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;
 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         header = ib_chunk->kdata[idx];
 733         pkt->idx = idx;
 734         pkt->type = CP_PACKET_GET_TYPE(header);
 735         pkt->count = CP_PACKET_GET_COUNT(header);
 736         switch (pkt->type) {
 737         case PACKET_TYPE0:
 738                 pkt->reg = CP_PACKET0_GET_REG(header);
 739                 pkt->one_reg_wr = CP_PACKET0_GET_ONE_REG_WR(header);
 740                 break;
 741         case PACKET_TYPE3:
 742                 pkt->opcode = CP_PACKET3_GET_OPCODE(header);
 743                 break;
 744         case PACKET_TYPE2:
 745                 pkt->count = -1;
 746                 break;
 747         default:
 748                 DRM_ERROR("Unknown packet type %d at %d !\n", pkt->type, idx);
 749                 return -EINVAL;
 750         }
 751         if ((pkt->count + 1 + pkt->idx) >= ib_chunk->length_dw) {
 752                 DRM_ERROR("Packet (%d:%d:%d) end after CS buffer (%d) !\n",
 753                           pkt->idx, pkt->type, pkt->count, ib_chunk->length_dw);
 754                 return -EINVAL;
 755         }
 756         return 0;
 757 }
 758
 759 /**
 760  * r100_cs_packet_next_vline() - parse userspace VLINE packet
 761  * @parser:             parser structure holding parsing context.
 762  *
 763  * Userspace sends a special sequence for VLINE waits.
 764  * PACKET0 - VLINE_START_END + value
 765  * PACKET0 - WAIT_UNTIL +_value
 766  * RELOC (P3) - crtc_id in reloc.
 767  *
 768  * This function parses this and relocates the VLINE START END
 769  * and WAIT UNTIL packets to the correct crtc.
 770  * It also detects a switched off crtc and nulls out the
 771  * wait in that case.
 772  */
 773 int r100_cs_packet_parse_vline(struct radeon_cs_parser *p)
 774 {
 775         struct radeon_cs_chunk *ib_chunk;
 776         struct drm_mode_object *obj;
 777         struct drm_crtc *crtc;
 778         struct radeon_crtc *radeon_crtc;
 779         struct radeon_cs_packet p3reloc, waitreloc;
 780         int crtc_id;
 781         int r;
 782         uint32_t header, h_idx, reg;
 783
 784         ib_chunk = &p->chunks[p->chunk_ib_idx];
 785
 786         /* parse the wait until */
 787         r = r100_cs_packet_parse(p, &waitreloc, p->idx);
 788         if (r)
 789                 return r;
 790
 791         /* check its a wait until and only 1 count */
 792         if (waitreloc.reg != RADEON_WAIT_UNTIL ||
 793             waitreloc.count != 0) {
 794                 DRM_ERROR("vline wait had illegal wait until segment\n");
 795                 r = -EINVAL;
 796                 return r;
 797         }
 798
 799         if (ib_chunk->kdata[waitreloc.idx + 1] != RADEON_WAIT_CRTC_VLINE) {
 800                 DRM_ERROR("vline wait had illegal wait until\n");
 801                 r = -EINVAL;
 802                 return r;
 803         }
 804
 805         /* jump over the NOP */
 806         r = r100_cs_packet_parse(p, &p3reloc, p->idx);
 807         if (r)
 808                 return r;
 809
 810         h_idx = p->idx - 2;
 811         p->idx += waitreloc.count;
 812         p->idx += p3reloc.count;
 813
 814         header = ib_chunk->kdata[h_idx];
 815         crtc_id = ib_chunk->kdata[h_idx + 5];
 816         reg = ib_chunk->kdata[h_idx] >> 2;
 817         mutex_lock(&p->rdev->ddev->mode_config.mutex);
 818         obj = drm_mode_object_find(p->rdev->ddev, crtc_id, DRM_MODE_OBJECT_CRTC);
 819         if (!obj) {
 820                 DRM_ERROR("cannot find crtc %d\n", crtc_id);
 821                 r = -EINVAL;
 822                 goto out;
 823         }
 824         crtc = obj_to_crtc(obj);
 825         radeon_crtc = to_radeon_crtc(crtc);
 826         crtc_id = radeon_crtc->crtc_id;
 827
 828         if (!crtc->enabled) {
 829                 /* if the CRTC isn't enabled - we need to nop out the wait until */
 830                 ib_chunk->kdata[h_idx + 2] = PACKET2(0);
 831                 ib_chunk->kdata[h_idx + 3] = PACKET2(0);
 832         } else if (crtc_id == 1) {
 833                 switch (reg) {
 834                 case AVIVO_D1MODE_VLINE_START_END:
 835                         header &= R300_CP_PACKET0_REG_MASK;
 836                         header |= AVIVO_D2MODE_VLINE_START_END >> 2;
 837                         break;
 838                 case RADEON_CRTC_GUI_TRIG_VLINE:
 839                         header &= R300_CP_PACKET0_REG_MASK;
 840                         header |= RADEON_CRTC2_GUI_TRIG_VLINE >> 2;
 841                         break;
 842                 default:
 843                         DRM_ERROR("unknown crtc reloc\n");
 844                         r = -EINVAL;
 845                         goto out;
 846                 }
 847                 ib_chunk->kdata[h_idx] = header;
 848                 ib_chunk->kdata[h_idx + 3] |= RADEON_ENG_DISPLAY_SELECT_CRTC1;
 849         }
 850 out:
 851         mutex_unlock(&p->rdev->ddev->mode_config.mutex);
 852         return r;
 853 }
 854
 855 /**
 856  * r100_cs_packet_next_reloc() - parse next packet which should be reloc packet3
 857  * @parser:             parser structure holding parsing context.
 858  * @data:               pointer to relocation data
 859  * @offset_start:       starting offset
 860  * @offset_mask:        offset mask (to align start offset on)
 861  * @reloc:              reloc informations
 862  *
 863  * Check next packet is relocation packet3, do bo validation and compute
 864  * GPU offset using the provided start.
 865  **/
 866 int r100_cs_packet_next_reloc(struct radeon_cs_parser *p,
 867                               struct radeon_cs_reloc **cs_reloc)
 868 {
 869         struct radeon_cs_chunk *ib_chunk;
 870         struct radeon_cs_chunk *relocs_chunk;
 871         struct radeon_cs_packet p3reloc;
 872         unsigned idx;
 873         int r;
 874
 875         if (p->chunk_relocs_idx == -1) {
 876                 DRM_ERROR("No relocation chunk !\n");
 877                 return -EINVAL;
 878         }
 879         *cs_reloc = NULL;
 880         ib_chunk = &p->chunks[p->chunk_ib_idx];
 881         relocs_chunk = &p->chunks[p->chunk_relocs_idx];
 882         r = r100_cs_packet_parse(p, &p3reloc, p->idx);
 883         if (r) {
 884                 return r;
 885         }
 886         p->idx += p3reloc.count + 2;
 887         if (p3reloc.type != PACKET_TYPE3 || p3reloc.opcode != PACKET3_NOP) {
 888                 DRM_ERROR("No packet3 for relocation for packet at %d.\n",
 889                           p3reloc.idx);
 890                 r100_cs_dump_packet(p, &p3reloc);
 891                 return -EINVAL;
 892         }
 893         idx = ib_chunk->kdata[p3reloc.idx + 1];
 894         if (idx >= relocs_chunk->length_dw) {
 895                 DRM_ERROR("Relocs at %d after relocations chunk end %d !\n",
 896                           idx, relocs_chunk->length_dw);
 897                 r100_cs_dump_packet(p, &p3reloc);
 898                 return -EINVAL;
 899         }
 900         /* FIXME: we assume reloc size is 4 dwords */
 901         *cs_reloc = p->relocs_ptr[(idx / 4)];
 902         return 0;
 903 }
 904
 905 static int r100_packet0_check(struct radeon_cs_parser *p,
 906                               struct radeon_cs_packet *pkt)
 907 {
 908         struct radeon_cs_chunk *ib_chunk;
 909         struct radeon_cs_reloc *reloc;
 910         volatile uint32_t *ib;
 911         uint32_t tmp;
 912         unsigned reg;
 913         unsigned i;
 914         unsigned idx;
 915         bool onereg;
 916         int r;
 917         u32 tile_flags = 0;
 918
 919         ib = p->ib->ptr;
 920         ib_chunk = &p->chunks[p->chunk_ib_idx];
 921         idx = pkt->idx + 1;
 922         reg = pkt->reg;
 923         onereg = false;
 924         if (CP_PACKET0_GET_ONE_REG_WR(ib_chunk->kdata[pkt->idx])) {
 925                 onereg = true;
 926         }
 927         for (i = 0; i <= pkt->count; i++, idx++, reg += 4) {
 928                 switch (reg) {
 929                 case RADEON_CRTC_GUI_TRIG_VLINE:
 930                         r = r100_cs_packet_parse_vline(p);
 931                         if (r) {
 932                                 DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
 933                                                 idx, reg);
 934                                 r100_cs_dump_packet(p, pkt);
 935                                 return r;
 936                         }
 937                         break;
 938                 /* FIXME: only allow PACKET3 blit? easier to check for out of
 939                  * range access */
 940                 case RADEON_DST_PITCH_OFFSET:
 941                 case RADEON_SRC_PITCH_OFFSET:
 942                         r = r100_cs_packet_next_reloc(p, &reloc);
 943                         if (r) {
 944                                 DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
 945                                           idx, reg);
 946                                 r100_cs_dump_packet(p, pkt);
 947                                 return r;
 948                         }
 949                         tmp = ib_chunk->kdata[idx] & 0x003fffff;
 950                         tmp += (((u32)reloc->lobj.gpu_offset) >> 10);
 951
 952                         if (reloc->lobj.tiling_flags & RADEON_TILING_MACRO)
 953                                 tile_flags |= RADEON_DST_TILE_MACRO;
 954                         if (reloc->lobj.tiling_flags & RADEON_TILING_MICRO) {
 955                                 if (reg == RADEON_SRC_PITCH_OFFSET) {
 956                                         DRM_ERROR("Cannot src blit from microtiled surface\n");
 957                                         r100_cs_dump_packet(p, pkt);
 958                                         return -EINVAL;
 959                                 }
 960                                 tile_flags |= RADEON_DST_TILE_MICRO;
 961                         }
 962
 963                         tmp |= tile_flags;
 964                         ib[idx] = (ib_chunk->kdata[idx] & 0x3fc00000) | tmp;
 965                         break;
 966                 case RADEON_RB3D_DEPTHOFFSET:
 967                 case RADEON_RB3D_COLOROFFSET:
 968                 case R300_RB3D_COLOROFFSET0:
 969                 case R300_ZB_DEPTHOFFSET:
 970                 case R200_PP_TXOFFSET_0:
 971                 case R200_PP_TXOFFSET_1:
 972                 case R200_PP_TXOFFSET_2:
 973                 case R200_PP_TXOFFSET_3:
 974                 case R200_PP_TXOFFSET_4:
 975                 case R200_PP_TXOFFSET_5:
 976                 case RADEON_PP_TXOFFSET_0:
 977                 case RADEON_PP_TXOFFSET_1:
 978                 case RADEON_PP_TXOFFSET_2:
 979                 case R300_TX_OFFSET_0:
 980                 case R300_TX_OFFSET_0+4:
 981                 case R300_TX_OFFSET_0+8:
 982                 case R300_TX_OFFSET_0+12:
 983                 case R300_TX_OFFSET_0+16:
 984                 case R300_TX_OFFSET_0+20:
 985                 case R300_TX_OFFSET_0+24:
 986                 case R300_TX_OFFSET_0+28:
 987                 case R300_TX_OFFSET_0+32:
 988                 case R300_TX_OFFSET_0+36:
 989                 case R300_TX_OFFSET_0+40:
 990                 case R300_TX_OFFSET_0+44:
 991                 case R300_TX_OFFSET_0+48:
 992                 case R300_TX_OFFSET_0+52:
 993                 case R300_TX_OFFSET_0+56:
 994                 case R300_TX_OFFSET_0+60:
 995                         /* rn50 has no 3D engine so fail on any 3d setup */
 996                         if (ASIC_IS_RN50(p->rdev)) {
 997                                 DRM_ERROR("attempt to use RN50 3D engine failed\n");
 998                                 return -EINVAL;
 999                         }
1000                         r = r100_cs_packet_next_reloc(p, &reloc);
1001                         if (r) {
1002                                 DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1003                                           idx, reg);
1004                                 r100_cs_dump_packet(p, pkt);
1005                                 return r;
1006                         }
1007                         ib[idx] = ib_chunk->kdata[idx] + ((u32)reloc->lobj.gpu_offset);
1008                         break;
1009                 case R300_RB3D_COLORPITCH0:
1010                 case RADEON_RB3D_COLORPITCH:
1011                         r = r100_cs_packet_next_reloc(p, &reloc);
1012                         if (r) {
1013                                 DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1014                                           idx, reg);
1015                                 r100_cs_dump_packet(p, pkt);
1016                                 return r;
1017                         }
1018
1019                         if (reloc->lobj.tiling_flags & RADEON_TILING_MACRO)
1020                                 tile_flags |= RADEON_COLOR_TILE_ENABLE;
1021                         if (reloc->lobj.tiling_flags & RADEON_TILING_MICRO)
1022                                 tile_flags |= RADEON_COLOR_MICROTILE_ENABLE;
1023
1024                         tmp = ib_chunk->kdata[idx] & ~(0x7 << 16);
1025                         tmp |= tile_flags;
1026                         ib[idx] = tmp;
1027                         break;
1028                 default:
1029                         /* FIXME: we don't want to allow anyothers packet */
1030                         break;
1031                 }
1032                 if (onereg) {
1033                         /* FIXME: forbid onereg write to register on relocate */
1034                         break;
1035                 }
1036         }
1037         return 0;
1038 }
1039
1040 int r100_cs_track_check_pkt3_indx_buffer(struct radeon_cs_parser *p,
1041                                          struct radeon_cs_packet *pkt,
1042                                          struct radeon_object *robj)
1043 {
1044         struct radeon_cs_chunk *ib_chunk;
1045         unsigned idx;
1046
1047         ib_chunk = &p->chunks[p->chunk_ib_idx];
1048         idx = pkt->idx + 1;
1049         if ((ib_chunk->kdata[idx+2] + 1) > radeon_object_size(robj)) {
1050                 DRM_ERROR("[drm] Buffer too small for PACKET3 INDX_BUFFER "
1051                           "(need %u have %lu) !\n",
1052                           ib_chunk->kdata[idx+2] + 1,
1053                           radeon_object_size(robj));
1054                 return -EINVAL;
1055         }
1056         return 0;
1057 }
1058
1059 static int r100_packet3_check(struct radeon_cs_parser *p,
1060                               struct radeon_cs_packet *pkt)
1061 {
1062         struct radeon_cs_chunk *ib_chunk;
1063         struct radeon_cs_reloc *reloc;
1064         unsigned idx;
1065         unsigned i, c;
1066         volatile uint32_t *ib;
1067         int r;
1068
1069         ib = p->ib->ptr;
1070         ib_chunk = &p->chunks[p->chunk_ib_idx];
1071         idx = pkt->idx + 1;
1072         switch (pkt->opcode) {
1073         case PACKET3_3D_LOAD_VBPNTR:
1074                 c = ib_chunk->kdata[idx++];
1075                 for (i = 0; i < (c - 1); i += 2, idx += 3) {
1076                         r = r100_cs_packet_next_reloc(p, &reloc);
1077                         if (r) {
1078                                 DRM_ERROR("No reloc for packet3 %d\n",
1079                                           pkt->opcode);
1080                                 r100_cs_dump_packet(p, pkt);
1081                                 return r;
1082                         }
1083                         ib[idx+1] = ib_chunk->kdata[idx+1] + ((u32)reloc->lobj.gpu_offset);
1084                         r = r100_cs_packet_next_reloc(p, &reloc);
1085                         if (r) {
1086                                 DRM_ERROR("No reloc for packet3 %d\n",
1087                                           pkt->opcode);
1088                                 r100_cs_dump_packet(p, pkt);
1089                                 return r;
1090                         }
1091                         ib[idx+2] = ib_chunk->kdata[idx+2] + ((u32)reloc->lobj.gpu_offset);
1092                 }
1093                 if (c & 1) {
1094                         r = r100_cs_packet_next_reloc(p, &reloc);
1095                         if (r) {
1096                                 DRM_ERROR("No reloc for packet3 %d\n",
1097                                           pkt->opcode);
1098                                 r100_cs_dump_packet(p, pkt);
1099                                 return r;
1100                         }
1101                         ib[idx+1] = ib_chunk->kdata[idx+1] + ((u32)reloc->lobj.gpu_offset);
1102                 }
1103                 break;
1104         case PACKET3_INDX_BUFFER:
1105                 r = r100_cs_packet_next_reloc(p, &reloc);
1106                 if (r) {
1107                         DRM_ERROR("No reloc for packet3 %d\n", pkt->opcode);
1108                         r100_cs_dump_packet(p, pkt);
1109                         return r;
1110                 }
1111                 ib[idx+1] = ib_chunk->kdata[idx+1] + ((u32)reloc->lobj.gpu_offset);
1112                 r = r100_cs_track_check_pkt3_indx_buffer(p, pkt, reloc->robj);
1113                 if (r) {
1114                         return r;
1115                 }
1116                 break;
1117         case 0x23:
1118                 /* FIXME: cleanup */
1119                 /* 3D_RNDR_GEN_INDX_PRIM on r100/r200 */
1120                 r = r100_cs_packet_next_reloc(p, &reloc);
1121                 if (r) {
1122                         DRM_ERROR("No reloc for packet3 %d\n", pkt->opcode);
1123                         r100_cs_dump_packet(p, pkt);
1124                         return r;
1125                 }
1126                 ib[idx] = ib_chunk->kdata[idx] + ((u32)reloc->lobj.gpu_offset);
1127                 break;
1128         case PACKET3_3D_DRAW_IMMD:
1129                 /* triggers drawing using in-packet vertex data */
1130         case PACKET3_3D_DRAW_IMMD_2:
1131                 /* triggers drawing using in-packet vertex data */
1132         case PACKET3_3D_DRAW_VBUF_2:
1133                 /* triggers drawing of vertex buffers setup elsewhere */
1134         case PACKET3_3D_DRAW_INDX_2:
1135                 /* triggers drawing using indices to vertex buffer */
1136         case PACKET3_3D_DRAW_VBUF:
1137                 /* triggers drawing of vertex buffers setup elsewhere */
1138         case PACKET3_3D_DRAW_INDX:
1139                 /* triggers drawing using indices to vertex buffer */
1140         case PACKET3_NOP:
1141                 break;
1142         default:
1143                 DRM_ERROR("Packet3 opcode %x not supported\n", pkt->opcode);
1144                 return -EINVAL;
1145         }
1146         return 0;
1147 }
1148
1149 int r100_cs_parse(struct radeon_cs_parser *p)
1150 {
1151         struct radeon_cs_packet pkt;
1152         int r;
1153
1154         do {
1155                 r = r100_cs_packet_parse(p, &pkt, p->idx);
1156                 if (r) {
1157                         return r;
1158                 }
1159                 p->idx += pkt.count + 2;
1160                 switch (pkt.type) {
1161                         case PACKET_TYPE0:
1162                                 r = r100_packet0_check(p, &pkt);
1163                                 break;
1164                         case PACKET_TYPE2:
1165                                 break;
1166                         case PACKET_TYPE3:
1167                                 r = r100_packet3_check(p, &pkt);
1168                                 break;
1169                         default:
1170                                 DRM_ERROR("Unknown packet type %d !\n",
1171                                           pkt.type);
1172                                 return -EINVAL;
1173                 }
1174                 if (r) {
1175                         return r;
1176                 }
1177         } while (p->idx < p->chunks[p->chunk_ib_idx].length_dw);
1178         return 0;
1179 }
1180
1181
1182 /*
1183  * Global GPU functions
1184  */
1185 void r100_errata(struct radeon_device *rdev)
1186 {
1187         rdev->pll_errata = 0;
1188
1189         if (rdev->family == CHIP_RV200 || rdev->family == CHIP_RS200) {
1190                 rdev->pll_errata |= CHIP_ERRATA_PLL_DUMMYREADS;
1191         }
1192
1193         if (rdev->family == CHIP_RV100 ||
1194             rdev->family == CHIP_RS100 ||
1195             rdev->family == CHIP_RS200) {
1196                 rdev->pll_errata |= CHIP_ERRATA_PLL_DELAY;
1197         }
1198 }
1199
1200 /* Wait for vertical sync on primary CRTC */
1201 void r100_gpu_wait_for_vsync(struct radeon_device *rdev)
1202 {
1203         uint32_t crtc_gen_cntl, tmp;
1204         int i;
1205
1206         crtc_gen_cntl = RREG32(RADEON_CRTC_GEN_CNTL);
1207         if ((crtc_gen_cntl & RADEON_CRTC_DISP_REQ_EN_B) ||
1208             !(crtc_gen_cntl & RADEON_CRTC_EN)) {
1209                 return;
1210         }
1211         /* Clear the CRTC_VBLANK_SAVE bit */
1212         WREG32(RADEON_CRTC_STATUS, RADEON_CRTC_VBLANK_SAVE_CLEAR);
1213         for (i = 0; i < rdev->usec_timeout; i++) {
1214                 tmp = RREG32(RADEON_CRTC_STATUS);
1215                 if (tmp & RADEON_CRTC_VBLANK_SAVE) {
1216                         return;
1217                 }
1218                 DRM_UDELAY(1);
1219         }
1220 }
1221
1222 /* Wait for vertical sync on secondary CRTC */
1223 void r100_gpu_wait_for_vsync2(struct radeon_device *rdev)
1224 {
1225         uint32_t crtc2_gen_cntl, tmp;
1226         int i;
1227
1228         crtc2_gen_cntl = RREG32(RADEON_CRTC2_GEN_CNTL);
1229         if ((crtc2_gen_cntl & RADEON_CRTC2_DISP_REQ_EN_B) ||
1230             !(crtc2_gen_cntl & RADEON_CRTC2_EN))
1231                 return;
1232
1233         /* Clear the CRTC_VBLANK_SAVE bit */
1234         WREG32(RADEON_CRTC2_STATUS, RADEON_CRTC2_VBLANK_SAVE_CLEAR);
1235         for (i = 0; i < rdev->usec_timeout; i++) {
1236                 tmp = RREG32(RADEON_CRTC2_STATUS);
1237                 if (tmp & RADEON_CRTC2_VBLANK_SAVE) {
1238                         return;
1239                 }
1240                 DRM_UDELAY(1);
1241         }
1242 }
1243
1244 int r100_rbbm_fifo_wait_for_entry(struct radeon_device *rdev, unsigned n)
1245 {
1246         unsigned i;
1247         uint32_t tmp;
1248
1249         for (i = 0; i < rdev->usec_timeout; i++) {
1250                 tmp = RREG32(RADEON_RBBM_STATUS) & RADEON_RBBM_FIFOCNT_MASK;
1251                 if (tmp >= n) {
1252                         return 0;
1253                 }
1254                 DRM_UDELAY(1);
1255         }
1256         return -1;
1257 }
1258
1259 int r100_gui_wait_for_idle(struct radeon_device *rdev)
1260 {
1261         unsigned i;
1262         uint32_t tmp;
1263
1264         if (r100_rbbm_fifo_wait_for_entry(rdev, 64)) {
1265                 printk(KERN_WARNING "radeon: wait for empty RBBM fifo failed !"
1266                        " Bad things might happen.\n");
1267         }
1268         for (i = 0; i < rdev->usec_timeout; i++) {
1269                 tmp = RREG32(RADEON_RBBM_STATUS);
1270                 if (!(tmp & (1 << 31))) {
1271                         return 0;
1272                 }
1273                 DRM_UDELAY(1);
1274         }
1275         return -1;
1276 }
1277
1278 int r100_mc_wait_for_idle(struct radeon_device *rdev)
1279 {
1280         unsigned i;
1281         uint32_t tmp;
1282
1283         for (i = 0; i < rdev->usec_timeout; i++) {
1284                 /* read MC_STATUS */
1285                 tmp = RREG32(0x0150);
1286                 if (tmp & (1 << 2)) {
1287                         return 0;
1288                 }
1289                 DRM_UDELAY(1);
1290         }
1291         return -1;
1292 }
1293
1294 void r100_gpu_init(struct radeon_device *rdev)
1295 {
1296         /* TODO: anythings to do here ? pipes ? */
1297         r100_hdp_reset(rdev);
1298 }
1299
1300 void r100_hdp_reset(struct radeon_device *rdev)
1301 {
1302         uint32_t tmp;
1303
1304         tmp = RREG32(RADEON_HOST_PATH_CNTL) & RADEON_HDP_APER_CNTL;
1305         tmp |= (7 << 28);
1306         WREG32(RADEON_HOST_PATH_CNTL, tmp | RADEON_HDP_SOFT_RESET | RADEON_HDP_READ_BUFFER_INVALIDATE);
1307         (void)RREG32(RADEON_HOST_PATH_CNTL);
1308         udelay(200);
1309         WREG32(RADEON_RBBM_SOFT_RESET, 0);
1310         WREG32(RADEON_HOST_PATH_CNTL, tmp);
1311         (void)RREG32(RADEON_HOST_PATH_CNTL);
1312 }
1313
1314 int r100_rb2d_reset(struct radeon_device *rdev)
1315 {
1316         uint32_t tmp;
1317         int i;
1318
1319         WREG32(RADEON_RBBM_SOFT_RESET, RADEON_SOFT_RESET_E2);
1320         (void)RREG32(RADEON_RBBM_SOFT_RESET);
1321         udelay(200);
1322         WREG32(RADEON_RBBM_SOFT_RESET, 0);
1323         /* Wait to prevent race in RBBM_STATUS */
1324         mdelay(1);
1325         for (i = 0; i < rdev->usec_timeout; i++) {
1326                 tmp = RREG32(RADEON_RBBM_STATUS);
1327                 if (!(tmp & (1 << 26))) {
1328                         DRM_INFO("RB2D reset succeed (RBBM_STATUS=0x%08X)\n",
1329                                  tmp);
1330                         return 0;
1331                 }
1332                 DRM_UDELAY(1);
1333         }
1334         tmp = RREG32(RADEON_RBBM_STATUS);
1335         DRM_ERROR("Failed to reset RB2D (RBBM_STATUS=0x%08X)!\n", tmp);
1336         return -1;
1337 }
1338
1339 int r100_gpu_reset(struct radeon_device *rdev)
1340 {
1341         uint32_t status;
1342
1343         /* reset order likely matter */
1344         status = RREG32(RADEON_RBBM_STATUS);
1345         /* reset HDP */
1346         r100_hdp_reset(rdev);
1347         /* reset rb2d */
1348         if (status & ((1 << 17) | (1 << 18) | (1 << 27))) {
1349                 r100_rb2d_reset(rdev);
1350         }
1351         /* TODO: reset 3D engine */
1352         /* reset CP */
1353         status = RREG32(RADEON_RBBM_STATUS);
1354         if (status & (1 << 16)) {
1355                 r100_cp_reset(rdev);
1356         }
1357         /* Check if GPU is idle */
1358         status = RREG32(RADEON_RBBM_STATUS);
1359         if (status & (1 << 31)) {
1360                 DRM_ERROR("Failed to reset GPU (RBBM_STATUS=0x%08X)\n", status);
1361                 return -1;
1362         }
1363         DRM_INFO("GPU reset succeed (RBBM_STATUS=0x%08X)\n", status);
1364         return 0;
1365 }
1366
1367
1368 /*
1369  * VRAM info
1370  */
1371 static void r100_vram_get_type(struct radeon_device *rdev)
1372 {
1373         uint32_t tmp;
1374
1375         rdev->mc.vram_is_ddr = false;
1376         if (rdev->flags & RADEON_IS_IGP)
1377                 rdev->mc.vram_is_ddr = true;
1378         else if (RREG32(RADEON_MEM_SDRAM_MODE_REG) & RADEON_MEM_CFG_TYPE_DDR)
1379                 rdev->mc.vram_is_ddr = true;
1380         if ((rdev->family == CHIP_RV100) ||
1381             (rdev->family == CHIP_RS100) ||
1382             (rdev->family == CHIP_RS200)) {
1383                 tmp = RREG32(RADEON_MEM_CNTL);
1384                 if (tmp & RV100_HALF_MODE) {
1385                         rdev->mc.vram_width = 32;
1386                 } else {
1387                         rdev->mc.vram_width = 64;
1388                 }
1389                 if (rdev->flags & RADEON_SINGLE_CRTC) {
1390                         rdev->mc.vram_width /= 4;
1391                         rdev->mc.vram_is_ddr = true;
1392                 }
1393         } else if (rdev->family <= CHIP_RV280) {
1394                 tmp = RREG32(RADEON_MEM_CNTL);
1395                 if (tmp & RADEON_MEM_NUM_CHANNELS_MASK) {
1396                         rdev->mc.vram_width = 128;
1397                 } else {
1398                         rdev->mc.vram_width = 64;
1399                 }
1400         } else {
1401                 /* newer IGPs */
1402                 rdev->mc.vram_width = 128;
1403         }
1404 }
1405
1406 static u32 r100_get_accessible_vram(struct radeon_device *rdev)
1407 {
1408         u32 aper_size;
1409         u8 byte;
1410
1411         aper_size = RREG32(RADEON_CONFIG_APER_SIZE);
1412
1413         /* Set HDP_APER_CNTL only on cards that are known not to be broken,
1414          * that is has the 2nd generation multifunction PCI interface
1415          */
1416         if (rdev->family == CHIP_RV280 ||
1417             rdev->family >= CHIP_RV350) {
1418                 WREG32_P(RADEON_HOST_PATH_CNTL, RADEON_HDP_APER_CNTL,
1419                        ~RADEON_HDP_APER_CNTL);
1420                 DRM_INFO("Generation 2 PCI interface, using max accessible memory\n");
1421                 return aper_size * 2;
1422         }
1423
1424         /* Older cards have all sorts of funny issues to deal with. First
1425          * check if it's a multifunction card by reading the PCI config
1426          * header type... Limit those to one aperture size
1427          */
1428         pci_read_config_byte(rdev->pdev, 0xe, &byte);
1429         if (byte & 0x80) {
1430                 DRM_INFO("Generation 1 PCI interface in multifunction mode\n");
1431                 DRM_INFO("Limiting VRAM to one aperture\n");
1432                 return aper_size;
1433         }
1434
1435         /* Single function older card. We read HDP_APER_CNTL to see how the BIOS
1436          * have set it up. We don't write this as it's broken on some ASICs but
1437          * we expect the BIOS to have done the right thing (might be too optimistic...)
1438          */
1439         if (RREG32(RADEON_HOST_PATH_CNTL) & RADEON_HDP_APER_CNTL)
1440                 return aper_size * 2;
1441         return aper_size;
1442 }
1443
1444 void r100_vram_init_sizes(struct radeon_device *rdev)
1445 {
1446         u64 config_aper_size;
1447         u32 accessible;
1448
1449         config_aper_size = RREG32(RADEON_CONFIG_APER_SIZE);
1450
1451         if (rdev->flags & RADEON_IS_IGP) {
1452                 uint32_t tom;
1453                 /* read NB_TOM to get the amount of ram stolen for the GPU */
1454                 tom = RREG32(RADEON_NB_TOM);
1455                 rdev->mc.real_vram_size = (((tom >> 16) - (tom & 0xffff) + 1) << 16);
1456                 /* for IGPs we need to keep VRAM where it was put by the BIOS */
1457                 rdev->mc.vram_location = (tom & 0xffff) << 16;
1458                 WREG32(RADEON_CONFIG_MEMSIZE, rdev->mc.real_vram_size);
1459                 rdev->mc.mc_vram_size = rdev->mc.real_vram_size;
1460         } else {
1461                 rdev->mc.real_vram_size = RREG32(RADEON_CONFIG_MEMSIZE);
1462                 /* Some production boards of m6 will report 0
1463                  * if it's 8 MB
1464                  */
1465                 if (rdev->mc.real_vram_size == 0) {
1466                         rdev->mc.real_vram_size = 8192 * 1024;
1467                         WREG32(RADEON_CONFIG_MEMSIZE, rdev->mc.real_vram_size);
1468                 }
1469                 /* let driver place VRAM */
1470                 rdev->mc.vram_location = 0xFFFFFFFFUL;
1471                  /* Fix for RN50, M6, M7 with 8/16/32(??) MBs of VRAM -
1472                   * Novell bug 204882 + along with lots of ubuntu ones */
1473                 if (config_aper_size > rdev->mc.real_vram_size)
1474                         rdev->mc.mc_vram_size = config_aper_size;
1475                 else
1476                         rdev->mc.mc_vram_size = rdev->mc.real_vram_size;
1477         }
1478
1479         /* work out accessible VRAM */
1480         accessible = r100_get_accessible_vram(rdev);
1481
1482         rdev->mc.aper_base = drm_get_resource_start(rdev->ddev, 0);
1483         rdev->mc.aper_size = drm_get_resource_len(rdev->ddev, 0);
1484
1485         if (accessible > rdev->mc.aper_size)
1486                 accessible = rdev->mc.aper_size;
1487
1488         if (rdev->mc.mc_vram_size > rdev->mc.aper_size)
1489                 rdev->mc.mc_vram_size = rdev->mc.aper_size;
1490
1491         if (rdev->mc.real_vram_size > rdev->mc.aper_size)
1492                 rdev->mc.real_vram_size = rdev->mc.aper_size;
1493 }
1494
1495 void r100_vram_info(struct radeon_device *rdev)
1496 {
1497         r100_vram_get_type(rdev);
1498
1499         r100_vram_init_sizes(rdev);
1500 }
1501
1502
1503 /*
1504  * Indirect registers accessor
1505  */
1506 void r100_pll_errata_after_index(struct radeon_device *rdev)
1507 {
1508         if (!(rdev->pll_errata & CHIP_ERRATA_PLL_DUMMYREADS)) {
1509                 return;
1510         }
1511         (void)RREG32(RADEON_CLOCK_CNTL_DATA);
1512         (void)RREG32(RADEON_CRTC_GEN_CNTL);
1513 }
1514
1515 static void r100_pll_errata_after_data(struct radeon_device *rdev)
1516 {
1517         /* This workarounds is necessary on RV100, RS100 and RS200 chips
1518          * or the chip could hang on a subsequent access
1519          */
1520         if (rdev->pll_errata & CHIP_ERRATA_PLL_DELAY) {
1521                 udelay(5000);
1522         }
1523
1524         /* This function is required to workaround a hardware bug in some (all?)
1525          * revisions of the R300.  This workaround should be called after every
1526          * CLOCK_CNTL_INDEX register access.  If not, register reads afterward
1527          * may not be correct.
1528          */
1529         if (rdev->pll_errata & CHIP_ERRATA_R300_CG) {
1530                 uint32_t save, tmp;
1531
1532                 save = RREG32(RADEON_CLOCK_CNTL_INDEX);
1533                 tmp = save & ~(0x3f | RADEON_PLL_WR_EN);
1534                 WREG32(RADEON_CLOCK_CNTL_INDEX, tmp);
1535                 tmp = RREG32(RADEON_CLOCK_CNTL_DATA);
1536                 WREG32(RADEON_CLOCK_CNTL_INDEX, save);
1537         }
1538 }
1539
1540 uint32_t r100_pll_rreg(struct radeon_device *rdev, uint32_t reg)
1541 {
1542         uint32_t data;
1543
1544         WREG8(RADEON_CLOCK_CNTL_INDEX, reg & 0x3f);
1545         r100_pll_errata_after_index(rdev);
1546         data = RREG32(RADEON_CLOCK_CNTL_DATA);
1547         r100_pll_errata_after_data(rdev);
1548         return data;
1549 }
1550
1551 void r100_pll_wreg(struct radeon_device *rdev, uint32_t reg, uint32_t v)
1552 {
1553         WREG8(RADEON_CLOCK_CNTL_INDEX, ((reg & 0x3f) | RADEON_PLL_WR_EN));
1554         r100_pll_errata_after_index(rdev);
1555         WREG32(RADEON_CLOCK_CNTL_DATA, v);
1556         r100_pll_errata_after_data(rdev);
1557 }
1558
1559 uint32_t r100_mm_rreg(struct radeon_device *rdev, uint32_t reg)
1560 {
1561         if (reg < 0x10000)
1562                 return readl(((void __iomem *)rdev->rmmio) + reg);
1563         else {
1564                 writel(reg, ((void __iomem *)rdev->rmmio) + RADEON_MM_INDEX);
1565                 return readl(((void __iomem *)rdev->rmmio) + RADEON_MM_DATA);
1566         }
1567 }
1568
1569 void r100_mm_wreg(struct radeon_device *rdev, uint32_t reg, uint32_t v)
1570 {
1571         if (reg < 0x10000)
1572                 writel(v, ((void __iomem *)rdev->rmmio) + reg);
1573         else {
1574                 writel(reg, ((void __iomem *)rdev->rmmio) + RADEON_MM_INDEX);
1575                 writel(v, ((void __iomem *)rdev->rmmio) + RADEON_MM_DATA);
1576         }
1577 }
1578
1579 int r100_init(struct radeon_device *rdev)
1580 {
1581         return 0;
1582 }
1583
1584 /*
1585  * Debugfs info
1586  */
1587 #if defined(CONFIG_DEBUG_FS)
1588 static int r100_debugfs_rbbm_info(struct seq_file *m, void *data)
1589 {
1590         struct drm_info_node *node = (struct drm_info_node *) m->private;
1591         struct drm_device *dev = node->minor->dev;
1592         struct radeon_device *rdev = dev->dev_private;
1593         uint32_t reg, value;
1594         unsigned i;
1595
1596         seq_printf(m, "RBBM_STATUS 0x%08x\n", RREG32(RADEON_RBBM_STATUS));
1597         seq_printf(m, "RBBM_CMDFIFO_STAT 0x%08x\n", RREG32(0xE7C));
1598         seq_printf(m, "CP_STAT 0x%08x\n", RREG32(RADEON_CP_STAT));
1599         for (i = 0; i < 64; i++) {
1600                 WREG32(RADEON_RBBM_CMDFIFO_ADDR, i | 0x100);
1601                 reg = (RREG32(RADEON_RBBM_CMDFIFO_DATA) - 1) >> 2;
1602                 WREG32(RADEON_RBBM_CMDFIFO_ADDR, i);
1603                 value = RREG32(RADEON_RBBM_CMDFIFO_DATA);
1604                 seq_printf(m, "[0x%03X] 0x%04X=0x%08X\n", i, reg, value);
1605         }
1606         return 0;
1607 }
1608
1609 static int r100_debugfs_cp_ring_info(struct seq_file *m, void *data)
1610 {
1611         struct drm_info_node *node = (struct drm_info_node *) m->private;
1612         struct drm_device *dev = node->minor->dev;
1613         struct radeon_device *rdev = dev->dev_private;
1614         uint32_t rdp, wdp;
1615         unsigned count, i, j;
1616
1617         radeon_ring_free_size(rdev);
1618         rdp = RREG32(RADEON_CP_RB_RPTR);
1619         wdp = RREG32(RADEON_CP_RB_WPTR);
1620         count = (rdp + rdev->cp.ring_size - wdp) & rdev->cp.ptr_mask;
1621         seq_printf(m, "CP_STAT 0x%08x\n", RREG32(RADEON_CP_STAT));
1622         seq_printf(m, "CP_RB_WPTR 0x%08x\n", wdp);
1623         seq_printf(m, "CP_RB_RPTR 0x%08x\n", rdp);
1624         seq_printf(m, "%u free dwords in ring\n", rdev->cp.ring_free_dw);
1625         seq_printf(m, "%u dwords in ring\n", count);
1626         for (j = 0; j <= count; j++) {
1627                 i = (rdp + j) & rdev->cp.ptr_mask;
1628                 seq_printf(m, "r[%04d]=0x%08x\n", i, rdev->cp.ring[i]);
1629         }
1630         return 0;
1631 }
1632
1633
1634 static int r100_debugfs_cp_csq_fifo(struct seq_file *m, void *data)
1635 {
1636         struct drm_info_node *node = (struct drm_info_node *) m->private;
1637         struct drm_device *dev = node->minor->dev;
1638         struct radeon_device *rdev = dev->dev_private;
1639         uint32_t csq_stat, csq2_stat, tmp;
1640         unsigned r_rptr, r_wptr, ib1_rptr, ib1_wptr, ib2_rptr, ib2_wptr;
1641         unsigned i;
1642
1643         seq_printf(m, "CP_STAT 0x%08x\n", RREG32(RADEON_CP_STAT));
1644         seq_printf(m, "CP_CSQ_MODE 0x%08x\n", RREG32(RADEON_CP_CSQ_MODE));
1645         csq_stat = RREG32(RADEON_CP_CSQ_STAT);
1646         csq2_stat = RREG32(RADEON_CP_CSQ2_STAT);
1647         r_rptr = (csq_stat >> 0) & 0x3ff;
1648         r_wptr = (csq_stat >> 10) & 0x3ff;
1649         ib1_rptr = (csq_stat >> 20) & 0x3ff;
1650         ib1_wptr = (csq2_stat >> 0) & 0x3ff;
1651         ib2_rptr = (csq2_stat >> 10) & 0x3ff;
1652         ib2_wptr = (csq2_stat >> 20) & 0x3ff;
1653         seq_printf(m, "CP_CSQ_STAT 0x%08x\n", csq_stat);
1654         seq_printf(m, "CP_CSQ2_STAT 0x%08x\n", csq2_stat);
1655         seq_printf(m, "Ring rptr %u\n", r_rptr);
1656         seq_printf(m, "Ring wptr %u\n", r_wptr);
1657         seq_printf(m, "Indirect1 rptr %u\n", ib1_rptr);
1658         seq_printf(m, "Indirect1 wptr %u\n", ib1_wptr);
1659         seq_printf(m, "Indirect2 rptr %u\n", ib2_rptr);
1660         seq_printf(m, "Indirect2 wptr %u\n", ib2_wptr);
1661         /* FIXME: 0, 128, 640 depends on fifo setup see cp_init_kms
1662          * 128 = indirect1_start * 8 & 640 = indirect2_start * 8 */
1663         seq_printf(m, "Ring fifo:\n");
1664         for (i = 0; i < 256; i++) {
1665                 WREG32(RADEON_CP_CSQ_ADDR, i << 2);
1666                 tmp = RREG32(RADEON_CP_CSQ_DATA);
1667                 seq_printf(m, "rfifo[%04d]=0x%08X\n", i, tmp);
1668         }
1669         seq_printf(m, "Indirect1 fifo:\n");
1670         for (i = 256; i <= 512; i++) {
1671                 WREG32(RADEON_CP_CSQ_ADDR, i << 2);
1672                 tmp = RREG32(RADEON_CP_CSQ_DATA);
1673                 seq_printf(m, "ib1fifo[%04d]=0x%08X\n", i, tmp);
1674         }
1675         seq_printf(m, "Indirect2 fifo:\n");
1676         for (i = 640; i < ib1_wptr; i++) {
1677                 WREG32(RADEON_CP_CSQ_ADDR, i << 2);
1678                 tmp = RREG32(RADEON_CP_CSQ_DATA);
1679                 seq_printf(m, "ib2fifo[%04d]=0x%08X\n", i, tmp);
1680         }
1681         return 0;
1682 }
1683
1684 static int r100_debugfs_mc_info(struct seq_file *m, void *data)
1685 {
1686         struct drm_info_node *node = (struct drm_info_node *) m->private;
1687         struct drm_device *dev = node->minor->dev;
1688         struct radeon_device *rdev = dev->dev_private;
1689         uint32_t tmp;
1690
1691         tmp = RREG32(RADEON_CONFIG_MEMSIZE);
1692         seq_printf(m, "CONFIG_MEMSIZE 0x%08x\n", tmp);
1693         tmp = RREG32(RADEON_MC_FB_LOCATION);
1694         seq_printf(m, "MC_FB_LOCATION 0x%08x\n", tmp);
1695         tmp = RREG32(RADEON_BUS_CNTL);
1696         seq_printf(m, "BUS_CNTL 0x%08x\n", tmp);
1697         tmp = RREG32(RADEON_MC_AGP_LOCATION);
1698         seq_printf(m, "MC_AGP_LOCATION 0x%08x\n", tmp);
1699         tmp = RREG32(RADEON_AGP_BASE);
1700         seq_printf(m, "AGP_BASE 0x%08x\n", tmp);
1701         tmp = RREG32(RADEON_HOST_PATH_CNTL);
1702         seq_printf(m, "HOST_PATH_CNTL 0x%08x\n", tmp);
1703         tmp = RREG32(0x01D0);
1704         seq_printf(m, "AIC_CTRL 0x%08x\n", tmp);
1705         tmp = RREG32(RADEON_AIC_LO_ADDR);
1706         seq_printf(m, "AIC_LO_ADDR 0x%08x\n", tmp);
1707         tmp = RREG32(RADEON_AIC_HI_ADDR);
1708         seq_printf(m, "AIC_HI_ADDR 0x%08x\n", tmp);
1709         tmp = RREG32(0x01E4);
1710         seq_printf(m, "AIC_TLB_ADDR 0x%08x\n", tmp);
1711         return 0;
1712 }
1713
1714 static struct drm_info_list r100_debugfs_rbbm_list[] = {
1715         {"r100_rbbm_info", r100_debugfs_rbbm_info, 0, NULL},
1716 };
1717
1718 static struct drm_info_list r100_debugfs_cp_list[] = {
1719         {"r100_cp_ring_info", r100_debugfs_cp_ring_info, 0, NULL},
1720         {"r100_cp_csq_fifo", r100_debugfs_cp_csq_fifo, 0, NULL},
1721 };
1722
1723 static struct drm_info_list r100_debugfs_mc_info_list[] = {
1724         {"r100_mc_info", r100_debugfs_mc_info, 0, NULL},
1725 };
1726 #endif
1727
1728 int r100_debugfs_rbbm_init(struct radeon_device *rdev)
1729 {
1730 #if defined(CONFIG_DEBUG_FS)
1731         return radeon_debugfs_add_files(rdev, r100_debugfs_rbbm_list, 1);
1732 #else
1733         return 0;
1734 #endif
1735 }
1736
1737 int r100_debugfs_cp_init(struct radeon_device *rdev)
1738 {
1739 #if defined(CONFIG_DEBUG_FS)
1740         return radeon_debugfs_add_files(rdev, r100_debugfs_cp_list, 2);
1741 #else
1742         return 0;
1743 #endif
1744 }
1745
1746 int r100_debugfs_mc_info_init(struct radeon_device *rdev)
1747 {
1748 #if defined(CONFIG_DEBUG_FS)
1749         return radeon_debugfs_add_files(rdev, r100_debugfs_mc_info_list, 1);
1750 #else
1751         return 0;
1752 #endif
1753 }
1754
1755 int r100_set_surface_reg(struct radeon_device *rdev, int reg,
1756                          uint32_t tiling_flags, uint32_t pitch,
1757                          uint32_t offset, uint32_t obj_size)
1758 {
1759         int surf_index = reg * 16;
1760         int flags = 0;
1761
1762         /* r100/r200 divide by 16 */
1763         if (rdev->family < CHIP_R300)
1764                 flags = pitch / 16;
1765         else
1766                 flags = pitch / 8;
1767
1768         if (rdev->family <= CHIP_RS200) {
1769                 if ((tiling_flags & (RADEON_TILING_MACRO|RADEON_TILING_MICRO))
1770                                  == (RADEON_TILING_MACRO|RADEON_TILING_MICRO))
1771                         flags |= RADEON_SURF_TILE_COLOR_BOTH;
1772                 if (tiling_flags & RADEON_TILING_MACRO)
1773                         flags |= RADEON_SURF_TILE_COLOR_MACRO;
1774         } else if (rdev->family <= CHIP_RV280) {
1775                 if (tiling_flags & (RADEON_TILING_MACRO))
1776                         flags |= R200_SURF_TILE_COLOR_MACRO;
1777                 if (tiling_flags & RADEON_TILING_MICRO)
1778                         flags |= R200_SURF_TILE_COLOR_MICRO;
1779         } else {
1780                 if (tiling_flags & RADEON_TILING_MACRO)
1781                         flags |= R300_SURF_TILE_MACRO;
1782                 if (tiling_flags & RADEON_TILING_MICRO)
1783                         flags |= R300_SURF_TILE_MICRO;
1784         }
1785
1786         DRM_DEBUG("writing surface %d %d %x %x\n", reg, flags, offset, offset+obj_size-1);
1787         WREG32(RADEON_SURFACE0_INFO + surf_index, flags);
1788         WREG32(RADEON_SURFACE0_LOWER_BOUND + surf_index, offset);
1789         WREG32(RADEON_SURFACE0_UPPER_BOUND + surf_index, offset + obj_size - 1);
1790         return 0;
1791 }
1792
1793 void r100_clear_surface_reg(struct radeon_device *rdev, int reg)
1794 {
1795         int surf_index = reg * 16;
1796         WREG32(RADEON_SURFACE0_INFO + surf_index, 0);
1797 }
1798
1799 void r100_bandwidth_update(struct radeon_device *rdev)
1800 {
1801         fixed20_12 trcd_ff, trp_ff, tras_ff, trbs_ff, tcas_ff;
1802         fixed20_12 sclk_ff, mclk_ff, sclk_eff_ff, sclk_delay_ff;
1803         fixed20_12 peak_disp_bw, mem_bw, pix_clk, pix_clk2, temp_ff, crit_point_ff;
1804         uint32_t temp, data, mem_trcd, mem_trp, mem_tras;
1805         fixed20_12 memtcas_ff[8] = {
1806                 fixed_init(1),
1807                 fixed_init(2),
1808                 fixed_init(3),
1809                 fixed_init(0),
1810                 fixed_init_half(1),
1811                 fixed_init_half(2),
1812                 fixed_init(0),
1813         };
1814         fixed20_12 memtcas_rs480_ff[8] = {
1815                 fixed_init(0),
1816                 fixed_init(1),
1817                 fixed_init(2),
1818                 fixed_init(3),
1819                 fixed_init(0),
1820                 fixed_init_half(1),
1821                 fixed_init_half(2),
1822                 fixed_init_half(3),
1823         };
1824         fixed20_12 memtcas2_ff[8] = {
1825                 fixed_init(0),
1826                 fixed_init(1),
1827                 fixed_init(2),
1828                 fixed_init(3),
1829                 fixed_init(4),
1830                 fixed_init(5),
1831                 fixed_init(6),
1832                 fixed_init(7),
1833         };
1834         fixed20_12 memtrbs[8] = {
1835                 fixed_init(1),
1836                 fixed_init_half(1),
1837                 fixed_init(2),
1838                 fixed_init_half(2),
1839                 fixed_init(3),
1840                 fixed_init_half(3),
1841                 fixed_init(4),
1842                 fixed_init_half(4)
1843         };
1844         fixed20_12 memtrbs_r4xx[8] = {
1845                 fixed_init(4),
1846                 fixed_init(5),
1847                 fixed_init(6),
1848                 fixed_init(7),
1849                 fixed_init(8),
1850                 fixed_init(9),
1851                 fixed_init(10),
1852                 fixed_init(11)
1853         };
1854         fixed20_12 min_mem_eff;
1855         fixed20_12 mc_latency_sclk, mc_latency_mclk, k1;
1856         fixed20_12 cur_latency_mclk, cur_latency_sclk;
1857         fixed20_12 disp_latency, disp_latency_overhead, disp_drain_rate,
1858                 disp_drain_rate2, read_return_rate;
1859         fixed20_12 time_disp1_drop_priority;
1860         int c;
1861         int cur_size = 16;       /* in octawords */
1862         int critical_point = 0, critical_point2;
1863 /*      uint32_t read_return_rate, time_disp1_drop_priority; */
1864         int stop_req, max_stop_req;
1865         struct drm_display_mode *mode1 = NULL;
1866         struct drm_display_mode *mode2 = NULL;
1867         uint32_t pixel_bytes1 = 0;
1868         uint32_t pixel_bytes2 = 0;
1869
1870         if (rdev->mode_info.crtcs[0]->base.enabled) {
1871                 mode1 = &rdev->mode_info.crtcs[0]->base.mode;
1872                 pixel_bytes1 = rdev->mode_info.crtcs[0]->base.fb->bits_per_pixel / 8;
1873         }
1874         if (rdev->mode_info.crtcs[1]->base.enabled) {
1875                 mode2 = &rdev->mode_info.crtcs[1]->base.mode;
1876                 pixel_bytes2 = rdev->mode_info.crtcs[1]->base.fb->bits_per_pixel / 8;
1877         }
1878
1879         min_mem_eff.full = rfixed_const_8(0);
1880         /* get modes */
1881         if ((rdev->disp_priority == 2) && ASIC_IS_R300(rdev)) {
1882                 uint32_t mc_init_misc_lat_timer = RREG32(R300_MC_INIT_MISC_LAT_TIMER);
1883                 mc_init_misc_lat_timer &= ~(R300_MC_DISP1R_INIT_LAT_MASK << R300_MC_DISP1R_INIT_LAT_SHIFT);
1884                 mc_init_misc_lat_timer &= ~(R300_MC_DISP0R_INIT_LAT_MASK << R300_MC_DISP0R_INIT_LAT_SHIFT);
1885                 /* check crtc enables */
1886                 if (mode2)
1887                         mc_init_misc_lat_timer |= (1 << R300_MC_DISP1R_INIT_LAT_SHIFT);
1888                 if (mode1)
1889                         mc_init_misc_lat_timer |= (1 << R300_MC_DISP0R_INIT_LAT_SHIFT);
1890                 WREG32(R300_MC_INIT_MISC_LAT_TIMER, mc_init_misc_lat_timer);
1891         }
1892
1893         /*
1894          * determine is there is enough bw for current mode
1895          */
1896         mclk_ff.full = rfixed_const(rdev->clock.default_mclk);
1897         temp_ff.full = rfixed_const(100);
1898         mclk_ff.full = rfixed_div(mclk_ff, temp_ff);
1899         sclk_ff.full = rfixed_const(rdev->clock.default_sclk);
1900         sclk_ff.full = rfixed_div(sclk_ff, temp_ff);
1901
1902         temp = (rdev->mc.vram_width / 8) * (rdev->mc.vram_is_ddr ? 2 : 1);
1903         temp_ff.full = rfixed_const(temp);
1904         mem_bw.full = rfixed_mul(mclk_ff, temp_ff);
1905
1906         pix_clk.full = 0;
1907         pix_clk2.full = 0;
1908         peak_disp_bw.full = 0;
1909         if (mode1) {
1910                 temp_ff.full = rfixed_const(1000);
1911                 pix_clk.full = rfixed_const(mode1->clock); /* convert to fixed point */
1912                 pix_clk.full = rfixed_div(pix_clk, temp_ff);
1913                 temp_ff.full = rfixed_const(pixel_bytes1);
1914                 peak_disp_bw.full += rfixed_mul(pix_clk, temp_ff);
1915         }
1916         if (mode2) {
1917                 temp_ff.full = rfixed_const(1000);
1918                 pix_clk2.full = rfixed_const(mode2->clock); /* convert to fixed point */
1919                 pix_clk2.full = rfixed_div(pix_clk2, temp_ff);
1920                 temp_ff.full = rfixed_const(pixel_bytes2);
1921                 peak_disp_bw.full += rfixed_mul(pix_clk2, temp_ff);
1922         }
1923
1924         mem_bw.full = rfixed_mul(mem_bw, min_mem_eff);
1925         if (peak_disp_bw.full >= mem_bw.full) {
1926                 DRM_ERROR("You may not have enough display bandwidth for current mode\n"
1927                           "If you have flickering problem, try to lower resolution, refresh rate, or color depth\n");
1928         }
1929
1930         /*  Get values from the EXT_MEM_CNTL register...converting its contents. */
1931         temp = RREG32(RADEON_MEM_TIMING_CNTL);
1932         if ((rdev->family == CHIP_RV100) || (rdev->flags & RADEON_IS_IGP)) { /* RV100, M6, IGPs */
1933                 mem_trcd = ((temp >> 2) & 0x3) + 1;
1934                 mem_trp  = ((temp & 0x3)) + 1;
1935                 mem_tras = ((temp & 0x70) >> 4) + 1;
1936         } else if (rdev->family == CHIP_R300 ||
1937                    rdev->family == CHIP_R350) { /* r300, r350 */
1938                 mem_trcd = (temp & 0x7) + 1;
1939                 mem_trp = ((temp >> 8) & 0x7) + 1;
1940                 mem_tras = ((temp >> 11) & 0xf) + 4;
1941         } else if (rdev->family == CHIP_RV350 ||
1942                    rdev->family <= CHIP_RV380) {
1943                 /* rv3x0 */
1944                 mem_trcd = (temp & 0x7) + 3;
1945                 mem_trp = ((temp >> 8) & 0x7) + 3;
1946                 mem_tras = ((temp >> 11) & 0xf) + 6;
1947         } else if (rdev->family == CHIP_R420 ||
1948                    rdev->family == CHIP_R423 ||
1949                    rdev->family == CHIP_RV410) {
1950                 /* r4xx */
1951                 mem_trcd = (temp & 0xf) + 3;
1952                 if (mem_trcd > 15)
1953                         mem_trcd = 15;
1954                 mem_trp = ((temp >> 8) & 0xf) + 3;
1955                 if (mem_trp > 15)
1956                         mem_trp = 15;
1957                 mem_tras = ((temp >> 12) & 0x1f) + 6;
1958                 if (mem_tras > 31)
1959                         mem_tras = 31;
1960         } else { /* RV200, R200 */
1961                 mem_trcd = (temp & 0x7) + 1;
1962                 mem_trp = ((temp >> 8) & 0x7) + 1;
1963                 mem_tras = ((temp >> 12) & 0xf) + 4;
1964         }
1965         /* convert to FF */
1966         trcd_ff.full = rfixed_const(mem_trcd);
1967         trp_ff.full = rfixed_const(mem_trp);
1968         tras_ff.full = rfixed_const(mem_tras);
1969
1970         /* Get values from the MEM_SDRAM_MODE_REG register...converting its */
1971         temp = RREG32(RADEON_MEM_SDRAM_MODE_REG);
1972         data = (temp & (7 << 20)) >> 20;
1973         if ((rdev->family == CHIP_RV100) || rdev->flags & RADEON_IS_IGP) {
1974                 if (rdev->family == CHIP_RS480) /* don't think rs400 */
1975                         tcas_ff = memtcas_rs480_ff[data];
1976                 else
1977                         tcas_ff = memtcas_ff[data];
1978         } else
1979                 tcas_ff = memtcas2_ff[data];
1980
1981         if (rdev->family == CHIP_RS400 ||
1982             rdev->family == CHIP_RS480) {
1983                 /* extra cas latency stored in bits 23-25 0-4 clocks */
1984                 data = (temp >> 23) & 0x7;
1985                 if (data < 5)
1986                         tcas_ff.full += rfixed_const(data);
1987         }
1988
1989         if (ASIC_IS_R300(rdev) && !(rdev->flags & RADEON_IS_IGP)) {
1990                 /* on the R300, Tcas is included in Trbs.
1991                  */
1992                 temp = RREG32(RADEON_MEM_CNTL);
1993                 data = (R300_MEM_NUM_CHANNELS_MASK & temp);
1994                 if (data == 1) {
1995                         if (R300_MEM_USE_CD_CH_ONLY & temp) {
1996                                 temp = RREG32(R300_MC_IND_INDEX);
1997                                 temp &= ~R300_MC_IND_ADDR_MASK;
1998                                 temp |= R300_MC_READ_CNTL_CD_mcind;
1999                                 WREG32(R300_MC_IND_INDEX, temp);
2000                                 temp = RREG32(R300_MC_IND_DATA);
2001                                 data = (R300_MEM_RBS_POSITION_C_MASK & temp);
2002                         } else {
2003                                 temp = RREG32(R300_MC_READ_CNTL_AB);
2004                                 data = (R300_MEM_RBS_POSITION_A_MASK & temp);
2005                         }
2006                 } else {
2007                         temp = RREG32(R300_MC_READ_CNTL_AB);
2008                         data = (R300_MEM_RBS_POSITION_A_MASK & temp);
2009                 }
2010                 if (rdev->family == CHIP_RV410 ||
2011                     rdev->family == CHIP_R420 ||
2012                     rdev->family == CHIP_R423)
2013                         trbs_ff = memtrbs_r4xx[data];
2014                 else
2015                         trbs_ff = memtrbs[data];
2016                 tcas_ff.full += trbs_ff.full;
2017         }
2018
2019         sclk_eff_ff.full = sclk_ff.full;
2020
2021         if (rdev->flags & RADEON_IS_AGP) {
2022                 fixed20_12 agpmode_ff;
2023                 agpmode_ff.full = rfixed_const(radeon_agpmode);
2024                 temp_ff.full = rfixed_const_666(16);
2025                 sclk_eff_ff.full -= rfixed_mul(agpmode_ff, temp_ff);
2026         }
2027         /* TODO PCIE lanes may affect this - agpmode == 16?? */
2028
2029         if (ASIC_IS_R300(rdev)) {
2030                 sclk_delay_ff.full = rfixed_const(250);
2031         } else {
2032                 if ((rdev->family == CHIP_RV100) ||
2033                     rdev->flags & RADEON_IS_IGP) {
2034                         if (rdev->mc.vram_is_ddr)
2035                                 sclk_delay_ff.full = rfixed_const(41);
2036                         else
2037                                 sclk_delay_ff.full = rfixed_const(33);
2038                 } else {
2039                         if (rdev->mc.vram_width == 128)
2040                                 sclk_delay_ff.full = rfixed_const(57);
2041                         else
2042                                 sclk_delay_ff.full = rfixed_const(41);
2043                 }
2044         }
2045
2046         mc_latency_sclk.full = rfixed_div(sclk_delay_ff, sclk_eff_ff);
2047
2048         if (rdev->mc.vram_is_ddr) {
2049                 if (rdev->mc.vram_width == 32) {
2050                         k1.full = rfixed_const(40);
2051                         c  = 3;
2052                 } else {
2053                         k1.full = rfixed_const(20);
2054                         c  = 1;
2055                 }
2056         } else {
2057                 k1.full = rfixed_const(40);
2058                 c  = 3;
2059         }
2060
2061         temp_ff.full = rfixed_const(2);
2062         mc_latency_mclk.full = rfixed_mul(trcd_ff, temp_ff);
2063         temp_ff.full = rfixed_const(c);
2064         mc_latency_mclk.full += rfixed_mul(tcas_ff, temp_ff);
2065         temp_ff.full = rfixed_const(4);
2066         mc_latency_mclk.full += rfixed_mul(tras_ff, temp_ff);
2067         mc_latency_mclk.full += rfixed_mul(trp_ff, temp_ff);
2068         mc_latency_mclk.full += k1.full;
2069
2070         mc_latency_mclk.full = rfixed_div(mc_latency_mclk, mclk_ff);
2071         mc_latency_mclk.full += rfixed_div(temp_ff, sclk_eff_ff);
2072
2073         /*
2074           HW cursor time assuming worst case of full size colour cursor.
2075         */
2076         temp_ff.full = rfixed_const((2 * (cur_size - (rdev->mc.vram_is_ddr + 1))));
2077         temp_ff.full += trcd_ff.full;
2078         if (temp_ff.full < tras_ff.full)
2079                 temp_ff.full = tras_ff.full;
2080         cur_latency_mclk.full = rfixed_div(temp_ff, mclk_ff);
2081
2082         temp_ff.full = rfixed_const(cur_size);
2083         cur_latency_sclk.full = rfixed_div(temp_ff, sclk_eff_ff);
2084         /*
2085           Find the total latency for the display data.
2086         */
2087         disp_latency_overhead.full = rfixed_const(80);
2088         disp_latency_overhead.full = rfixed_div(disp_latency_overhead, sclk_ff);
2089         mc_latency_mclk.full += disp_latency_overhead.full + cur_latency_mclk.full;
2090         mc_latency_sclk.full += disp_latency_overhead.full + cur_latency_sclk.full;
2091
2092         if (mc_latency_mclk.full > mc_latency_sclk.full)
2093                 disp_latency.full = mc_latency_mclk.full;
2094         else
2095                 disp_latency.full = mc_latency_sclk.full;
2096
2097         /* setup Max GRPH_STOP_REQ default value */
2098         if (ASIC_IS_RV100(rdev))
2099                 max_stop_req = 0x5c;
2100         else
2101                 max_stop_req = 0x7c;
2102
2103         if (mode1) {
2104                 /*  CRTC1
2105                     Set GRPH_BUFFER_CNTL register using h/w defined optimal values.
2106                     GRPH_STOP_REQ <= MIN[ 0x7C, (CRTC_H_DISP + 1) * (bit depth) / 0x10 ]
2107                 */
2108                 stop_req = mode1->hdisplay * pixel_bytes1 / 16;
2109
2110                 if (stop_req > max_stop_req)
2111                         stop_req = max_stop_req;
2112
2113                 /*
2114                   Find the drain rate of the display buffer.
2115                 */
2116                 temp_ff.full = rfixed_const((16/pixel_bytes1));
2117                 disp_drain_rate.full = rfixed_div(pix_clk, temp_ff);
2118
2119                 /*
2120                   Find the critical point of the display buffer.
2121                 */
2122                 crit_point_ff.full = rfixed_mul(disp_drain_rate, disp_latency);
2123                 crit_point_ff.full += rfixed_const_half(0);
2124
2125                 critical_point = rfixed_trunc(crit_point_ff);
2126
2127                 if (rdev->disp_priority == 2) {
2128                         critical_point = 0;
2129                 }
2130
2131                 /*
2132                   The critical point should never be above max_stop_req-4.  Setting
2133                   GRPH_CRITICAL_CNTL = 0 will thus force high priority all the time.
2134                 */
2135                 if (max_stop_req - critical_point < 4)
2136                         critical_point = 0;
2137
2138                 if (critical_point == 0 && mode2 && rdev->family == CHIP_R300) {
2139                         /* some R300 cards have problem with this set to 0, when CRTC2 is enabled.*/
2140                         critical_point = 0x10;
2141                 }
2142
2143                 temp = RREG32(RADEON_GRPH_BUFFER_CNTL);
2144                 temp &= ~(RADEON_GRPH_STOP_REQ_MASK);
2145                 temp |= (stop_req << RADEON_GRPH_STOP_REQ_SHIFT);
2146                 temp &= ~(RADEON_GRPH_START_REQ_MASK);
2147                 if ((rdev->family == CHIP_R350) &&
2148                     (stop_req > 0x15)) {
2149                         stop_req -= 0x10;
2150                 }
2151                 temp |= (stop_req << RADEON_GRPH_START_REQ_SHIFT);
2152                 temp |= RADEON_GRPH_BUFFER_SIZE;
2153                 temp &= ~(RADEON_GRPH_CRITICAL_CNTL   |
2154                           RADEON_GRPH_CRITICAL_AT_SOF |
2155                           RADEON_GRPH_STOP_CNTL);
2156                 /*
2157                   Write the result into the register.
2158                 */
2159                 WREG32(RADEON_GRPH_BUFFER_CNTL, ((temp & ~RADEON_GRPH_CRITICAL_POINT_MASK) |
2160                                                        (critical_point << RADEON_GRPH_CRITICAL_POINT_SHIFT)));
2161
2162 #if 0
2163                 if ((rdev->family == CHIP_RS400) ||
2164                     (rdev->family == CHIP_RS480)) {
2165                         /* attempt to program RS400 disp regs correctly ??? */
2166                         temp = RREG32(RS400_DISP1_REG_CNTL);
2167                         temp &= ~(RS400_DISP1_START_REQ_LEVEL_MASK |
2168                                   RS400_DISP1_STOP_REQ_LEVEL_MASK);
2169                         WREG32(RS400_DISP1_REQ_CNTL1, (temp |
2170                                                        (critical_point << RS400_DISP1_START_REQ_LEVEL_SHIFT) |
2171                                                        (critical_point << RS400_DISP1_STOP_REQ_LEVEL_SHIFT)));
2172                         temp = RREG32(RS400_DMIF_MEM_CNTL1);
2173                         temp &= ~(RS400_DISP1_CRITICAL_POINT_START_MASK |
2174                                   RS400_DISP1_CRITICAL_POINT_STOP_MASK);
2175                         WREG32(RS400_DMIF_MEM_CNTL1, (temp |
2176                                                       (critical_point << RS400_DISP1_CRITICAL_POINT_START_SHIFT) |
2177                                                       (critical_point << RS400_DISP1_CRITICAL_POINT_STOP_SHIFT)));
2178                 }
2179 #endif
2180
2181                 DRM_DEBUG("GRPH_BUFFER_CNTL from to %x\n",
2182                           /*      (unsigned int)info->SavedReg->grph_buffer_cntl, */
2183                           (unsigned int)RREG32(RADEON_GRPH_BUFFER_CNTL));
2184         }
2185
2186         if (mode2) {
2187                 u32 grph2_cntl;
2188                 stop_req = mode2->hdisplay * pixel_bytes2 / 16;
2189
2190                 if (stop_req > max_stop_req)
2191                         stop_req = max_stop_req;
2192
2193                 /*
2194                   Find the drain rate of the display buffer.
2195                 */
2196                 temp_ff.full = rfixed_const((16/pixel_bytes2));
2197                 disp_drain_rate2.full = rfixed_div(pix_clk2, temp_ff);
2198
2199                 grph2_cntl = RREG32(RADEON_GRPH2_BUFFER_CNTL);
2200                 grph2_cntl &= ~(RADEON_GRPH_STOP_REQ_MASK);
2201                 grph2_cntl |= (stop_req << RADEON_GRPH_STOP_REQ_SHIFT);
2202                 grph2_cntl &= ~(RADEON_GRPH_START_REQ_MASK);
2203                 if ((rdev->family == CHIP_R350) &&
2204                     (stop_req > 0x15)) {
2205                         stop_req -= 0x10;
2206                 }
2207                 grph2_cntl |= (stop_req << RADEON_GRPH_START_REQ_SHIFT);
2208                 grph2_cntl |= RADEON_GRPH_BUFFER_SIZE;
2209                 grph2_cntl &= ~(RADEON_GRPH_CRITICAL_CNTL   |
2210                           RADEON_GRPH_CRITICAL_AT_SOF |
2211                           RADEON_GRPH_STOP_CNTL);
2212
2213                 if ((rdev->family == CHIP_RS100) ||
2214                     (rdev->family == CHIP_RS200))
2215                         critical_point2 = 0;
2216                 else {
2217                         temp = (rdev->mc.vram_width * rdev->mc.vram_is_ddr + 1)/128;
2218                         temp_ff.full = rfixed_const(temp);
2219                         temp_ff.full = rfixed_mul(mclk_ff, temp_ff);
2220                         if (sclk_ff.full < temp_ff.full)
2221                                 temp_ff.full = sclk_ff.full;
2222
2223                         read_return_rate.full = temp_ff.full;
2224
2225                         if (mode1) {
2226                                 temp_ff.full = read_return_rate.full - disp_drain_rate.full;
2227                                 time_disp1_drop_priority.full = rfixed_div(crit_point_ff, temp_ff);
2228                         } else {
2229                                 time_disp1_drop_priority.full = 0;
2230                         }
2231                         crit_point_ff.full = disp_latency.full + time_disp1_drop_priority.full + disp_latency.full;
2232                         crit_point_ff.full = rfixed_mul(crit_point_ff, disp_drain_rate2);
2233                         crit_point_ff.full += rfixed_const_half(0);
2234
2235                         critical_point2 = rfixed_trunc(crit_point_ff);
2236
2237                         if (rdev->disp_priority == 2) {
2238                                 critical_point2 = 0;
2239                         }
2240
2241                         if (max_stop_req - critical_point2 < 4)
2242                                 critical_point2 = 0;
2243
2244                 }
2245
2246                 if (critical_point2 == 0 && rdev->family == CHIP_R300) {
2247                         /* some R300 cards have problem with this set to 0 */
2248                         critical_point2 = 0x10;
2249                 }
2250
2251                 WREG32(RADEON_GRPH2_BUFFER_CNTL, ((grph2_cntl & ~RADEON_GRPH_CRITICAL_POINT_MASK) |
2252                                                   (critical_point2 << RADEON_GRPH_CRITICAL_POINT_SHIFT)));
2253
2254                 if ((rdev->family == CHIP_RS400) ||
2255                     (rdev->family == CHIP_RS480)) {
2256 #if 0
2257                         /* attempt to program RS400 disp2 regs correctly ??? */
2258                         temp = RREG32(RS400_DISP2_REQ_CNTL1);
2259                         temp &= ~(RS400_DISP2_START_REQ_LEVEL_MASK |
2260                                   RS400_DISP2_STOP_REQ_LEVEL_MASK);
2261                         WREG32(RS400_DISP2_REQ_CNTL1, (temp |
2262                                                        (critical_point2 << RS400_DISP1_START_REQ_LEVEL_SHIFT) |
2263                                                        (critical_point2 << RS400_DISP1_STOP_REQ_LEVEL_SHIFT)));
2264                         temp = RREG32(RS400_DISP2_REQ_CNTL2);
2265                         temp &= ~(RS400_DISP2_CRITICAL_POINT_START_MASK |
2266                                   RS400_DISP2_CRITICAL_POINT_STOP_MASK);
2267                         WREG32(RS400_DISP2_REQ_CNTL2, (temp |
2268                                                        (critical_point2 << RS400_DISP2_CRITICAL_POINT_START_SHIFT) |
2269                                                        (critical_point2 << RS400_DISP2_CRITICAL_POINT_STOP_SHIFT)));
2270 #endif
2271                         WREG32(RS400_DISP2_REQ_CNTL1, 0x105DC1CC);
2272                         WREG32(RS400_DISP2_REQ_CNTL2, 0x2749D000);
2273                         WREG32(RS400_DMIF_MEM_CNTL1,  0x29CA71DC);
2274                         WREG32(RS400_DISP1_REQ_CNTL1, 0x28FBC3AC);
2275                 }
2276
2277                 DRM_DEBUG("GRPH2_BUFFER_CNTL from to %x\n",
2278                           (unsigned int)RREG32(RADEON_GRPH2_BUFFER_CNTL));
2279         }
2280 }