtcs[1]->base.enabled) {
1874                 mode2 = &rdev->mode_info.crtcs[1]->base.mode;
1875                 pixel_bytes2 = rdev->mode_info.crtcs[1]->base.fb->bits_per_pixel / 8;
1876         }
1877
1878         min_mem_eff.full = rfixed_const_8(0);
1879         /* get modes */
1880         if ((rdev->disp_priority == 2) && ASIC_IS_R300(rdev)) {
1881                 uint32_t mc_init_misc_lat_timer = RREG32(R300_MC_INIT_MISC_LAT_TIMER);
1882                 mc_init_misc_lat_timer &= ~(R300_MC_DISP1R_INIT_LAT_MASK << R300_MC_DISP1R_INIT_LAT_SHIFT);
1883                 mc_init_misc_lat_timer &= ~(R300_MC_DISP0R_INIT_LAT_MASK << R300_MC_DISP0R_INIT_LAT_SHIFT);
1884                 /* check crtc enables */
1885                 if (mode2)
1886                         mc_init_misc_lat_timer |= (1 << R300_MC_DISP1R_INIT_LAT_SHIFT);
1887                 if (mode1)
1888                         mc_init_misc_lat_timer |= (1 << R300_MC_DISP0R_INIT_LAT_SHIFT);
1889                 WREG32(R300_MC_INIT_MISC_LAT_TIMER, mc_init_misc_lat_timer);
1890         }
1891
1892         /*
1893          * determine is there is enough bw for current mode
1894          */
1895         mclk_ff.full = rfixed_const(rdev->clock.default_mclk);
1896         temp_ff.full = rfixed_const(100);
1897         mclk_ff.full = rfixed_div(mclk_ff, temp_ff);
1898         sclk_ff.full = rfixed_const(rdev->clock.default_sclk);
1899         sclk_ff.full = rfixed_div(sclk_ff, temp_ff);
1900
1901         temp = (rdev->mc.vram_width / 8) * (rdev->mc.vram_is_ddr ? 2 : 1);
1902         temp_ff.full = rfixed_const(temp);
1903         mem_bw.full = rfixed_mul(mclk_ff, temp_ff);
1904
1905         pix_clk.full = 0;
1906         pix_clk2.full = 0;
1907         peak_disp_bw.full = 0;
1908         if (mode1) {
1909                 temp_ff.full = rfixed_const(1000);
1910                 pix_clk.full = rfixed_const(mode1->clock); /* convert to fixed point */
1911                 pix_clk.full = rfixed_div(pix_clk, temp_ff);
1912                 temp_ff.full = rfixed_const(pixel_bytes1);
1913                 peak_disp_bw.full += rfixed_mul(pix_clk, temp_ff);
1914         }
1915         if (mode2) {
1916                 temp_ff.full = rfixed_const(1000);
1917                 pix_clk2.full = rfixed_const(mode2->clock); /* convert to fixed point */
1918                 pix_clk2.full = rfixed_div(pix_clk2, temp_ff);
1919                 temp_ff.full = rfixed_const(pixel_bytes2);
1920                 peak_disp_bw.full += rfixed_mul(pix_clk2, temp_ff);
1921         }
1922
1923         mem_bw.full = rfixed_mul(mem_bw, min_mem_eff);
1924         if (peak_disp_bw.full >= mem_bw.full) {
1925                 DRM_ERROR("You may not have enough display bandwidth for current mode\n"
1926                           "If you have flickering problem, try to lower resolution, refresh rate, or color depth\n");
1927         }
1928
1929         /*  Get values from the EXT_MEM_CNTL register...converting its contents. */
1930         temp = RREG32(RADEON_MEM_TIMING_CNTL);
1931         if ((rdev->family == CHIP_RV100) || (rdev->flags & RADEON_IS_IGP)) { /* RV100, M6, IGPs */
1932                 mem_trcd = ((temp >> 2) & 0x3) + 1;
1933                 mem_trp  = ((temp & 0x3)) + 1;
1934                 mem_tras = ((temp & 0x70) >> 4) + 1;
1935         } else if (rdev->family == CHIP_R300 ||
1936                    rdev->family == CHIP_R350) { /* r300, r350 */
1937                 mem_trcd = (temp & 0x7) + 1;
1938                 mem_trp = ((temp >> 8) & 0x7) + 1;
1939                 mem_tras = ((temp >> 11) & 0xf) + 4;
1940         } else if (rdev->family == CHIP_RV350 ||
1941                    rdev->family <= CHIP_RV380) {
1942                 /* rv3x0 */
1943                 mem_trcd = (temp & 0x7) + 3;
1944                 mem_trp = ((temp >> 8) & 0x7) + 3;
1945                 mem_tras = ((temp >> 11) & 0xf) + 6;
1946         } else if (rdev->family == CHIP_R420 ||
1947                    rdev->family == CHIP_R423 ||
1948                    rdev->family == CHIP_RV410) {
1949                 /* r4xx */
1950                 mem_trcd = (temp & 0xf) + 3;
1951                 if (mem_trcd > 15)
1952                         mem_trcd = 15;
1953                 mem_trp = ((temp >> 8) & 0xf) + 3;
1954                 if (mem_trp > 15)
1955                         mem_trp = 15;
1956                 mem_tras = ((temp >> 12) & 0x1f) + 6;
1957                 if (mem_tras > 31)
1958                         mem_tras = 31;
1959         } else { /* RV200, R200 */
1960                 mem_trcd = (temp & 0x7) + 1;
1961                 mem_trp = ((temp >> 8) & 0x7) + 1;
1962                 mem_tras = ((temp >> 12) & 0xf) + 4;
1963         }
1964         /* convert to FF */
1965         trcd_ff.full = rfixed_const(mem_trcd);
1966         trp_ff.full = rfixed_const(mem_trp);
1967         tras_ff.full = rfixed_const(mem_tras);
1968
1969         /* Get values from the MEM_SDRAM_MODE_REG register...converting its */
1970         temp = RREG32(RADEON_MEM_SDRAM_MODE_REG);
1971         data = (temp & (7 << 20)) >> 20;
1972         if ((rdev->family == CHIP_RV100) || rdev->flags & RADEON_IS_IGP) {
1973                 if (rdev->family == CHIP_RS480) /* don't think rs400 */
1974                         tcas_ff = memtcas_rs480_ff[data];
1975                 else
1976                         tcas_ff = memtcas_ff[data];
1977         } else
1978                 tcas_ff = memtcas2_ff[data];
1979
1980         if (rdev->family == CHIP_RS400 ||
1981             rdev->family == CHIP_RS480) {
1982                 /* extra cas latency stored in bits 23-25 0-4 clocks */
1983                 data = (temp >> 23) & 0x7;
1984                 if (data < 5)
1985                         tcas_ff.full += rfixed_const(data);
1986         }
1987
1988         if (ASIC_IS_R300(rdev) && !(rdev->flags & RADEON_IS_IGP)) {
1989                 /* on the R300, Tcas is included in Trbs.
1990                  */
1991                 temp = RREG32(RADEON_MEM_CNTL);
1992                 data = (R300_MEM_NUM_CHANNELS_MASK & temp);
1993                 if (data == 1) {
1994                         if (R300_MEM_USE_CD_CH_ONLY & temp) {
1995                                 temp = RREG32(R300_MC_IND_INDEX);
1996                                 temp &= ~R300_MC_IND_ADDR_MASK;
1997                                 temp |= R300_MC_READ_CNTL_CD_mcind;
1998                                 WREG32(R300_MC_IND_INDEX, temp);
1999                                 temp = RREG32(R300_MC_IND_DATA);
2000                                 data = (R300_MEM_RBS_POSITION_C_MASK & temp);
2001                         } else {
2002                                 temp = RREG32(R300_MC_READ_CNTL_AB);
2003                                 data = (R300_MEM_RBS_POSITION_A_MASK & temp);
2004                         }
2005                 } else {
2006                         temp = RREG32(R300_MC_READ_CNTL_AB);
2007                         data = (R300_MEM_RBS_POSITION_A_MASK & temp);
2008                 }
2009                 if (rdev->family == CHIP_RV410 ||
2010                     rdev->family == CHIP_R420 ||
2011                     rdev->family == CHIP_R423)
2012                         trbs_ff = memtrbs_r4xx[data];
2013                 else
2014                         trbs_ff = memtrbs[data];
2015                 tcas_ff.full += trbs_ff.full;
2016         }
2017
2018         sclk_eff_ff.full = sclk_ff.full;
2019
2020         if (rdev->flags & RADEON_IS_AGP) {
2021                 fixed20_12 agpmode_ff;
2022                 agpmode_ff.full = rfixed_const(radeon_agpmode);
2023                 temp_ff.full = rfixed_const_666(16);
2024                 sclk_eff_ff.full -= rfixed_mul(agpmode_ff, temp_ff);
2025         }
2026         /* TODO PCIE lanes may affect this - agpmode == 16?? */
2027
2028         if (ASIC_IS_R300(rdev)) {
2029                 sclk_delay_ff.full = rfixed_const(250);
2030         } else {
2031                 if ((rdev->family == CHIP_RV100) ||
2032                     rdev->flags & RADEON_IS_IGP) {
2033                         if (rdev->mc.vram_is_ddr)
2034                                 sclk_delay_ff.full = rfixed_const(41);
2035                         else
2036                                 sclk_delay_ff.full = rfixed_const(33);
2037                 } else {
2038                         if (rdev->mc.vram_width == 128)
2039                                 sclk_delay_ff.full = rfixed_const(57);
2040                         else
2041                                 sclk_delay_ff.full = rfixed_const(41);
2042                 }
2043         }
2044
2045         mc_latency_sclk.full = rfixed_div(sclk_delay_ff, sclk_eff_ff);
2046
2047         if (rdev->mc.vram_is_ddr) {
2048                 if (rdev->mc.vram_width == 32) {
2049                         k1.full = rfixed_const(40);
2050                         c  = 3;
2051                 } else {
2052                         k1.full = rfixed_const(20);
2053                         c  = 1;
2054                 }
2055         } else {
2056                 k1.full = rfixed_const(40);
2057                 c  = 3;
2058         }
2059
2060         temp_ff.full = rfixed_const(2);
2061         mc_latency_mclk.full = rfixed_mul(trcd_ff, temp_ff);
2062         temp_ff.full = rfixed_const(c);
2063         mc_latency_mclk.full += rfixed_mul(tcas_ff, temp_ff);
2064         temp_ff.full = rfixed_const(4);
2065         mc_latency_mclk.full += rfixed_mul(tras_ff, temp_ff);
2066         mc_latency_mclk.full += rfixed_mul(trp_ff, temp_ff);
2067         mc_latency_mclk.full += k1.full;
2068
2069         mc_latency_mclk.full = rfixed_div(mc_latency_mclk, mclk_ff);
2070         mc_latency_mclk.full += rfixed_div(temp_ff, sclk_eff_ff);
2071
2072         /*
2073           HW cursor time assuming worst case of full size colour cursor.
2074         */
2075         temp_ff.full = rfixed_const((2 * (cur_size - (rdev->mc.vram_is_ddr + 1))));
2076         temp_ff.full += trcd_ff.full;
2077         if (temp_ff.full < tras_ff.full)
2078                 temp_ff.full = tras_ff.full;
2079         cur_latency_mclk.full = rfixed_div(temp_ff, mclk_ff);
2080
2081         temp_ff.full = rfixed_const(cur_size);
2082         cur_latency_sclk.full = rfixed_div(temp_ff, sclk_eff_ff);
2083         /*
2084           Find the total latency for the display data.
2085         */
2086         disp_latency_overhead.full = rfixed_const(80);
2087         disp_latency_overhead.full = rfixed_div(disp_latency_overhead, sclk_ff);
2088         mc_latency_mclk.full += disp_latency_overhead.full + cur_latency_mclk.full;
2089         mc_latency_sclk.full += disp_latency_overhead.full + cur_latency_sclk.full;
2090
2091         if (mc_latency_mclk.full > mc_latency_sclk.full)
2092                 disp_latency.full = mc_latency_mclk.full;
2093         else
2094                 disp_latency.full = mc_latency_sclk.full;
2095
2096         /* setup Max GRPH_STOP_REQ default value */
2097         if (ASIC_IS_RV100(rdev))
2098                 max_stop_req = 0x5c;
2099         else
2100                 max_stop_req = 0x7c;
2101
2102         if (mode1) {
2103                 /*  CRTC1
2104                     Set GRPH_BUFFER_CNTL register using h/w defined optimal values.
2105                     GRPH_STOP_REQ <= MIN[ 0x7C, (CRTC_H_DISP + 1) * (bit depth) / 0x10 ]
2106                 */
2107                 stop_req = mode1->hdisplay * pixel_bytes1 / 16;
2108
2109                 if (stop_req > max_stop_req)
2110                         stop_req = max_stop_req;
2111
2112                 /*
2113                   Find the drain rate of the display buffer.
2114                 */
2115                 temp_ff.full = rfixed_const((16/pixel_bytes1));
2116                 disp_drain_rate.full = rfixed_div(pix_clk, temp_ff);
2117
2118                 /*
2119                   Find the critical point of the display buffer.
2120                 */
2121                 crit_point_ff.full = rfixed_mul(disp_drain_rate, disp_latency);
2122                 crit_point_ff.full += rfixed_const_half(0);
2123
2124                 critical_point = rfixed_trunc(crit_point_ff);
2125
2126                 if (rdev->disp_priority == 2) {
2127                         critical_point = 0;
2128                 }
2129
2130                 /*
2131                   The critical point should never be above max_stop_req-4.  Setting
2132                   GRPH_CRITICAL_CNTL = 0 will thus force high priority all the time.
2133                 */
2134                 if (max_stop_req - critical_point < 4)
2135                         critical_point = 0;
2136
2137                 if (critical_point == 0 && mode2 && rdev->family == CHIP_R300) {
2138                         /* some R300 cards have problem with this set to 0, when CRTC2 is enabled.*/
2139                         critical_point = 0x10;
2140                 }
2141
2142                 temp = RREG32(RADEON_GRPH_BUFFER_CNTL);
2143                 temp &= ~(RADEON_GRPH_STOP_REQ_MASK);
2144                 temp |= (stop_req << RADEON_GRPH_STOP_REQ_SHIFT);
2145                 temp &= ~(RADEON_GRPH_START_REQ_MASK);
2146                 if ((rdev->family == CHIP_R350) &&
2147                     (stop_req > 0x15)) {
2148                         stop_req -= 0x10;
2149                 }
2150                 temp |= (stop_req << RADEON_GRPH_START_REQ_SHIFT);
2151                 temp |= RADEON_GRPH_BUFFER_SIZE;
2152                 temp &= ~(RADEON_GRPH_CRITICAL_CNTL   |
2153                           RADEON_GRPH_CRITICAL_AT_SOF |
2154                           RADEON_GRPH_STOP_CNTL);
2155                 /*
2156                   Write the result into the register.
2157                 */
2158                 WREG32(RADEON_GRPH_BUFFER_CNTL, ((temp & ~RADEON_GRPH_CRITICAL_POINT_MASK) |
2159                                                        (critical_point << RADEON_GRPH_CRITICAL_POINT_SHIFT)));
2160
2161 #if 0
2162                 if ((rdev->family == CHIP_RS400) ||
2163                     (rdev->family == CHIP_RS480)) {
2164                         /* attempt to program RS400 disp regs correctly ??? */
2165                         temp = RREG32(RS400_DISP1_REG_CNTL);
2166                         temp &= ~(RS400_DISP1_START_REQ_LEVEL_MASK |
2167                                   RS400_DISP1_STOP_REQ_LEVEL_MASK);
2168                         WREG32(RS400_DISP1_REQ_CNTL1, (temp |
2169                                                        (critical_point << RS400_DISP1_START_REQ_LEVEL_SHIFT) |
2170                                                        (critical_point << RS400_DISP1_STOP_REQ_LEVEL_SHIFT)));
2171                         temp = RREG32(RS400_DMIF_MEM_CNTL1);
2172                         temp &= ~(RS400_DISP1_CRITICAL_POINT_START_MASK |
2173                                   RS400_DISP1_CRITICAL_POINT_STOP_MASK);
2174                         WREG32(RS400_DMIF_MEM_CNTL1, (temp |
2175                                                       (critical_point << RS400_DISP1_CRITICAL_POINT_START_SHIFT) |
2176                                                       (critical_point << RS400_DISP1_CRITICAL_POINT_STOP_SHIFT)));
2177                 }
2178 #endif
2179
2180                 DRM_DEBUG("GRPH_BUFFER_CNTL from to %x\n",
2181                           /*      (unsigned int)info->SavedReg->grph_buffer_cntl, */
2182                           (unsigned int)RREG32(RADEON_GRPH_BUFFER_CNTL));
2183         }
2184
2185         if (mode2) {
2186                 u32 grph2_cntl;
2187                 stop_req = mode2->hdisplay * pixel_bytes2 / 16;
2188
2189                 if (stop_req > max_stop_req)
2190                         stop_req = max_stop_req;
2191
2192                 /*
2193                   Find the drain rate of the display buffer.
2194                 */
2195                 temp_ff.full = rfixed_const((16/pixel_bytes2));
2196                 disp_drain_rate2.full = rfixed_div(pix_clk2, temp_ff);
2197
2198                 grph2_cntl = RREG32(RADEON_GRPH2_BUFFER_CNTL);
2199                 grph2_cntl &= ~(RADEON_GRPH_STOP_REQ_MASK);
2200                 grph2_cntl |= (stop_req << RADEON_GRPH_STOP_REQ_SHIFT);
2201                 grph2_cntl &= ~(RADEON_GRPH_START_REQ_MASK);
2202                 if ((rdev->family == CHIP_R350) &&
2203                     (stop_req > 0x15)) {
2204                         stop_req -= 0x10;
2205                 }
2206                 grph2_cntl |= (stop_req << RADEON_GRPH_START_REQ_SHIFT);
2207                 grph2_cntl |= RADEON_GRPH_BUFFER_SIZE;
2208                 grph2_cntl &= ~(RADEON_GRPH_CRITICAL_CNTL   |
2209                           RADEON_GRPH_CRITICAL_AT_SOF |
2210                           RADEON_GRPH_STOP_CNTL);
2211
2212                 if ((rdev->family == CHIP_RS100) ||
2213                     (rdev->family == CHIP_RS200))
2214                         critical_point2 = 0;
2215                 else {
2216                         temp = (rdev->mc.vram_width * rdev->mc.vram_is_ddr + 1)/128;
2217                         temp_ff.full = rfixed_const(temp);
2218                         temp_ff.full = rfixed_mul(mclk_ff, temp_ff);
2219                         if (sclk_ff.full < temp_ff.full)
2220                                 temp_ff.full = sclk_ff.full;
2221
2222                         read_return_rate.full = temp_ff.full;
2223
2224                         if (mode1) {
2225                                 temp_ff.full = read_return_rate.full - disp_drain_rate.full;
2226                                 time_disp1_drop_priority.full = rfixed_div(crit_point_ff, temp_ff);
2227                         } else {
2228                                 time_disp1_drop_priority.full = 0;
2229                         }
2230                         crit_point_ff.full = disp_latency.full + time_disp1_drop_priority.full + disp_latency.full;
2231                         crit_point_ff.full = rfixed_mul(crit_point_ff, disp_drain_rate2);
2232                         crit_point_ff.full += rfixed_const_half(0);
2233
2234                         critical_point2 = rfixed_trunc(crit_point_ff);
2235
2236                         if (rdev->disp_priority == 2) {
2237                                 critical_point2 = 0;
2238                         }
2239
2240                         if (max_stop_req - critical_point2 < 4)
2241                                 critical_point2 = 0;
2242
2243                 }
2244
2245                 if (critical_point2 == 0 && rdev->family == CHIP_R300) {
2246                         /* some R300 cards have problem with this set to 0 */
2247                         critical_point2 = 0x10;
2248                 }
2249
2250                 WREG32(RADEON_GRPH2_BUFFER_CNTL, ((grph2_cntl & ~RADEON_GRPH_CRITICAL_POINT_MASK) |
2251                                                   (critical_point2 << RADEON_GRPH_CRITICAL_POINT_SHIFT)));
2252
2253                 if ((rdev->family == CHIP_RS400) ||
2254                     (rdev->family == CHIP_RS480)) {
2255 #if 0
2256                         /* attempt to program RS400 disp2 regs correctly ??? */
2257                         temp = RREG32(RS400_DISP2_REQ_CNTL1);
2258                         temp &= ~(RS400_DISP2_START_REQ_LEVEL_MASK |
2259                                   RS400_DISP2_STOP_REQ_LEVEL_MASK);
2260                         WREG32(RS400_DISP2_REQ_CNTL1, (temp |
2261                                                        (critical_point2 << RS400_DISP1_START_REQ_LEVEL_SHIFT) |
2262                                                        (critical_point2 << RS400_DISP1_STOP_REQ_LEVEL_SHIFT)));
2263                         temp = RREG32(RS400_DISP2_REQ_CNTL2);
2264                         temp &= ~(RS400_DISP2_CRITICAL_POINT_START_MASK |
2265                                   RS400_DISP2_CRITICAL_POINT_STOP_MASK);
2266                         WREG32(RS400_DISP2_REQ_CNTL2, (temp |
2267                                                        (critical_point2 << RS400_DISP2_CRITICAL_POINT_START_SHIFT) |
2268                                                        (critical_point2 << RS400_DISP2_CRITICAL_POINT_STOP_SHIFT)));
2269 #endif
2270                         WREG32(RS400_DISP2_REQ_CNTL1, 0x105DC1CC);
2271                         WREG32(RS400_DISP2_REQ_CNTL2, 0x2749D000);
2272                         WREG32(RS400_DMIF_MEM_CNTL1,  0x29CA71DC);
2273                         WREG32(RS400_DISP1_REQ_CNTL1, 0x28FBC3AC);
2274                 }
2275
2276                 DRM_DEBUG("GRPH2_BUFFER_CNTL from to %x\n",
2277                           (unsigned int)RREG32(RADEON_GRPH2_BUFFER_CNTL));
2278         }
2279 }