git.oblomov.eu Git - linux-2.6/blob - 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.vram_size);
 177         tmp = rdev->mc.vram_location + rdev->mc.vram_size - 1;
 178         tmp = REG_SET(RADEON_MC_FB_TOP, tmp >> 16);
 179         tmp |= REG_SET(RADEON_MC_FB_START, rdev->mc.vram_location >> 16);
 180         WREG32(RADEON_MC_FB_LOCATION, tmp);
 181
 182         /* Enable bus mastering */
 183         tmp = RREG32(RADEON_BUS_CNTL) & ~RADEON_BUS_MASTER_DIS;
 184         WREG32(RADEON_BUS_CNTL, tmp);
 185
 186         if (rdev->flags & RADEON_IS_AGP) {
 187                 tmp = rdev->mc.gtt_location + rdev->mc.gtt_size - 1;
 188                 tmp = REG_SET(RADEON_MC_AGP_TOP, tmp >> 16);
 189                 tmp |= REG_SET(RADEON_MC_AGP_START, rdev->mc.gtt_location >> 16);
 190                 WREG32(RADEON_MC_AGP_LOCATION, tmp);
 191                 WREG32(RADEON_AGP_BASE, rdev->mc.agp_base);
 192         } else {
 193                 WREG32(RADEON_MC_AGP_LOCATION, 0x0FFFFFFF);
 194                 WREG32(RADEON_AGP_BASE, 0);
 195         }
 196
 197         tmp = RREG32(RADEON_HOST_PATH_CNTL) & RADEON_HDP_APER_CNTL;
 198         tmp |= (7 << 28);
 199         WREG32(RADEON_HOST_PATH_CNTL, tmp | RADEON_HDP_SOFT_RESET | RADEON_HDP_READ_BUFFER_INVALIDATE);
 200         (void)RREG32(RADEON_HOST_PATH_CNTL);
 201         WREG32(RADEON_HOST_PATH_CNTL, tmp);
 202         (void)RREG32(RADEON_HOST_PATH_CNTL);
 203 }
 204
 205 int r100_mc_init(struct radeon_device *rdev)
 206 {
 207         int r;
 208
 209         if (r100_debugfs_rbbm_init(rdev)) {
 210                 DRM_ERROR("Failed to register debugfs file for RBBM !\n");
 211         }
 212
 213         r100_gpu_init(rdev);
 214         /* Disable gart which also disable out of gart access */
 215         r100_pci_gart_disable(rdev);
 216
 217         /* Setup GPU memory space */
 218         rdev->mc.gtt_location = 0xFFFFFFFFUL;
 219         if (rdev->flags & RADEON_IS_AGP) {
 220                 r = radeon_agp_init(rdev);
 221                 if (r) {
 222                         printk(KERN_WARNING "[drm] Disabling AGP\n");
 223                         rdev->flags &= ~RADEON_IS_AGP;
 224                         rdev->mc.gtt_size = radeon_gart_size * 1024 * 1024;
 225                 } else {
 226                         rdev->mc.gtt_location = rdev->mc.agp_base;
 227                 }
 228         }
 229         r = radeon_mc_setup(rdev);
 230         if (r) {
 231                 return r;
 232         }
 233
 234         r100_mc_disable_clients(rdev);
 235         if (r100_mc_wait_for_idle(rdev)) {
 236                 printk(KERN_WARNING "Failed to wait MC idle while "
 237                        "programming pipes. Bad things might happen.\n");
 238         }
 239
 240         r100_mc_setup(rdev);
 241         return 0;
 242 }
 243
 244 void r100_mc_fini(struct radeon_device *rdev)
 245 {
 246         r100_pci_gart_disable(rdev);
 247         radeon_gart_table_ram_free(rdev);
 248         radeon_gart_fini(rdev);
 249 }
 250
 251
 252 /*
 253  * Fence emission
 254  */
 255 void r100_fence_ring_emit(struct radeon_device *rdev,
 256                           struct radeon_fence *fence)
 257 {
 258         /* Who ever call radeon_fence_emit should call ring_lock and ask
 259          * for enough space (today caller are ib schedule and buffer move) */
 260         /* Wait until IDLE & CLEAN */
 261         radeon_ring_write(rdev, PACKET0(0x1720, 0));
 262         radeon_ring_write(rdev, (1 << 16) | (1 << 17));
 263         /* Emit fence sequence & fire IRQ */
 264         radeon_ring_write(rdev, PACKET0(rdev->fence_drv.scratch_reg, 0));
 265         radeon_ring_write(rdev, fence->seq);
 266         radeon_ring_write(rdev, PACKET0(RADEON_GEN_INT_STATUS, 0));
 267         radeon_ring_write(rdev, RADEON_SW_INT_FIRE);
 268 }
 269
 270
 271 /*
 272  * Writeback
 273  */
 274 int r100_wb_init(struct radeon_device *rdev)
 275 {
 276         int r;
 277
 278         if (rdev->wb.wb_obj == NULL) {
 279                 r = radeon_object_create(rdev, NULL, 4096,
 280                                          true,
 281                                          RADEON_GEM_DOMAIN_GTT,
 282                                          false, &rdev->wb.wb_obj);
 283                 if (r) {
 284                         DRM_ERROR("radeon: failed to create WB buffer (%d).\n", r);
 285                         return r;
 286                 }
 287                 r = radeon_object_pin(rdev->wb.wb_obj,
 288                                       RADEON_GEM_DOMAIN_GTT,
 289                                       &rdev->wb.gpu_addr);
 290                 if (r) {
 291                         DRM_ERROR("radeon: failed to pin WB buffer (%d).\n", r);
 292                         return r;
 293                 }
 294                 r = radeon_object_kmap(rdev->wb.wb_obj, (void **)&rdev->wb.wb);
 295                 if (r) {
 296                         DRM_ERROR("radeon: failed to map WB buffer (%d).\n", r);
 297                         return r;
 298                 }
 299         }
 300         WREG32(0x774, rdev->wb.gpu_addr);
 301         WREG32(0x70C, rdev->wb.gpu_addr + 1024);
 302         WREG32(0x770, 0xff);
 303         return 0;
 304 }
 305
 306 void r100_wb_fini(struct radeon_device *rdev)
 307 {
 308         if (rdev->wb.wb_obj) {
 309                 radeon_object_kunmap(rdev->wb.wb_obj);
 310                 radeon_object_unpin(rdev->wb.wb_obj);
 311                 radeon_object_unref(&rdev->wb.wb_obj);
 312                 rdev->wb.wb = NULL;
 313                 rdev->wb.wb_obj = NULL;
 314         }
 315 }
 316
 317 int r100_copy_blit(struct radeon_device *rdev,
 318                    uint64_t src_offset,
 319                    uint64_t dst_offset,
 320                    unsigned num_pages,
 321                    struct radeon_fence *fence)
 322 {
 323         uint32_t cur_pages;
 324         uint32_t stride_bytes = PAGE_SIZE;
 325         uint32_t pitch;
 326         uint32_t stride_pixels;
 327         unsigned ndw;
 328         int num_loops;
 329         int r = 0;
 330
 331         /* radeon limited to 16k stride */
 332         stride_bytes &= 0x3fff;
 333         /* radeon pitch is /64 */
 334         pitch = stride_bytes / 64;
 335         stride_pixels = stride_bytes / 4;
 336         num_loops = DIV_ROUND_UP(num_pages, 8191);
 337
 338         /* Ask for enough room for blit + flush + fence */
 339         ndw = 64 + (10 * num_loops);
 340         r = radeon_ring_lock(rdev, ndw);
 341         if (r) {
 342                 DRM_ERROR("radeon: moving bo (%d) asking for %u dw.\n", r, ndw);
 343                 return -EINVAL;
 344         }
 345         while (num_pages > 0) {
 346                 cur_pages = num_pages;
 347                 if (cur_pages > 8191) {
 348                         cur_pages = 8191;
 349                 }
 350                 num_pages -= cur_pages;
 351
 352                 /* pages are in Y direction - height
 353                    page width in X direction - width */
 354                 radeon_ring_write(rdev, PACKET3(PACKET3_BITBLT_MULTI, 8));
 355                 radeon_ring_write(rdev,
 356                                   RADEON_GMC_SRC_PITCH_OFFSET_CNTL |
 357                                   RADEON_GMC_DST_PITCH_OFFSET_CNTL |
 358                                   RADEON_GMC_SRC_CLIPPING |
 359                                   RADEON_GMC_DST_CLIPPING |
 360                                   RADEON_GMC_BRUSH_NONE |
 361                                   (RADEON_COLOR_FORMAT_ARGB8888 << 8) |
 362                                   RADEON_GMC_SRC_DATATYPE_COLOR |
 363                                   RADEON_ROP3_S |
 364                                   RADEON_DP_SRC_SOURCE_MEMORY |
 365                                   RADEON_GMC_CLR_CMP_CNTL_DIS |
 366                                   RADEON_GMC_WR_MSK_DIS);
 367                 radeon_ring_write(rdev, (pitch << 22) | (src_offset >> 10));
 368                 radeon_ring_write(rdev, (pitch << 22) | (dst_offset >> 10));
 369                 radeon_ring_write(rdev, (0x1fff) | (0x1fff << 16));
 370                 radeon_ring_write(rdev, 0);
 371                 radeon_ring_write(rdev, (0x1fff) | (0x1fff << 16));
 372                 radeon_ring_write(rdev, num_pages);
 373                 radeon_ring_write(rdev, num_pages);
 374                 radeon_ring_write(rdev, cur_pages | (stride_pixels << 16));
 375         }
 376         radeon_ring_write(rdev, PACKET0(RADEON_DSTCACHE_CTLSTAT, 0));
 377         radeon_ring_write(rdev, RADEON_RB2D_DC_FLUSH_ALL);
 378         radeon_ring_write(rdev, PACKET0(RADEON_WAIT_UNTIL, 0));
 379         radeon_ring_write(rdev,
 380                           RADEON_WAIT_2D_IDLECLEAN |
 381                           RADEON_WAIT_HOST_IDLECLEAN |
 382                           RADEON_WAIT_DMA_GUI_IDLE);
 383         if (fence) {
 384                 r = radeon_fence_emit(rdev, fence);
 385         }
 386         radeon_ring_unlock_commit(rdev);
 387         return r;
 388 }
 389
 390
 391 /*
 392  * CP
 393  */
 394 void r100_ring_start(struct radeon_device *rdev)
 395 {
 396         int r;
 397
 398         r = radeon_ring_lock(rdev, 2);
 399         if (r) {
 400                 return;
 401         }
 402         radeon_ring_write(rdev, PACKET0(RADEON_ISYNC_CNTL, 0));
 403         radeon_ring_write(rdev,
 404                           RADEON_ISYNC_ANY2D_IDLE3D |
 405                           RADEON_ISYNC_ANY3D_IDLE2D |
 406                           RADEON_ISYNC_WAIT_IDLEGUI |
 407                           RADEON_ISYNC_CPSCRATCH_IDLEGUI);
 408         radeon_ring_unlock_commit(rdev);
 409 }
 410
 411 static void r100_cp_load_microcode(struct radeon_device *rdev)
 412 {
 413         int i;
 414
 415         if (r100_gui_wait_for_idle(rdev)) {
 416                 printk(KERN_WARNING "Failed to wait GUI idle while "
 417                        "programming pipes. Bad things might happen.\n");
 418         }
 419
 420         WREG32(RADEON_CP_ME_RAM_ADDR, 0);
 421         if ((rdev->family == CHIP_R100) || (rdev->family == CHIP_RV100) ||
 422             (rdev->family == CHIP_RV200) || (rdev->family == CHIP_RS100) ||
 423             (rdev->family == CHIP_RS200)) {
 424                 DRM_INFO("Loading R100 Microcode\n");
 425                 for (i = 0; i < 256; i++) {
 426                         WREG32(RADEON_CP_ME_RAM_DATAH, R100_cp_microcode[i][1]);
 427                         WREG32(RADEON_CP_ME_RAM_DATAL, R100_cp_microcode[i][0]);
 428                 }
 429         } else if ((rdev->family == CHIP_R200) ||
 430                    (rdev->family == CHIP_RV250) ||
 431                    (rdev->family == CHIP_RV280) ||
 432                    (rdev->family == CHIP_RS300)) {
 433                 DRM_INFO("Loading R200 Microcode\n");
 434                 for (i = 0; i < 256; i++) {
 435                         WREG32(RADEON_CP_ME_RAM_DATAH, R200_cp_microcode[i][1]);
 436                         WREG32(RADEON_CP_ME_RAM_DATAL, R200_cp_microcode[i][0]);
 437                 }
 438         } else if ((rdev->family == CHIP_R300) ||
 439                    (rdev->family == CHIP_R350) ||
 440                    (rdev->family == CHIP_RV350) ||
 441                    (rdev->family == CHIP_RV380) ||
 442                    (rdev->family == CHIP_RS400) ||
 443                    (rdev->family == CHIP_RS480)) {
 444                 DRM_INFO("Loading R300 Microcode\n");
 445                 for (i = 0; i < 256; i++) {
 446                         WREG32(RADEON_CP_ME_RAM_DATAH, R300_cp_microcode[i][1]);
 447                         WREG32(RADEON_CP_ME_RAM_DATAL, R300_cp_microcode[i][0]);
 448                 }
 449         } else if ((rdev->family == CHIP_R420) ||
 450                    (rdev->family == CHIP_R423) ||
 451                    (rdev->family == CHIP_RV410)) {
 452                 DRM_INFO("Loading R400 Microcode\n");
 453                 for (i = 0; i < 256; i++) {
 454                         WREG32(RADEON_CP_ME_RAM_DATAH, R420_cp_microcode[i][1]);
 455                         WREG32(RADEON_CP_ME_RAM_DATAL, R420_cp_microcode[i][0]);
 456                 }
 457         } else if ((rdev->family == CHIP_RS690) ||
 458                    (rdev->family == CHIP_RS740)) {
 459                 DRM_INFO("Loading RS690/RS740 Microcode\n");
 460                 for (i = 0; i < 256; i++) {
 461                         WREG32(RADEON_CP_ME_RAM_DATAH, RS690_cp_microcode[i][1]);
 462                         WREG32(RADEON_CP_ME_RAM_DATAL, RS690_cp_microcode[i][0]);
 463                 }
 464         } else if (rdev->family == CHIP_RS600) {
 465                 DRM_INFO("Loading RS600 Microcode\n");
 466                 for (i = 0; i < 256; i++) {
 467                         WREG32(RADEON_CP_ME_RAM_DATAH, RS600_cp_microcode[i][1]);
 468                         WREG32(RADEON_CP_ME_RAM_DATAL, RS600_cp_microcode[i][0]);
 469                 }
 470         } else if ((rdev->family == CHIP_RV515) ||
 471                    (rdev->family == CHIP_R520) ||
 472                    (rdev->family == CHIP_RV530) ||
 473                    (rdev->family == CHIP_R580) ||
 474                    (rdev->family == CHIP_RV560) ||
 475                    (rdev->family == CHIP_RV570)) {
 476                 DRM_INFO("Loading R500 Microcode\n");
 477                 for (i = 0; i < 256; i++) {
 478                         WREG32(RADEON_CP_ME_RAM_DATAH, R520_cp_microcode[i][1]);
 479                         WREG32(RADEON_CP_ME_RAM_DATAL, R520_cp_microcode[i][0]);
 480                 }
 481         }
 482 }
 483
 484 int r100_cp_init(struct radeon_device *rdev, unsigned ring_size)
 485 {
 486         unsigned rb_bufsz;
 487         unsigned rb_blksz;
 488         unsigned max_fetch;
 489         unsigned pre_write_timer;
 490         unsigned pre_write_limit;
 491         unsigned indirect2_start;
 492         unsigned indirect1_start;
 493         uint32_t tmp;
 494         int r;
 495
 496         if (r100_debugfs_cp_init(rdev)) {
 497                 DRM_ERROR("Failed to register debugfs file for CP !\n");
 498         }
 499         /* Reset CP */
 500         tmp = RREG32(RADEON_CP_CSQ_STAT);
 501         if ((tmp & (1 << 31))) {
 502                 DRM_INFO("radeon: cp busy (0x%08X) resetting\n", tmp);
 503                 WREG32(RADEON_CP_CSQ_MODE, 0);
 504                 WREG32(RADEON_CP_CSQ_CNTL, 0);
 505                 WREG32(RADEON_RBBM_SOFT_RESET, RADEON_SOFT_RESET_CP);
 506                 tmp = RREG32(RADEON_RBBM_SOFT_RESET);
 507                 mdelay(2);
 508                 WREG32(RADEON_RBBM_SOFT_RESET, 0);
 509                 tmp = RREG32(RADEON_RBBM_SOFT_RESET);
 510                 mdelay(2);
 511                 tmp = RREG32(RADEON_CP_CSQ_STAT);
 512                 if ((tmp & (1 << 31))) {
 513                         DRM_INFO("radeon: cp reset failed (0x%08X)\n", tmp);
 514                 }
 515         } else {
 516                 DRM_INFO("radeon: cp idle (0x%08X)\n", tmp);
 517         }
 518         /* Align ring size */
 519         rb_bufsz = drm_order(ring_size / 8);
 520         ring_size = (1 << (rb_bufsz + 1)) * 4;
 521         r100_cp_load_microcode(rdev);
 522         r = radeon_ring_init(rdev, ring_size);
 523         if (r) {
 524                 return r;
 525         }
 526         /* Each time the cp read 1024 bytes (16 dword/quadword) update
 527          * the rptr copy in system ram */
 528         rb_blksz = 9;
 529         /* cp will read 128bytes at a time (4 dwords) */
 530         max_fetch = 1;
 531         rdev->cp.align_mask = 16 - 1;
 532         /* Write to CP_RB_WPTR will be delayed for pre_write_timer clocks */
 533         pre_write_timer = 64;
 534         /* Force CP_RB_WPTR write if written more than one time before the
 535          * delay expire
 536          */
 537         pre_write_limit = 0;
 538         /* Setup the cp cache like this (cache size is 96 dwords) :
 539          *      RING            0  to 15
 540          *      INDIRECT1       16 to 79
 541          *      INDIRECT2       80 to 95
 542          * So ring cache size is 16dwords (> (2 * max_fetch = 2 * 4dwords))
 543          *    indirect1 cache size is 64dwords (> (2 * max_fetch = 2 * 4dwords))
 544          *    indirect2 cache size is 16dwords (> (2 * max_fetch = 2 * 4dwords))
 545          * Idea being that most of the gpu cmd will be through indirect1 buffer
 546          * so it gets the bigger cache.
 547          */
 548         indirect2_start = 80;
 549         indirect1_start = 16;
 550         /* cp setup */
 551         WREG32(0x718, pre_write_timer | (pre_write_limit << 28));
 552         WREG32(RADEON_CP_RB_CNTL,
 553 #ifdef __BIG_ENDIAN
 554                RADEON_BUF_SWAP_32BIT |
 555 #endif
 556                REG_SET(RADEON_RB_BUFSZ, rb_bufsz) |
 557                REG_SET(RADEON_RB_BLKSZ, rb_blksz) |
 558                REG_SET(RADEON_MAX_FETCH, max_fetch) |
 559                RADEON_RB_NO_UPDATE);
 560         /* Set ring address */
 561         DRM_INFO("radeon: ring at 0x%016lX\n", (unsigned long)rdev->cp.gpu_addr);
 562         WREG32(RADEON_CP_RB_BASE, rdev->cp.gpu_addr);
 563         /* Force read & write ptr to 0 */
 564         tmp = RREG32(RADEON_CP_RB_CNTL);
 565         WREG32(RADEON_CP_RB_CNTL, tmp | RADEON_RB_RPTR_WR_ENA);
 566         WREG32(RADEON_CP_RB_RPTR_WR, 0);
 567         WREG32(RADEON_CP_RB_WPTR, 0);
 568         WREG32(RADEON_CP_RB_CNTL, tmp);
 569         udelay(10);
 570         rdev->cp.rptr = RREG32(RADEON_CP_RB_RPTR);
 571         rdev->cp.wptr = RREG32(RADEON_CP_RB_WPTR);
 572         /* Set cp mode to bus mastering & enable cp*/
 573         WREG32(RADEON_CP_CSQ_MODE,
 574                REG_SET(RADEON_INDIRECT2_START, indirect2_start) |
 575                REG_SET(RADEON_INDIRECT1_START, indirect1_start));
 576         WREG32(0x718, 0);
 577         WREG32(0x744, 0x00004D4D);
 578         WREG32(RADEON_CP_CSQ_CNTL, RADEON_CSQ_PRIBM_INDBM);
 579         radeon_ring_start(rdev);
 580         r = radeon_ring_test(rdev);
 581         if (r) {
 582                 DRM_ERROR("radeon: cp isn't working (%d).\n", r);
 583                 return r;
 584         }
 585         rdev->cp.ready = true;
 586         return 0;
 587 }
 588
 589 void r100_cp_fini(struct radeon_device *rdev)
 590 {
 591         /* Disable ring */
 592         rdev->cp.ready = false;
 593         WREG32(RADEON_CP_CSQ_CNTL, 0);
 594         radeon_ring_fini(rdev);
 595         DRM_INFO("radeon: cp finalized\n");
 596 }
 597
 598 void r100_cp_disable(struct radeon_device *rdev)
 599 {
 600         /* Disable ring */
 601         rdev->cp.ready = false;
 602         WREG32(RADEON_CP_CSQ_MODE, 0);
 603         WREG32(RADEON_CP_CSQ_CNTL, 0);
 604         if (r100_gui_wait_for_idle(rdev)) {
 605                 printk(KERN_WARNING "Failed to wait GUI idle while "
 606                        "programming pipes. Bad things might happen.\n");
 607         }
 608 }
 609
 610 int r100_cp_reset(struct radeon_device *rdev)
 611 {
 612         uint32_t tmp;
 613         bool reinit_cp;
 614         int i;
 615
 616         reinit_cp = rdev->cp.ready;
 617         rdev->cp.ready = false;
 618         WREG32(RADEON_CP_CSQ_MODE, 0);
 619         WREG32(RADEON_CP_CSQ_CNTL, 0);
 620         WREG32(RADEON_RBBM_SOFT_RESET, RADEON_SOFT_RESET_CP);
 621         (void)RREG32(RADEON_RBBM_SOFT_RESET);
 622         udelay(200);
 623         WREG32(RADEON_RBBM_SOFT_RESET, 0);
 624         /* Wait to prevent race in RBBM_STATUS */
 625         mdelay(1);
 626         for (i = 0; i < rdev->usec_timeout; i++) {
 627                 tmp = RREG32(RADEON_RBBM_STATUS);
 628                 if (!(tmp & (1 << 16))) {
 629                         DRM_INFO("CP reset succeed (RBBM_STATUS=0x%08X)\n",
 630                                  tmp);
 631                         if (reinit_cp) {
 632                                 return r100_cp_init(rdev, rdev->cp.ring_size);
 633                         }
 634                         return 0;
 635                 }
 636                 DRM_UDELAY(1);
 637         }
 638         tmp = RREG32(RADEON_RBBM_STATUS);
 639         DRM_ERROR("Failed to reset CP (RBBM_STATUS=0x%08X)!\n", tmp);
 640         return -1;
 641 }
 642
 643
 644 /*
 645  * CS functions
 646  */
 647 int r100_cs_parse_packet0(struct radeon_cs_parser *p,
 648                           struct radeon_cs_packet *pkt,
 649                           const unsigned *auth, unsigned n,
 650                           radeon_packet0_check_t check)
 651 {
 652         unsigned reg;
 653         unsigned i, j, m;
 654         unsigned idx;
 655         int r;
 656
 657         idx = pkt->idx + 1;
 658         reg = pkt->reg;
 659         /* Check that register fall into register range
 660          * determined by the number of entry (n) in the
 661          * safe register bitmap.
 662          */
 663         if (pkt->one_reg_wr) {
 664                 if ((reg >> 7) > n) {
 665                         return -EINVAL;
 666                 }
 667         } else {
 668                 if (((reg + (pkt->count << 2)) >> 7) > n) {
 669                         return -EINVAL;
 670                 }
 671         }
 672         for (i = 0; i <= pkt->count; i++, idx++) {
 673                 j = (reg >> 7);
 674                 m = 1 << ((reg >> 2) & 31);
 675                 if (auth[j] & m) {
 676                         r = check(p, pkt, idx, reg);
 677                         if (r) {
 678                                 return r;
 679                         }
 680                 }
 681                 if (pkt->one_reg_wr) {
 682                         if (!(auth[j] & m)) {
 683                                 break;
 684                         }
 685                 } else {
 686                         reg += 4;
 687                 }
 688         }
 689         return 0;
 690 }
 691
 692 void r100_cs_dump_packet(struct radeon_cs_parser *p,
 693                          struct radeon_cs_packet *pkt)
 694 {
 695         struct radeon_cs_chunk *ib_chunk;
 696         volatile uint32_t *ib;
 697         unsigned i;
 698         unsigned idx;
 699
 700         ib = p->ib->ptr;
 701         ib_chunk = &p->chunks[p->chunk_ib_idx];
 702         idx = pkt->idx;
 703         for (i = 0; i <= (pkt->count + 1); i++, idx++) {
 704                 DRM_INFO("ib[%d]=0x%08X\n", idx, ib[idx]);
 705         }
 706 }
 707
 708 /**
 709  * r100_cs_packet_parse() - parse cp packet and point ib index to next packet
 710  * @parser:     parser structure holding parsing context.
 711  * @pkt:        where to store packet informations
 712  *
 713  * Assume that chunk_ib_index is properly set. Will return -EINVAL
 714  * if packet is bigger than remaining ib size. or if packets is unknown.
 715  **/
 716 int r100_cs_packet_parse(struct radeon_cs_parser *p,
 717                          struct radeon_cs_packet *pkt,
 718                          unsigned idx)
 719 {
 720         struct radeon_cs_chunk *ib_chunk = &p->chunks[p->chunk_ib_idx];
 721         uint32_t header = ib_chunk->kdata[idx];
 722
 723         if (idx >= ib_chunk->length_dw) {
 724                 DRM_ERROR("Can not parse packet at %d after CS end %d !\n",
 725                           idx, ib_chunk->length_dw);
 726                 return -EINVAL;
 727         }
 728         pkt->idx = idx;
 729         pkt->type = CP_PACKET_GET_TYPE(header);
 730         pkt->count = CP_PACKET_GET_COUNT(header);
 731         switch (pkt->type) {
 732         case PACKET_TYPE0:
 733                 pkt->reg = CP_PACKET0_GET_REG(header);
 734                 pkt->one_reg_wr = CP_PACKET0_GET_ONE_REG_WR(header);
 735                 break;
 736         case PACKET_TYPE3:
 737                 pkt->opcode = CP_PACKET3_GET_OPCODE(header);
 738                 break;
 739         case PACKET_TYPE2:
 740                 pkt->count = -1;
 741                 break;
 742         default:
 743                 DRM_ERROR("Unknown packet type %d at %d !\n", pkt->type, idx);
 744                 return -EINVAL;
 745         }
 746         if ((pkt->count + 1 + pkt->idx) >= ib_chunk->length_dw) {
 747                 DRM_ERROR("Packet (%d:%d:%d) end after CS buffer (%d) !\n",
 748                           pkt->idx, pkt->type, pkt->count, ib_chunk->length_dw);
 749                 return -EINVAL;
 750         }
 751         return 0;
 752 }
 753
 754 /**
 755  * r100_cs_packet_next_vline() - parse userspace VLINE packet
 756  * @parser:             parser structure holding parsing context.
 757  *
 758  * Userspace sends a special sequence for VLINE waits.
 759  * PACKET0 - VLINE_START_END + value
 760  * PACKET0 - WAIT_UNTIL +_value
 761  * RELOC (P3) - crtc_id in reloc.
 762  *
 763  * This function parses this and relocates the VLINE START END
 764  * and WAIT UNTIL packets to the correct crtc.
 765  * It also detects a switched off crtc and nulls out the
 766  * wait in that case.
 767  */
 768 int r100_cs_packet_parse_vline(struct radeon_cs_parser *p)
 769 {
 770         struct radeon_cs_chunk *ib_chunk;
 771         struct drm_mode_object *obj;
 772         struct drm_crtc *crtc;
 773         struct radeon_crtc *radeon_crtc;
 774         struct radeon_cs_packet p3reloc, waitreloc;
 775         int crtc_id;
 776         int r;
 777         uint32_t header, h_idx, reg;
 778
 779         ib_chunk = &p->chunks[p->chunk_ib_idx];
 780
 781         /* parse the wait until */
 782         r = r100_cs_packet_parse(p, &waitreloc, p->idx);
 783         if (r)
 784                 return r;
 785
 786         /* check its a wait until and only 1 count */
 787         if (waitreloc.reg != RADEON_WAIT_UNTIL ||
 788             waitreloc.count != 0) {
 789                 DRM_ERROR("vline wait had illegal wait until segment\n");
 790                 r = -EINVAL;
 791                 return r;
 792         }
 793
 794         if (ib_chunk->kdata[waitreloc.idx + 1] != RADEON_WAIT_CRTC_VLINE) {
 795                 DRM_ERROR("vline wait had illegal wait until\n");
 796                 r = -EINVAL;
 797                 return r;
 798         }
 799
 800         /* jump over the NOP */
 801         r = r100_cs_packet_parse(p, &p3reloc, p->idx);
 802         if (r)
 803                 return r;
 804
 805         h_idx = p->idx - 2;
 806         p->idx += waitreloc.count;
 807         p->idx += p3reloc.count;
 808
 809         header = ib_chunk->kdata[h_idx];
 810         crtc_id = ib_chunk->kdata[h_idx + 5];
 811         reg = ib_chunk->kdata[h_idx] >> 2;
 812         mutex_lock(&p->rdev->ddev->mode_config.mutex);
 813         obj = drm_mode_object_find(p->rdev->ddev, crtc_id, DRM_MODE_OBJECT_CRTC);
 814         if (!obj) {
 815                 DRM_ERROR("cannot find crtc %d\n", crtc_id);
 816                 r = -EINVAL;
 817                 goto out;
 818         }
 819         crtc = obj_to_crtc(obj);
 820         radeon_crtc = to_radeon_crtc(crtc);
 821         crtc_id = radeon_crtc->crtc_id;
 822
 823         if (!crtc->enabled) {
 824                 /* if the CRTC isn't enabled - we need to nop out the wait until */
 825                 ib_chunk->kdata[h_idx + 2] = PACKET2(0);
 826                 ib_chunk->kdata[h_idx + 3] = PACKET2(0);
 827         } else if (crtc_id == 1) {
 828                 switch (reg) {
 829                 case AVIVO_D1MODE_VLINE_START_END:
 830                         header &= R300_CP_PACKET0_REG_MASK;
 831                         header |= AVIVO_D2MODE_VLINE_START_END >> 2;
 832                         break;
 833                 case RADEON_CRTC_GUI_TRIG_VLINE:
 834                         header &= R300_CP_PACKET0_REG_MASK;
 835                         header |= RADEON_CRTC2_GUI_TRIG_VLINE >> 2;
 836                         break;
 837                 default:
 838                         DRM_ERROR("unknown crtc reloc\n");
 839                         r = -EINVAL;
 840                         goto out;
 841                 }
 842                 ib_chunk->kdata[h_idx] = header;
 843                 ib_chunk->kdata[h_idx + 3] |= RADEON_ENG_DISPLAY_SELECT_CRTC1;
 844         }
 845 out:
 846         mutex_unlock(&p->rdev->ddev->mode_config.mutex);
 847         return r;
 848 }
 849
 850 /**
 851  * r100_cs_packet_next_reloc() - parse next packet which should be reloc packet3
 852  * @parser:             parser structure holding parsing context.
 853  * @data:               pointer to relocation data
 854  * @offset_start:       starting offset
 855  * @offset_mask:        offset mask (to align start offset on)
 856  * @reloc:              reloc informations
 857  *
 858  * Check next packet is relocation packet3, do bo validation and compute
 859  * GPU offset using the provided start.
 860  **/
 861 int r100_cs_packet_next_reloc(struct radeon_cs_parser *p,
 862                               struct radeon_cs_reloc **cs_reloc)
 863 {
 864         struct radeon_cs_chunk *ib_chunk;
 865         struct radeon_cs_chunk *relocs_chunk;
 866         struct radeon_cs_packet p3reloc;
 867         unsigned idx;
 868         int r;
 869
 870         if (p->chunk_relocs_idx == -1) {
 871                 DRM_ERROR("No relocation chunk !\n");
 872                 return -EINVAL;
 873         }
 874         *cs_reloc = NULL;
 875         ib_chunk = &p->chunks[p->chunk_ib_idx];
 876         relocs_chunk = &p->chunks[p->chunk_relocs_idx];
 877         r = r100_cs_packet_parse(p, &p3reloc, p->idx);
 878         if (r) {
 879                 return r;
 880         }
 881         p->idx += p3reloc.count + 2;
 882         if (p3reloc.type != PACKET_TYPE3 || p3reloc.opcode != PACKET3_NOP) {
 883                 DRM_ERROR("No packet3 for relocation for packet at %d.\n",
 884                           p3reloc.idx);
 885                 r100_cs_dump_packet(p, &p3reloc);
 886                 return -EINVAL;
 887         }
 888         idx = ib_chunk->kdata[p3reloc.idx + 1];
 889         if (idx >= relocs_chunk->length_dw) {
 890                 DRM_ERROR("Relocs at %d after relocations chunk end %d !\n",
 891                           idx, relocs_chunk->length_dw);
 892                 r100_cs_dump_packet(p, &p3reloc);
 893                 return -EINVAL;
 894         }
 895         /* FIXME: we assume reloc size is 4 dwords */
 896         *cs_reloc = p->relocs_ptr[(idx / 4)];
 897         return 0;
 898 }
 899
 900 static int r100_packet0_check(struct radeon_cs_parser *p,
 901                               struct radeon_cs_packet *pkt)
 902 {
 903         struct radeon_cs_chunk *ib_chunk;
 904         struct radeon_cs_reloc *reloc;
 905         volatile uint32_t *ib;
 906         uint32_t tmp;
 907         unsigned reg;
 908         unsigned i;
 909         unsigned idx;
 910         bool onereg;
 911         int r;
 912         u32 tile_flags = 0;
 913
 914         ib = p->ib->ptr;
 915         ib_chunk = &p->chunks[p->chunk_ib_idx];
 916         idx = pkt->idx + 1;
 917         reg = pkt->reg;
 918         onereg = false;
 919         if (CP_PACKET0_GET_ONE_REG_WR(ib_chunk->kdata[pkt->idx])) {
 920                 onereg = true;
 921         }
 922         for (i = 0; i <= pkt->count; i++, idx++, reg += 4) {
 923                 switch (reg) {
 924                 case RADEON_CRTC_GUI_TRIG_VLINE:
 925                         r = r100_cs_packet_parse_vline(p);
 926                         if (r) {
 927                                 DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
 928                                                 idx, reg);
 929                                 r100_cs_dump_packet(p, pkt);
 930                                 return r;
 931                         }
 932                         break;
 933                 /* FIXME: only allow PACKET3 blit? easier to check for out of
 934                  * range access */
 935                 case RADEON_DST_PITCH_OFFSET:
 936                 case RADEON_SRC_PITCH_OFFSET:
 937                         r = r100_cs_packet_next_reloc(p, &reloc);
 938                         if (r) {
 939                                 DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
 940                                           idx, reg);
 941                                 r100_cs_dump_packet(p, pkt);
 942                                 return r;
 943                         }
 944                         tmp = ib_chunk->kdata[idx] & 0x003fffff;
 945                         tmp += (((u32)reloc->lobj.gpu_offset) >> 10);
 946
 947                         if (reloc->lobj.tiling_flags & RADEON_TILING_MACRO)
 948                                 tile_flags |= RADEON_DST_TILE_MACRO;
 949                         if (reloc->lobj.tiling_flags & RADEON_TILING_MICRO) {
 950                                 if (reg == RADEON_SRC_PITCH_OFFSET) {
 951                                         DRM_ERROR("Cannot src blit from microtiled surface\n");
 952                                         r100_cs_dump_packet(p, pkt);
 953                                         return -EINVAL;
 954                                 }
 955                                 tile_flags |= RADEON_DST_TILE_MICRO;
 956                         }
 957
 958                         tmp |= tile_flags;
 959                         ib[idx] = (ib_chunk->kdata[idx] & 0x3fc00000) | tmp;
 960                         break;
 961                 case RADEON_RB3D_DEPTHOFFSET:
 962                 case RADEON_RB3D_COLOROFFSET:
 963                 case R300_RB3D_COLOROFFSET0:
 964                 case R300_ZB_DEPTHOFFSET:
 965                 case R200_PP_TXOFFSET_0:
 966                 case R200_PP_TXOFFSET_1:
 967                 case R200_PP_TXOFFSET_2:
 968                 case R200_PP_TXOFFSET_3:
 969                 case R200_PP_TXOFFSET_4:
 970                 case R200_PP_TXOFFSET_5:
 971                 case RADEON_PP_TXOFFSET_0:
 972                 case RADEON_PP_TXOFFSET_1:
 973                 case RADEON_PP_TXOFFSET_2:
 974                 case R300_TX_OFFSET_0:
 975                 case R300_TX_OFFSET_0+4:
 976                 case R300_TX_OFFSET_0+8:
 977                 case R300_TX_OFFSET_0+12:
 978                 case R300_TX_OFFSET_0+16:
 979                 case R300_TX_OFFSET_0+20:
 980                 case R300_TX_OFFSET_0+24:
 981                 case R300_TX_OFFSET_0+28:
 982                 case R300_TX_OFFSET_0+32:
 983                 case R300_TX_OFFSET_0+36:
 984                 case R300_TX_OFFSET_0+40:
 985                 case R300_TX_OFFSET_0+44:
 986                 case R300_TX_OFFSET_0+48:
 987                 case R300_TX_OFFSET_0+52:
 988                 case R300_TX_OFFSET_0+56:
 989                 case R300_TX_OFFSET_0+60:
 990                         /* rn50 has no 3D engine so fail on any 3d setup */
 991                         if (ASIC_IS_RN50(p->rdev)) {
 992                                 DRM_ERROR("attempt to use RN50 3D engine failed\n");
 993                                 return -EINVAL;
 994                         }
 995                         r = r100_cs_packet_next_reloc(p, &reloc);
 996                         if (r) {
 997                                 DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
 998                                           idx, reg);
 999                                 r100_cs_dump_packet(p, pkt);
1000                                 return r;
1001                         }
1002                         ib[idx] = ib_chunk->kdata[idx] + ((u32)reloc->lobj.gpu_offset);
1003                         break;
1004                 case R300_RB3D_COLORPITCH0:
1005                 case RADEON_RB3D_COLORPITCH:
1006                         r = r100_cs_packet_next_reloc(p, &reloc);
1007                         if (r) {
1008                                 DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1009                                           idx, reg);
1010                                 r100_cs_dump_packet(p, pkt);
1011                                 return r;
1012                         }
1013
1014                         if (reloc->lobj.tiling_flags & RADEON_TILING_MACRO)
1015                                 tile_flags |= RADEON_COLOR_TILE_ENABLE;
1016                         if (reloc->lobj.tiling_flags & RADEON_TILING_MICRO)
1017                                 tile_flags |= RADEON_COLOR_MICROTILE_ENABLE;
1018
1019                         tmp = ib_chunk->kdata[idx] & ~(0x7 << 16);
1020                         tmp |= tile_flags;
1021                         ib[idx] = tmp;
1022                         break;
1023                 default:
1024                         /* FIXME: we don't want to allow anyothers packet */
1025                         break;
1026                 }
1027                 if (onereg) {
1028                         /* FIXME: forbid onereg write to register on relocate */
1029                         break;
1030                 }
1031         }
1032         return 0;
1033 }
1034
1035 int r100_cs_track_check_pkt3_indx_buffer(struct radeon_cs_parser *p,
1036                                          struct radeon_cs_packet *pkt,
1037                                          struct radeon_object *robj)
1038 {
1039         struct radeon_cs_chunk *ib_chunk;
1040         unsigned idx;
1041
1042         ib_chunk = &p->chunks[p->chunk_ib_idx];
1043         idx = pkt->idx + 1;
1044         if ((ib_chunk->kdata[idx+2] + 1) > radeon_object_size(robj)) {
1045                 DRM_ERROR("[drm] Buffer too small for PACKET3 INDX_BUFFER "
1046                           "(need %u have %lu) !\n",
1047                           ib_chunk->kdata[idx+2] + 1,
1048                           radeon_object_size(robj));
1049                 return -EINVAL;
1050         }
1051         return 0;
1052 }
1053
1054 static int r100_packet3_check(struct radeon_cs_parser *p,
1055                               struct radeon_cs_packet *pkt)
1056 {
1057         struct radeon_cs_chunk *ib_chunk;
1058         struct radeon_cs_reloc *reloc;
1059         unsigned idx;
1060         unsigned i, c;
1061         volatile uint32_t *ib;
1062         int r;
1063
1064         ib = p->ib->ptr;
1065         ib_chunk = &p->chunks[p->chunk_ib_idx];
1066         idx = pkt->idx + 1;
1067         switch (pkt->opcode) {
1068         case PACKET3_3D_LOAD_VBPNTR:
1069                 c = ib_chunk->kdata[idx++];
1070                 for (i = 0; i < (c - 1); i += 2, idx += 3) {
1071                         r = r100_cs_packet_next_reloc(p, &reloc);
1072                         if (r) {
1073                                 DRM_ERROR("No reloc for packet3 %d\n",
1074                                           pkt->opcode);
1075                                 r100_cs_dump_packet(p, pkt);
1076                                 return r;
1077                         }
1078                         ib[idx+1] = ib_chunk->kdata[idx+1] + ((u32)reloc->lobj.gpu_offset);
1079                         r = r100_cs_packet_next_reloc(p, &reloc);
1080                         if (r) {
1081                                 DRM_ERROR("No reloc for packet3 %d\n",
1082                                           pkt->opcode);
1083                                 r100_cs_dump_packet(p, pkt);
1084                                 return r;
1085                         }
1086                         ib[idx+2] = ib_chunk->kdata[idx+2] + ((u32)reloc->lobj.gpu_offset);
1087                 }
1088                 if (c & 1) {
1089                         r = r100_cs_packet_next_reloc(p, &reloc);
1090                         if (r) {
1091                                 DRM_ERROR("No reloc for packet3 %d\n",
1092                                           pkt->opcode);
1093                                 r100_cs_dump_packet(p, pkt);
1094                                 return r;
1095                         }
1096                         ib[idx+1] = ib_chunk->kdata[idx+1] + ((u32)reloc->lobj.gpu_offset);
1097                 }
1098                 break;
1099         case PACKET3_INDX_BUFFER:
1100                 r = r100_cs_packet_next_reloc(p, &reloc);
1101                 if (r) {
1102                         DRM_ERROR("No reloc for packet3 %d\n", pkt->opcode);
1103                         r100_cs_dump_packet(p, pkt);
1104                         return r;
1105                 }
1106                 ib[idx+1] = ib_chunk->kdata[idx+1] + ((u32)reloc->lobj.gpu_offset);
1107                 r = r100_cs_track_check_pkt3_indx_buffer(p, pkt, reloc->robj);
1108                 if (r) {
1109                         return r;
1110                 }
1111                 break;
1112         case 0x23:
1113                 /* FIXME: cleanup */
1114                 /* 3D_RNDR_GEN_INDX_PRIM on r100/r200 */
1115                 r = r100_cs_packet_next_reloc(p, &reloc);
1116                 if (r) {
1117                         DRM_ERROR("No reloc for packet3 %d\n", pkt->opcode);
1118                         r100_cs_dump_packet(p, pkt);
1119                         return r;
1120                 }
1121                 ib[idx] = ib_chunk->kdata[idx] + ((u32)reloc->lobj.gpu_offset);
1122                 break;
1123         case PACKET3_3D_DRAW_IMMD:
1124                 /* triggers drawing using in-packet vertex data */
1125         case PACKET3_3D_DRAW_IMMD_2:
1126                 /* triggers drawing using in-packet vertex data */
1127         case PACKET3_3D_DRAW_VBUF_2:
1128                 /* triggers drawing of vertex buffers setup elsewhere */
1129         case PACKET3_3D_DRAW_INDX_2:
1130                 /* triggers drawing using indices to vertex buffer */
1131         case PACKET3_3D_DRAW_VBUF:
1132                 /* triggers drawing of vertex buffers setup elsewhere */
1133         case PACKET3_3D_DRAW_INDX:
1134                 /* triggers drawing using indices to vertex buffer */
1135         case PACKET3_NOP:
1136                 break;
1137         default:
1138                 DRM_ERROR("Packet3 opcode %x not supported\n", pkt->opcode);
1139                 return -EINVAL;
1140         }
1141         return 0;
1142 }
1143
1144 int r100_cs_parse(struct radeon_cs_parser *p)
1145 {
1146         struct radeon_cs_packet pkt;
1147         int r;
1148
1149         do {
1150                 r = r100_cs_packet_parse(p, &pkt, p->idx);
1151                 if (r) {
1152                         return r;
1153                 }
1154                 p->idx += pkt.count + 2;
1155                 switch (pkt.type) {
1156                         case PACKET_TYPE0:
1157                                 r = r100_packet0_check(p, &pkt);
1158                                 break;
1159                         case PACKET_TYPE2:
1160                                 break;
1161                         case PACKET_TYPE3:
1162                                 r = r100_packet3_check(p, &pkt);
1163                                 break;
1164                         default:
1165                                 DRM_ERROR("Unknown packet type %d !\n",
1166                                           pkt.type);
1167                                 return -EINVAL;
1168                 }
1169                 if (r) {
1170                         return r;
1171                 }
1172         } while (p->idx < p->chunks[p->chunk_ib_idx].length_dw);
1173         return 0;
1174 }
1175
1176
1177 /*
1178  * Global GPU functions
1179  */
1180 void r100_errata(struct radeon_device *rdev)
1181 {
1182         rdev->pll_errata = 0;
1183
1184         if (rdev->family == CHIP_RV200 || rdev->family == CHIP_RS200) {
1185                 rdev->pll_errata |= CHIP_ERRATA_PLL_DUMMYREADS;
1186         }
1187
1188         if (rdev->family == CHIP_RV100 ||
1189             rdev->family == CHIP_RS100 ||
1190             rdev->family == CHIP_RS200) {
1191                 rdev->pll_errata |= CHIP_ERRATA_PLL_DELAY;
1192         }
1193 }
1194
1195 /* Wait for vertical sync on primary CRTC */
1196 void r100_gpu_wait_for_vsync(struct radeon_device *rdev)
1197 {
1198         uint32_t crtc_gen_cntl, tmp;
1199         int i;
1200
1201         crtc_gen_cntl = RREG32(RADEON_CRTC_GEN_CNTL);
1202         if ((crtc_gen_cntl & RADEON_CRTC_DISP_REQ_EN_B) ||
1203             !(crtc_gen_cntl & RADEON_CRTC_EN)) {
1204                 return;
1205         }
1206         /* Clear the CRTC_VBLANK_SAVE bit */
1207         WREG32(RADEON_CRTC_STATUS, RADEON_CRTC_VBLANK_SAVE_CLEAR);
1208         for (i = 0; i < rdev->usec_timeout; i++) {
1209                 tmp = RREG32(RADEON_CRTC_STATUS);
1210                 if (tmp & RADEON_CRTC_VBLANK_SAVE) {
1211                         return;
1212                 }
1213                 DRM_UDELAY(1);
1214         }
1215 }
1216
1217 /* Wait for vertical sync on secondary CRTC */
1218 void r100_gpu_wait_for_vsync2(struct radeon_device *rdev)
1219 {
1220         uint32_t crtc2_gen_cntl, tmp;
1221         int i;
1222
1223         crtc2_gen_cntl = RREG32(RADEON_CRTC2_GEN_CNTL);
1224         if ((crtc2_gen_cntl & RADEON_CRTC2_DISP_REQ_EN_B) ||
1225             !(crtc2_gen_cntl & RADEON_CRTC2_EN))
1226                 return;
1227
1228         /* Clear the CRTC_VBLANK_SAVE bit */
1229         WREG32(RADEON_CRTC2_STATUS, RADEON_CRTC2_VBLANK_SAVE_CLEAR);
1230         for (i = 0; i < rdev->usec_timeout; i++) {
1231                 tmp = RREG32(RADEON_CRTC2_STATUS);
1232                 if (tmp & RADEON_CRTC2_VBLANK_SAVE) {
1233                         return;
1234                 }
1235                 DRM_UDELAY(1);
1236         }
1237 }
1238
1239 int r100_rbbm_fifo_wait_for_entry(struct radeon_device *rdev, unsigned n)
1240 {
1241         unsigned i;
1242         uint32_t tmp;
1243
1244         for (i = 0; i < rdev->usec_timeout; i++) {
1245                 tmp = RREG32(RADEON_RBBM_STATUS) & RADEON_RBBM_FIFOCNT_MASK;
1246                 if (tmp >= n) {
1247                         return 0;
1248                 }
1249                 DRM_UDELAY(1);
1250         }
1251         return -1;
1252 }
1253
1254 int r100_gui_wait_for_idle(struct radeon_device *rdev)
1255 {
1256         unsigned i;
1257         uint32_t tmp;
1258
1259         if (r100_rbbm_fifo_wait_for_entry(rdev, 64)) {
1260                 printk(KERN_WARNING "radeon: wait for empty RBBM fifo failed !"
1261                        " Bad things might happen.\n");
1262         }
1263         for (i = 0; i < rdev->usec_timeout; i++) {
1264                 tmp = RREG32(RADEON_RBBM_STATUS);
1265                 if (!(tmp & (1 << 31))) {
1266                         return 0;
1267                 }
1268                 DRM_UDELAY(1);
1269         }
1270         return -1;
1271 }
1272
1273 int r100_mc_wait_for_idle(struct radeon_device *rdev)
1274 {
1275         unsigned i;
1276         uint32_t tmp;
1277
1278         for (i = 0; i < rdev->usec_timeout; i++) {
1279                 /* read MC_STATUS */
1280                 tmp = RREG32(0x0150);
1281                 if (tmp & (1 << 2)) {
1282                         return 0;
1283                 }
1284                 DRM_UDELAY(1);
1285         }
1286         return -1;
1287 }
1288
1289 void r100_gpu_init(struct radeon_device *rdev)
1290 {
1291         /* TODO: anythings to do here ? pipes ? */
1292         r100_hdp_reset(rdev);
1293 }
1294
1295 void r100_hdp_reset(struct radeon_device *rdev)
1296 {
1297         uint32_t tmp;
1298
1299         tmp = RREG32(RADEON_HOST_PATH_CNTL) & RADEON_HDP_APER_CNTL;
1300         tmp |= (7 << 28);
1301         WREG32(RADEON_HOST_PATH_CNTL, tmp | RADEON_HDP_SOFT_RESET | RADEON_HDP_READ_BUFFER_INVALIDATE);
1302         (void)RREG32(RADEON_HOST_PATH_CNTL);
1303         udelay(200);
1304         WREG32(RADEON_RBBM_SOFT_RESET, 0);
1305         WREG32(RADEON_HOST_PATH_CNTL, tmp);
1306         (void)RREG32(RADEON_HOST_PATH_CNTL);
1307 }
1308
1309 int r100_rb2d_reset(struct radeon_device *rdev)
1310 {
1311         uint32_t tmp;
1312         int i;
1313
1314         WREG32(RADEON_RBBM_SOFT_RESET, RADEON_SOFT_RESET_E2);
1315         (void)RREG32(RADEON_RBBM_SOFT_RESET);
1316         udelay(200);
1317         WREG32(RADEON_RBBM_SOFT_RESET, 0);
1318         /* Wait to prevent race in RBBM_STATUS */
1319         mdelay(1);
1320         for (i = 0; i < rdev->usec_timeout; i++) {
1321                 tmp = RREG32(RADEON_RBBM_STATUS);
1322                 if (!(tmp & (1 << 26))) {
1323                         DRM_INFO("RB2D reset succeed (RBBM_STATUS=0x%08X)\n",
1324                                  tmp);
1325                         return 0;
1326                 }
1327                 DRM_UDELAY(1);
1328         }
1329         tmp = RREG32(RADEON_RBBM_STATUS);
1330         DRM_ERROR("Failed to reset RB2D (RBBM_STATUS=0x%08X)!\n", tmp);
1331         return -1;
1332 }
1333
1334 int r100_gpu_reset(struct radeon_device *rdev)
1335 {
1336         uint32_t status;
1337
1338         /* reset order likely matter */
1339         status = RREG32(RADEON_RBBM_STATUS);
1340         /* reset HDP */
1341         r100_hdp_reset(rdev);
1342         /* reset rb2d */
1343         if (status & ((1 << 17) | (1 << 18) | (1 << 27))) {
1344                 r100_rb2d_reset(rdev);
1345         }
1346         /* TODO: reset 3D engine */
1347         /* reset CP */
1348         status = RREG32(RADEON_RBBM_STATUS);
1349         if (status & (1 << 16)) {
1350                 r100_cp_reset(rdev);
1351         }
1352         /* Check if GPU is idle */
1353         status = RREG32(RADEON_RBBM_STATUS);
1354         if (status & (1 << 31)) {
1355                 DRM_ERROR("Failed to reset GPU (RBBM_STATUS=0x%08X)\n", status);
1356                 return -1;
1357         }
1358         DRM_INFO("GPU reset succeed (RBBM_STATUS=0x%08X)\n", status);
1359         return 0;
1360 }
1361
1362
1363 /*
1364  * VRAM info
1365  */
1366 static void r100_vram_get_type(struct radeon_device *rdev)
1367 {
1368         uint32_t tmp;
1369
1370         rdev->mc.vram_is_ddr = false;
1371         if (rdev->flags & RADEON_IS_IGP)
1372                 rdev->mc.vram_is_ddr = true;
1373         else if (RREG32(RADEON_MEM_SDRAM_MODE_REG) & RADEON_MEM_CFG_TYPE_DDR)
1374                 rdev->mc.vram_is_ddr = true;
1375         if ((rdev->family == CHIP_RV100) ||
1376             (rdev->family == CHIP_RS100) ||
1377             (rdev->family == CHIP_RS200)) {
1378                 tmp = RREG32(RADEON_MEM_CNTL);
1379                 if (tmp & RV100_HALF_MODE) {
1380                         rdev->mc.vram_width = 32;
1381                 } else {
1382                         rdev->mc.vram_width = 64;
1383                 }
1384                 if (rdev->flags & RADEON_SINGLE_CRTC) {
1385                         rdev->mc.vram_width /= 4;
1386                         rdev->mc.vram_is_ddr = true;
1387                 }
1388         } else if (rdev->family <= CHIP_RV280) {
1389                 tmp = RREG32(RADEON_MEM_CNTL);
1390                 if (tmp & RADEON_MEM_NUM_CHANNELS_MASK) {
1391                         rdev->mc.vram_width = 128;
1392                 } else {
1393                         rdev->mc.vram_width = 64;
1394                 }
1395         } else {
1396                 /* newer IGPs */
1397                 rdev->mc.vram_width = 128;
1398         }
1399 }
1400
1401 static u32 r100_get_accessible_vram(struct radeon_device *rdev)
1402 {
1403         u32 aper_size;
1404         u8 byte;
1405
1406         aper_size = RREG32(RADEON_CONFIG_APER_SIZE);
1407
1408         /* Set HDP_APER_CNTL only on cards that are known not to be broken,
1409          * that is has the 2nd generation multifunction PCI interface
1410          */
1411         if (rdev->family == CHIP_RV280 ||
1412             rdev->family >= CHIP_RV350) {
1413                 WREG32_P(RADEON_HOST_PATH_CNTL, RADEON_HDP_APER_CNTL,
1414                        ~RADEON_HDP_APER_CNTL);
1415                 DRM_INFO("Generation 2 PCI interface, using max accessible memory\n");
1416                 return aper_size * 2;
1417         }
1418
1419         /* Older cards have all sorts of funny issues to deal with. First
1420          * check if it's a multifunction card by reading the PCI config
1421          * header type... Limit those to one aperture size
1422          */
1423         pci_read_config_byte(rdev->pdev, 0xe, &byte);
1424         if (byte & 0x80) {
1425                 DRM_INFO("Generation 1 PCI interface in multifunction mode\n");
1426                 DRM_INFO("Limiting VRAM to one aperture\n");
1427                 return aper_size;
1428         }
1429
1430         /* Single function older card. We read HDP_APER_CNTL to see how the BIOS
1431          * have set it up. We don't write this as it's broken on some ASICs but
1432          * we expect the BIOS to have done the right thing (might be too optimistic...)
1433          */
1434         if (RREG32(RADEON_HOST_PATH_CNTL) & RADEON_HDP_APER_CNTL)
1435                 return aper_size * 2;
1436         return aper_size;
1437 }
1438
1439 void r100_vram_init_sizes(struct radeon_device *rdev)
1440 {
1441         u64 config_aper_size;
1442         u32 accessible;
1443
1444         config_aper_size = RREG32(RADEON_CONFIG_APER_SIZE);
1445
1446         if (rdev->flags & RADEON_IS_IGP) {
1447                 uint32_t tom;
1448                 /* read NB_TOM to get the amount of ram stolen for the GPU */
1449                 tom = RREG32(RADEON_NB_TOM);
1450                 rdev->mc.vram_size = (((tom >> 16) - (tom & 0xffff) + 1) << 16);
1451                 /* for IGPs we need to keep VRAM where it was put by the BIOS */
1452                 rdev->mc.vram_location = (tom & 0xffff) << 16;
1453                 WREG32(RADEON_CONFIG_MEMSIZE, rdev->mc.vram_size);
1454         } else {
1455                 rdev->mc.vram_size = RREG32(RADEON_CONFIG_MEMSIZE);
1456                 /* Some production boards of m6 will report 0
1457                  * if it's 8 MB
1458                  */
1459                 if (rdev->mc.vram_size == 0) {
1460                         rdev->mc.vram_size = 8192 * 1024;
1461                         WREG32(RADEON_CONFIG_MEMSIZE, rdev->mc.vram_size);
1462                 }
1463                 /* let driver place VRAM */
1464                 rdev->mc.vram_location = 0xFFFFFFFFUL;
1465                  /* Fix for RN50, M6, M7 with 8/16/32(??) MBs of VRAM -
1466                   * Novell bug 204882 + along with lots of ubuntu ones */
1467                 if (config_aper_size > rdev->mc.vram_size)
1468                         rdev->mc.vram_size = config_aper_size;
1469         }
1470
1471         /* work out accessible VRAM */
1472         accessible = r100_get_accessible_vram(rdev);
1473
1474         rdev->mc.aper_base = drm_get_resource_start(rdev->ddev, 0);
1475         rdev->mc.aper_size = drm_get_resource_len(rdev->ddev, 0);
1476
1477         if (accessible > rdev->mc.aper_size)
1478                 accessible = rdev->mc.aper_size;
1479
1480         if (rdev->mc.vram_size > rdev->mc.aper_size)
1481                 rdev->mc.vram_size = rdev->mc.aper_size;
1482 }
1483
1484 void r100_vram_info(struct radeon_device *rdev)
1485 {
1486         r100_vram_get_type(rdev);
1487
1488         r100_vram_init_sizes(rdev);
1489 }
1490
1491
1492 /*
1493  * Indirect registers accessor
1494  */
1495 void r100_pll_errata_after_index(struct radeon_device *rdev)
1496 {
1497         if (!(rdev->pll_errata & CHIP_ERRATA_PLL_DUMMYREADS)) {
1498                 return;
1499         }
1500         (void)RREG32(RADEON_CLOCK_CNTL_DATA);
1501         (void)RREG32(RADEON_CRTC_GEN_CNTL);
1502 }
1503
1504 static void r100_pll_errata_after_data(struct radeon_device *rdev)
1505 {
1506         /* This workarounds is necessary on RV100, RS100 and RS200 chips
1507          * or the chip could hang on a subsequent access
1508          */
1509         if (rdev->pll_errata & CHIP_ERRATA_PLL_DELAY) {
1510                 udelay(5000);
1511         }
1512
1513         /* This function is required to workaround a hardware bug in some (all?)
1514          * revisions of the R300.  This workaround should be called after every
1515          * CLOCK_CNTL_INDEX register access.  If not, register reads afterward
1516          * may not be correct.
1517          */
1518         if (rdev->pll_errata & CHIP_ERRATA_R300_CG) {
1519                 uint32_t save, tmp;
1520
1521                 save = RREG32(RADEON_CLOCK_CNTL_INDEX);
1522                 tmp = save & ~(0x3f | RADEON_PLL_WR_EN);
1523                 WREG32(RADEON_CLOCK_CNTL_INDEX, tmp);
1524                 tmp = RREG32(RADEON_CLOCK_CNTL_DATA);
1525                 WREG32(RADEON_CLOCK_CNTL_INDEX, save);
1526         }
1527 }
1528
1529 uint32_t r100_pll_rreg(struct radeon_device *rdev, uint32_t reg)
1530 {
1531         uint32_t data;
1532
1533         WREG8(RADEON_CLOCK_CNTL_INDEX, reg & 0x3f);
1534         r100_pll_errata_after_index(rdev);
1535         data = RREG32(RADEON_CLOCK_CNTL_DATA);
1536         r100_pll_errata_after_data(rdev);
1537         return data;
1538 }
1539
1540 void r100_pll_wreg(struct radeon_device *rdev, uint32_t reg, uint32_t v)
1541 {
1542         WREG8(RADEON_CLOCK_CNTL_INDEX, ((reg & 0x3f) | RADEON_PLL_WR_EN));
1543         r100_pll_errata_after_index(rdev);
1544         WREG32(RADEON_CLOCK_CNTL_DATA, v);
1545         r100_pll_errata_after_data(rdev);
1546 }
1547
1548 uint32_t r100_mm_rreg(struct radeon_device *rdev, uint32_t reg)
1549 {
1550         if (reg < 0x10000)
1551                 return readl(((void __iomem *)rdev->rmmio) + reg);
1552         else {
1553                 writel(reg, ((void __iomem *)rdev->rmmio) + RADEON_MM_INDEX);
1554                 return readl(((void __iomem *)rdev->rmmio) + RADEON_MM_DATA);
1555         }
1556 }
1557
1558 void r100_mm_wreg(struct radeon_device *rdev, uint32_t reg, uint32_t v)
1559 {
1560         if (reg < 0x10000)
1561                 writel(v, ((void __iomem *)rdev->rmmio) + reg);
1562         else {
1563                 writel(reg, ((void __iomem *)rdev->rmmio) + RADEON_MM_INDEX);
1564                 writel(v, ((void __iomem *)rdev->rmmio) + RADEON_MM_DATA);
1565         }
1566 }
1567
1568 int r100_init(struct radeon_device *rdev)
1569 {
1570         return 0;
1571 }
1572
1573 /*
1574  * Debugfs info
1575  */
1576 #if defined(CONFIG_DEBUG_FS)
1577 static int r100_debugfs_rbbm_info(struct seq_file *m, void *data)
1578 {
1579         struct drm_info_node *node = (struct drm_info_node *) m->private;
1580         struct drm_device *dev = node->minor->dev;
1581         struct radeon_device *rdev = dev->dev_private;
1582         uint32_t reg, value;
1583         unsigned i;
1584
1585         seq_printf(m, "RBBM_STATUS 0x%08x\n", RREG32(RADEON_RBBM_STATUS));
1586         seq_printf(m, "RBBM_CMDFIFO_STAT 0x%08x\n", RREG32(0xE7C));
1587         seq_printf(m, "CP_STAT 0x%08x\n", RREG32(RADEON_CP_STAT));
1588         for (i = 0; i < 64; i++) {
1589                 WREG32(RADEON_RBBM_CMDFIFO_ADDR, i | 0x100);
1590                 reg = (RREG32(RADEON_RBBM_CMDFIFO_DATA) - 1) >> 2;
1591                 WREG32(RADEON_RBBM_CMDFIFO_ADDR, i);
1592                 value = RREG32(RADEON_RBBM_CMDFIFO_DATA);
1593                 seq_printf(m, "[0x%03X] 0x%04X=0x%08X\n", i, reg, value);
1594         }
1595         return 0;
1596 }
1597
1598 static int r100_debugfs_cp_ring_info(struct seq_file *m, void *data)
1599 {
1600         struct drm_info_node *node = (struct drm_info_node *) m->private;
1601         struct drm_device *dev = node->minor->dev;
1602         struct radeon_device *rdev = dev->dev_private;
1603         uint32_t rdp, wdp;
1604         unsigned count, i, j;
1605
1606         radeon_ring_free_size(rdev);
1607         rdp = RREG32(RADEON_CP_RB_RPTR);
1608         wdp = RREG32(RADEON_CP_RB_WPTR);
1609         count = (rdp + rdev->cp.ring_size - wdp) & rdev->cp.ptr_mask;
1610         seq_printf(m, "CP_STAT 0x%08x\n", RREG32(RADEON_CP_STAT));
1611         seq_printf(m, "CP_RB_WPTR 0x%08x\n", wdp);
1612         seq_printf(m, "CP_RB_RPTR 0x%08x\n", rdp);
1613         seq_printf(m, "%u free dwords in ring\n", rdev->cp.ring_free_dw);
1614         seq_printf(m, "%u dwords in ring\n", count);
1615         for (j = 0; j <= count; j++) {
1616                 i = (rdp + j) & rdev->cp.ptr_mask;
1617                 seq_printf(m, "r[%04d]=0x%08x\n", i, rdev->cp.ring[i]);
1618         }
1619         return 0;
1620 }
1621
1622
1623 static int r100_debugfs_cp_csq_fifo(struct seq_file *m, void *data)
1624 {
1625         struct drm_info_node *node = (struct drm_info_node *) m->private;
1626         struct drm_device *dev = node->minor->dev;
1627         struct radeon_device *rdev = dev->dev_private;
1628         uint32_t csq_stat, csq2_stat, tmp;
1629         unsigned r_rptr, r_wptr, ib1_rptr, ib1_wptr, ib2_rptr, ib2_wptr;
1630         unsigned i;
1631
1632         seq_printf(m, "CP_STAT 0x%08x\n", RREG32(RADEON_CP_STAT));
1633         seq_printf(m, "CP_CSQ_MODE 0x%08x\n", RREG32(RADEON_CP_CSQ_MODE));
1634         csq_stat = RREG32(RADEON_CP_CSQ_STAT);
1635         csq2_stat = RREG32(RADEON_CP_CSQ2_STAT);
1636         r_rptr = (csq_stat >> 0) & 0x3ff;
1637         r_wptr = (csq_stat >> 10) & 0x3ff;
1638         ib1_rptr = (csq_stat >> 20) & 0x3ff;
1639         ib1_wptr = (csq2_stat >> 0) & 0x3ff;
1640         ib2_rptr = (csq2_stat >> 10) & 0x3ff;
1641         ib2_wptr = (csq2_stat >> 20) & 0x3ff;
1642         seq_printf(m, "CP_CSQ_STAT 0x%08x\n", csq_stat);
1643         seq_printf(m, "CP_CSQ2_STAT 0x%08x\n", csq2_stat);
1644         seq_printf(m, "Ring rptr %u\n", r_rptr);
1645         seq_printf(m, "Ring wptr %u\n", r_wptr);
1646         seq_printf(m, "Indirect1 rptr %u\n", ib1_rptr);
1647         seq_printf(m, "Indirect1 wptr %u\n", ib1_wptr);
1648         seq_printf(m, "Indirect2 rptr %u\n", ib2_rptr);
1649         seq_printf(m, "Indirect2 wptr %u\n", ib2_wptr);
1650         /* FIXME: 0, 128, 640 depends on fifo setup see cp_init_kms
1651          * 128 = indirect1_start * 8 & 640 = indirect2_start * 8 */
1652         seq_printf(m, "Ring fifo:\n");
1653         for (i = 0; i < 256; i++) {
1654                 WREG32(RADEON_CP_CSQ_ADDR, i << 2);
1655                 tmp = RREG32(RADEON_CP_CSQ_DATA);
1656                 seq_printf(m, "rfifo[%04d]=0x%08X\n", i, tmp);
1657         }
1658         seq_printf(m, "Indirect1 fifo:\n");
1659         for (i = 256; i <= 512; i++) {
1660                 WREG32(RADEON_CP_CSQ_ADDR, i << 2);
1661                 tmp = RREG32(RADEON_CP_CSQ_DATA);
1662                 seq_printf(m, "ib1fifo[%04d]=0x%08X\n", i, tmp);
1663         }
1664         seq_printf(m, "Indirect2 fifo:\n");
1665         for (i = 640; i < ib1_wptr; i++) {
1666                 WREG32(RADEON_CP_CSQ_ADDR, i << 2);
1667                 tmp = RREG32(RADEON_CP_CSQ_DATA);
1668                 seq_printf(m, "ib2fifo[%04d]=0x%08X\n", i, tmp);
1669         }
1670         return 0;
1671 }
1672
1673 static int r100_debugfs_mc_info(struct seq_file *m, void *data)
1674 {
1675         struct drm_info_node *node = (struct drm_info_node *) m->private;
1676         struct drm_device *dev = node->minor->dev;
1677         struct radeon_device *rdev = dev->dev_private;
1678         uint32_t tmp;
1679
1680         tmp = RREG32(RADEON_CONFIG_MEMSIZE);
1681         seq_printf(m, "CONFIG_MEMSIZE 0x%08x\n", tmp);
1682         tmp = RREG32(RADEON_MC_FB_LOCATION);
1683         seq_printf(m, "MC_FB_LOCATION 0x%08x\n", tmp);
1684         tmp = RREG32(RADEON_BUS_CNTL);
1685         seq_printf(m, "BUS_CNTL 0x%08x\n", tmp);
1686         tmp = RREG32(RADEON_MC_AGP_LOCATION);
1687         seq_printf(m, "MC_AGP_LOCATION 0x%08x\n", tmp);
1688         tmp = RREG32(RADEON_AGP_BASE);
1689         seq_printf(m, "AGP_BASE 0x%08x\n", tmp);
1690         tmp = RREG32(RADEON_HOST_PATH_CNTL);
1691         seq_printf(m, "HOST_PATH_CNTL 0x%08x\n", tmp);
1692         tmp = RREG32(0x01D0);
1693         seq_printf(m, "AIC_CTRL 0x%08x\n", tmp);
1694         tmp = RREG32(RADEON_AIC_LO_ADDR);
1695         seq_printf(m, "AIC_LO_ADDR 0x%08x\n", tmp);
1696         tmp = RREG32(RADEON_AIC_HI_ADDR);
1697         seq_printf(m, "AIC_HI_ADDR 0x%08x\n", tmp);
1698         tmp = RREG32(0x01E4);
1699         seq_printf(m, "AIC_TLB_ADDR 0x%08x\n", tmp);
1700         return 0;
1701 }
1702
1703 static struct drm_info_list r100_debugfs_rbbm_list[] = {
1704         {"r100_rbbm_info", r100_debugfs_rbbm_info, 0, NULL},
1705 };
1706
1707 static struct drm_info_list r100_debugfs_cp_list[] = {
1708         {"r100_cp_ring_info", r100_debugfs_cp_ring_info, 0, NULL},
1709         {"r100_cp_csq_fifo", r100_debugfs_cp_csq_fifo, 0, NULL},
1710 };
1711
1712 static struct drm_info_list r100_debugfs_mc_info_list[] = {
1713         {"r100_mc_info", r100_debugfs_mc_info, 0, NULL},
1714 };
1715 #endif
1716
1717 int r100_debugfs_rbbm_init(struct radeon_device *rdev)
1718 {
1719 #if defined(CONFIG_DEBUG_FS)
1720         return radeon_debugfs_add_files(rdev, r100_debugfs_rbbm_list, 1);
1721 #else
1722         return 0;
1723 #endif
1724 }
1725
1726 int r100_debugfs_cp_init(struct radeon_device *rdev)
1727 {
1728 #if defined(CONFIG_DEBUG_FS)
1729         return radeon_debugfs_add_files(rdev, r100_debugfs_cp_list, 2);
1730 #else
1731         return 0;
1732 #endif
1733 }
1734
1735 int r100_debugfs_mc_info_init(struct radeon_device *rdev)
1736 {
1737 #if defined(CONFIG_DEBUG_FS)
1738         return radeon_debugfs_add_files(rdev, r100_debugfs_mc_info_list, 1);
1739 #else
1740         return 0;
1741 #endif
1742 }
1743
1744 int r100_set_surface_reg(struct radeon_device *rdev, int reg,
1745                          uint32_t tiling_flags, uint32_t pitch,
1746                          uint32_t offset, uint32_t obj_size)
1747 {
1748         int surf_index = reg * 16;
1749         int flags = 0;
1750
1751         /* r100/r200 divide by 16 */
1752         if (rdev->family < CHIP_R300)
1753                 flags = pitch / 16;
1754         else
1755                 flags = pitch / 8;
1756
1757         if (rdev->family <= CHIP_RS200) {
1758                 if ((tiling_flags & (RADEON_TILING_MACRO|RADEON_TILING_MICRO))
1759                                  == (RADEON_TILING_MACRO|RADEON_TILING_MICRO))
1760                         flags |= RADEON_SURF_TILE_COLOR_BOTH;
1761                 if (tiling_flags & RADEON_TILING_MACRO)
1762                         flags |= RADEON_SURF_TILE_COLOR_MACRO;
1763         } else if (rdev->family <= CHIP_RV280) {
1764                 if (tiling_flags & (RADEON_TILING_MACRO))
1765                         flags |= R200_SURF_TILE_COLOR_MACRO;
1766                 if (tiling_flags & RADEON_TILING_MICRO)
1767                         flags |= R200_SURF_TILE_COLOR_MICRO;
1768         } else {
1769                 if (tiling_flags & RADEON_TILING_MACRO)
1770                         flags |= R300_SURF_TILE_MACRO;
1771                 if (tiling_flags & RADEON_TILING_MICRO)
1772                         flags |= R300_SURF_TILE_MICRO;
1773         }
1774
1775         DRM_DEBUG("writing surface %d %d %x %x\n", reg, flags, offset, offset+obj_size-1);
1776         WREG32(RADEON_SURFACE0_INFO + surf_index, flags);
1777         WREG32(RADEON_SURFACE0_LOWER_BOUND + surf_index, offset);
1778         WREG32(RADEON_SURFACE0_UPPER_BOUND + surf_index, offset + obj_size - 1);
1779         return 0;
1780 }
1781
1782 void r100_clear_surface_reg(struct radeon_device *rdev, int reg)
1783 {
1784         int surf_index = reg * 16;
1785         WREG32(RADEON_SURFACE0_INFO + surf_index, 0);
1786 }
1787
1788 void r100_bandwidth_update(struct radeon_device *rdev)
1789 {
1790         fixed20_12 trcd_ff, trp_ff, tras_ff, trbs_ff, tcas_ff;
1791         fixed20_12 sclk_ff, mclk_ff, sclk_eff_ff, sclk_delay_ff;
1792         fixed20_12 peak_disp_bw, mem_bw, pix_clk, pix_clk2, temp_ff, crit_point_ff;
1793         uint32_t temp, data, mem_trcd, mem_trp, mem_tras;
1794         fixed20_12 memtcas_ff[8] = {
1795                 fixed_init(1),
1796                 fixed_init(2),
1797                 fixed_init(3),
1798                 fixed_init(0),
1799                 fixed_init_half(1),
1800                 fixed_init_half(2),
1801                 fixed_init(0),
1802         };
1803         fixed20_12 memtcas_rs480_ff[8] = {
1804                 fixed_init(0),
1805                 fixed_init(1),
1806                 fixed_init(2),
1807                 fixed_init(3),
1808                 fixed_init(0),
1809                 fixed_init_half(1),
1810                 fixed_init_half(2),
1811                 fixed_init_half(3),
1812         };
1813         fixed20_12 memtcas2_ff[8] = {
1814                 fixed_init(0),
1815                 fixed_init(1),
1816                 fixed_init(2),
1817                 fixed_init(3),
1818                 fixed_init(4),
1819                 fixed_init(5),
1820                 fixed_init(6),
1821                 fixed_init(7),
1822         };
1823         fixed20_12 memtrbs[8] = {
1824                 fixed_init(1),
1825                 fixed_init_half(1),
1826                 fixed_init(2),
1827                 fixed_init_half(2),
1828                 fixed_init(3),
1829                 fixed_init_half(3),
1830                 fixed_init(4),
1831                 fixed_init_half(4)
1832         };
1833         fixed20_12 memtrbs_r4xx[8] = {
1834                 fixed_init(4),
1835                 fixed_init(5),
1836                 fixed_init(6),
1837                 fixed_init(7),
1838                 fixed_init(8),
1839                 fixed_init(9),
1840                 fixed_init(10),
1841                 fixed_init(11)
1842         };
1843         fixed20_12 min_mem_eff;
1844         fixed20_12 mc_latency_sclk, mc_latency_mclk, k1;
1845         fixed20_12 cur_latency_mclk, cur_latency_sclk;
1846         fixed20_12 disp_latency, disp_latency_overhead, disp_drain_rate,
1847                 disp_drain_rate2, read_return_rate;
1848         fixed20_12 time_disp1_drop_priority;
1849         int c;
1850         int cur_size = 16;       /* in octawords */
1851         int critical_point = 0, critical_point2;
1852 /*      uint32_t read_return_rate, time_disp1_drop_priority; */
1853         int stop_req, max_stop_req;
1854         struct drm_display_mode *mode1 = NULL;
1855         struct drm_display_mode *mode2 = NULL;
1856         uint32_t pixel_bytes1 = 0;
1857         uint32_t pixel_bytes2 = 0;
1858
1859         if (rdev->mode_info.crtcs[0]->base.enabled) {
1860                 mode1 = &rdev->mode_info.crtcs[0]->base.mode;
1861                 pixel_bytes1 = rdev->mode_info.crtcs[0]->base.fb->bits_per_pixel / 8;
1862         }
1863         if (rdev->mode_info.crtcs[1]->base.enabled) {
1864                 mode2 = &rdev->mode_info.crtcs[1]->base.mode;
1865                 pixel_bytes2 = rdev->mode_info.crtcs[1]->base.fb->bits_per_pixel / 8;
1866         }
1867
1868         min_mem_eff.full = rfixed_const_8(0);
1869         /* get modes */
1870         if ((rdev->disp_priority == 2) && ASIC_IS_R300(rdev)) {
1871                 uint32_t mc_init_misc_lat_timer = RREG32(R300_MC_INIT_MISC_LAT_TIMER);
1872                 mc_init_misc_lat_timer &= ~(R300_MC_DISP1R_INIT_LAT_MASK << R300_MC_DISP1R_INIT_LAT_SHIFT);
1873                 mc_init_misc_lat_timer &= ~(R300_MC_DISP0R_INIT_LAT_MASK << R300_MC_DISP0R_INIT_LAT_SHIFT);
1874                 /* check crtc enables */
1875                 if (mode2)
1876                         mc_init_misc_lat_timer |= (1 << R300_MC_DISP1R_INIT_LAT_SHIFT);
1877                 if (mode1)
1878                         mc_init_misc_lat_timer |= (1 << R300_MC_DISP0R_INIT_LAT_SHIFT);
1879                 WREG32(R300_MC_INIT_MISC_LAT_TIMER, mc_init_misc_lat_timer);
1880         }
1881
1882         /*
1883          * determine is there is enough bw for current mode
1884          */
1885         mclk_ff.full = rfixed_const(rdev->clock.default_mclk);
1886         temp_ff.full = rfixed_const(100);
1887         mclk_ff.full = rfixed_div(mclk_ff, temp_ff);
1888         sclk_ff.full = rfixed_const(rdev->clock.default_sclk);
1889         sclk_ff.full = rfixed_div(sclk_ff, temp_ff);
1890
1891         temp = (rdev->mc.vram_width / 8) * (rdev->mc.vram_is_ddr ? 2 : 1);
1892         temp_ff.full = rfixed_const(temp);
1893         mem_bw.full = rfixed_mul(mclk_ff, temp_ff);
1894
1895         pix_clk.full = 0;
1896         pix_clk2.full = 0;
1897         peak_disp_bw.full = 0;
1898         if (mode1) {
1899                 temp_ff.full = rfixed_const(1000);
1900                 pix_clk.full = rfixed_const(mode1->clock); /* convert to fixed point */
1901                 pix_clk.full = rfixed_div(pix_clk, temp_ff);
1902                 temp_ff.full = rfixed_const(pixel_bytes1);
1903                 peak_disp_bw.full += rfixed_mul(pix_clk, temp_ff);
1904         }
1905         if (mode2) {
1906                 temp_ff.full = rfixed_const(1000);
1907                 pix_clk2.full = rfixed_const(mode2->clock); /* convert to fixed point */
1908                 pix_clk2.full = rfixed_div(pix_clk2, temp_ff);
1909                 temp_ff.full = rfixed_const(pixel_bytes2);
1910                 peak_disp_bw.full += rfixed_mul(pix_clk2, temp_ff);
1911         }
1912
1913         mem_bw.full = rfixed_mul(mem_bw, min_mem_eff);
1914         if (peak_disp_bw.full >= mem_bw.full) {
1915                 DRM_ERROR("You may not have enough display bandwidth for current mode\n"
1916                           "If you have flickering problem, try to lower resolution, refresh rate, or color depth\n");
1917         }
1918
1919         /*  Get values from the EXT_MEM_CNTL register...converting its contents. */
1920         temp = RREG32(RADEON_MEM_TIMING_CNTL);
1921         if ((rdev->family == CHIP_RV100) || (rdev->flags & RADEON_IS_IGP)) { /* RV100, M6, IGPs */
1922                 mem_trcd = ((temp >> 2) & 0x3) + 1;
1923                 mem_trp  = ((temp & 0x3)) + 1;
1924                 mem_tras = ((temp & 0x70) >> 4) + 1;
1925         } else if (rdev->family == CHIP_R300 ||
1926                    rdev->family == CHIP_R350) { /* r300, r350 */
1927                 mem_trcd = (temp & 0x7) + 1;
1928                 mem_trp = ((temp >> 8) & 0x7) + 1;
1929                 mem_tras = ((temp >> 11) & 0xf) + 4;
1930         } else if (rdev->family == CHIP_RV350 ||
1931                    rdev->family <= CHIP_RV380) {
1932                 /* rv3x0 */
1933                 mem_trcd = (temp & 0x7) + 3;
1934                 mem_trp = ((temp >> 8) & 0x7) + 3;
1935                 mem_tras = ((temp >> 11) & 0xf) + 6;
1936         } else if (rdev->family == CHIP_R420 ||
1937                    rdev->family == CHIP_R423 ||
1938                    rdev->family == CHIP_RV410) {
1939                 /* r4xx */
1940                 mem_trcd = (temp & 0xf) + 3;
1941                 if (mem_trcd > 15)
1942                         mem_trcd = 15;
1943                 mem_trp = ((temp >> 8) & 0xf) + 3;
1944                 if (mem_trp > 15)
1945                         mem_trp = 15;
1946                 mem_tras = ((temp >> 12) & 0x1f) + 6;
1947                 if (mem_tras > 31)
1948                         mem_tras = 31;
1949         } else { /* RV200, R200 */
1950                 mem_trcd = (temp & 0x7) + 1;
1951                 mem_trp = ((temp >> 8) & 0x7) + 1;
1952                 mem_tras = ((temp >> 12) & 0xf) + 4;
1953         }
1954         /* convert to FF */
1955         trcd_ff.full = rfixed_const(mem_trcd);
1956         trp_ff.full = rfixed_const(mem_trp);
1957         tras_ff.full = rfixed_const(mem_tras);
1958
1959         /* Get values from the MEM_SDRAM_MODE_REG register...converting its */
1960         temp = RREG32(RADEON_MEM_SDRAM_MODE_REG);
1961         data = (temp & (7 << 20)) >> 20;
1962         if ((rdev->family == CHIP_RV100) || rdev->flags & RADEON_IS_IGP) {
1963                 if (rdev->family == CHIP_RS480) /* don't think rs400 */
1964                         tcas_ff = memtcas_rs480_ff[data];
1965                 else
1966                         tcas_ff = memtcas_ff[data];
1967         } else
1968                 tcas_ff = memtcas2_ff[data];
1969
1970         if (rdev->family == CHIP_RS400 ||
1971             rdev->family == CHIP_RS480) {
1972                 /* extra cas latency stored in bits 23-25 0-4 clocks */
1973                 data = (temp >> 23) & 0x7;
1974                 if (data < 5)
1975                         tcas_ff.full += rfixed_const(data);
1976         }
1977
1978         if (ASIC_IS_R300(rdev) && !(rdev->flags & RADEON_IS_IGP)) {
1979                 /* on the R300, Tcas is included in Trbs.
1980                  */
1981                 temp = RREG32(RADEON_MEM_CNTL);
1982                 data = (R300_MEM_NUM_CHANNELS_MASK & temp);
1983                 if (data == 1) {
1984                         if (R300_MEM_USE_CD_CH_ONLY & temp) {
1985                                 temp = RREG32(R300_MC_IND_INDEX);
1986                                 temp &= ~R300_MC_IND_ADDR_MASK;
1987                                 temp |= R300_MC_READ_CNTL_CD_mcind;
1988                                 WREG32(R300_MC_IND_INDEX, temp);
1989                                 temp = RREG32(R300_MC_IND_DATA);
1990                                 data = (R300_MEM_RBS_POSITION_C_MASK & temp);
1991                         } else {
1992                                 temp = RREG32(R300_MC_READ_CNTL_AB);
1993                                 data = (R300_MEM_RBS_POSITION_A_MASK & temp);
1994                         }
1995                 } else {
1996                         temp = RREG32(R300_MC_READ_CNTL_AB);
1997                         data = (R300_MEM_RBS_POSITION_A_MASK & temp);
1998                 }
1999                 if (rdev->family == CHIP_RV410 ||
2000                     rdev->family == CHIP_R420 ||
2001                     rdev->family == CHIP_R423)
2002                         trbs_ff = memtrbs_r4xx[data];
2003                 else
2004                         trbs_ff = memtrbs[data];
2005                 tcas_ff.full += trbs_ff.full;
2006         }
2007
2008         sclk_eff_ff.full = sclk_ff.full;
2009
2010         if (rdev->flags & RADEON_IS_AGP) {
2011                 fixed20_12 agpmode_ff;
2012                 agpmode_ff.full = rfixed_const(radeon_agpmode);
2013                 temp_ff.full = rfixed_const_666(16);
2014                 sclk_eff_ff.full -= rfixed_mul(agpmode_ff, temp_ff);
2015         }
2016         /* TODO PCIE lanes may affect this - agpmode == 16?? */
2017
2018         if (ASIC_IS_R300(rdev)) {
2019                 sclk_delay_ff.full = rfixed_const(250);
2020         } else {
2021                 if ((rdev->family == CHIP_RV100) ||
2022                     rdev->flags & RADEON_IS_IGP) {
2023                         if (rdev->mc.vram_is_ddr)
2024                                 sclk_delay_ff.full = rfixed_const(41);
2025                         else
2026                                 sclk_delay_ff.full = rfixed_const(33);
2027                 } else {
2028                         if (rdev->mc.vram_width == 128)
2029                                 sclk_delay_ff.full = rfixed_const(57);
2030                         else
2031                                 sclk_delay_ff.full = rfixed_const(41);
2032                 }
2033         }
2034
2035         mc_latency_sclk.full = rfixed_div(sclk_delay_ff, sclk_eff_ff);
2036
2037         if (rdev->mc.vram_is_ddr) {
2038                 if (rdev->mc.vram_width == 32) {
2039                         k1.full = rfixed_const(40);
2040                         c  = 3;
2041                 } else {
2042                         k1.full = rfixed_const(20);
2043                         c  = 1;
2044                 }
2045         } else {
2046                 k1.full = rfixed_const(40);
2047                 c  = 3;
2048         }
2049
2050         temp_ff.full = rfixed_const(2);
2051         mc_latency_mclk.full = rfixed_mul(trcd_ff, temp_ff);
2052         temp_ff.full = rfixed_const(c);
2053         mc_latency_mclk.full += rfixed_mul(tcas_ff, temp_ff);
2054         temp_ff.full = rfixed_const(4);
2055         mc_latency_mclk.full += rfixed_mul(tras_ff, temp_ff);
2056         mc_latency_mclk.full += rfixed_mul(trp_ff, temp_ff);
2057         mc_latency_mclk.full += k1.full;
2058
2059         mc_latency_mclk.full = rfixed_div(mc_latency_mclk, mclk_ff);
2060         mc_latency_mclk.full += rfixed_div(temp_ff, sclk_eff_ff);
2061
2062         /*
2063           HW cursor time assuming worst case of full size colour cursor.
2064         */
2065         temp_ff.full = rfixed_const((2 * (cur_size - (rdev->mc.vram_is_ddr + 1))));
2066         temp_ff.full += trcd_ff.full;
2067         if (temp_ff.full < tras_ff.full)
2068                 temp_ff.full = tras_ff.full;
2069         cur_latency_mclk.full = rfixed_div(temp_ff, mclk_ff);
2070
2071         temp_ff.full = rfixed_const(cur_size);
2072         cur_latency_sclk.full = rfixed_div(temp_ff, sclk_eff_ff);
2073         /*
2074           Find the total latency for the display data.
2075         */
2076         disp_latency_overhead.full = rfixed_const(80);
2077         disp_latency_overhead.full = rfixed_div(disp_latency_overhead, sclk_ff);
2078         mc_latency_mclk.full += disp_latency_overhead.full + cur_latency_mclk.full;
2079         mc_latency_sclk.full += disp_latency_overhead.full + cur_latency_sclk.full;
2080
2081         if (mc_latency_mclk.full > mc_latency_sclk.full)
2082                 disp_latency.full = mc_latency_mclk.full;
2083         else
2084                 disp_latency.full = mc_latency_sclk.full;
2085
2086         /* setup Max GRPH_STOP_REQ default value */
2087         if (ASIC_IS_RV100(rdev))
2088                 max_stop_req = 0x5c;
2089         else
2090                 max_stop_req = 0x7c;
2091
2092         if (mode1) {
2093                 /*  CRTC1
2094                     Set GRPH_BUFFER_CNTL register using h/w defined optimal values.
2095                     GRPH_STOP_REQ <= MIN[ 0x7C, (CRTC_H_DISP + 1) * (bit depth) / 0x10 ]
2096                 */
2097                 stop_req = mode1->hdisplay * pixel_bytes1 / 16;
2098
2099                 if (stop_req > max_stop_req)
2100                         stop_req = max_stop_req;
2101
2102                 /*
2103                   Find the drain rate of the display buffer.
2104                 */
2105                 temp_ff.full = rfixed_const((16/pixel_bytes1));
2106                 disp_drain_rate.full = rfixed_div(pix_clk, temp_ff);
2107
2108                 /*
2109                   Find the critical point of the display buffer.
2110                 */
2111                 crit_point_ff.full = rfixed_mul(disp_drain_rate, disp_latency);
2112                 crit_point_ff.full += rfixed_const_half(0);
2113
2114                 critical_point = rfixed_trunc(crit_point_ff);
2115
2116                 if (rdev->disp_priority == 2) {
2117                         critical_point = 0;
2118                 }
2119
2120                 /*
2121                   The critical point should never be above max_stop_req-4.  Setting
2122                   GRPH_CRITICAL_CNTL = 0 will thus force high priority all the time.
2123                 */
2124                 if (max_stop_req - critical_point < 4)
2125                         critical_point = 0;
2126
2127                 if (critical_point == 0 && mode2 && rdev->family == CHIP_R300) {
2128                         /* some R300 cards have problem with this set to 0, when CRTC2 is enabled.*/
2129                         critical_point = 0x10;
2130                 }
2131
2132                 temp = RREG32(RADEON_GRPH_BUFFER_CNTL);
2133                 temp &= ~(RADEON_GRPH_STOP_REQ_MASK);
2134                 temp |= (stop_req << RADEON_GRPH_STOP_REQ_SHIFT);
2135                 temp &= ~(RADEON_GRPH_START_REQ_MASK);
2136                 if ((rdev->family == CHIP_R350) &&
2137                     (stop_req > 0x15)) {
2138                         stop_req -= 0x10;
2139                 }
2140                 temp |= (stop_req << RADEON_GRPH_START_REQ_SHIFT);
2141                 temp |= RADEON_GRPH_BUFFER_SIZE;
2142                 temp &= ~(RADEON_GRPH_CRITICAL_CNTL   |
2143                           RADEON_GRPH_CRITICAL_AT_SOF |
2144                           RADEON_GRPH_STOP_CNTL);
2145                 /*
2146                   Write the result into the register.
2147                 */
2148                 WREG32(RADEON_GRPH_BUFFER_CNTL, ((temp & ~RADEON_GRPH_CRITICAL_POINT_MASK) |
2149                                                        (critical_point << RADEON_GRPH_CRITICAL_POINT_SHIFT)));
2150
2151 #if 0
2152                 if ((rdev->family == CHIP_RS400) ||
2153                     (rdev->family == CHIP_RS480)) {
2154                         /* attempt to program RS400 disp regs correctly ??? */
2155                         temp = RREG32(RS400_DISP1_REG_CNTL);
2156                         temp &= ~(RS400_DISP1_START_REQ_LEVEL_MASK |
2157                                   RS400_DISP1_STOP_REQ_LEVEL_MASK);
2158                         WREG32(RS400_DISP1_REQ_CNTL1, (temp |
2159                                                        (critical_point << RS400_DISP1_START_REQ_LEVEL_SHIFT) |
2160                                                        (critical_point << RS400_DISP1_STOP_REQ_LEVEL_SHIFT)));
2161                         temp = RREG32(RS400_DMIF_MEM_CNTL1);
2162                         temp &= ~(RS400_DISP1_CRITICAL_POINT_START_MASK |
2163                                   RS400_DISP1_CRITICAL_POINT_STOP_MASK);
2164                         WREG32(RS400_DMIF_MEM_CNTL1, (temp |
2165                                                       (critical_point << RS400_DISP1_CRITICAL_POINT_START_SHIFT) |
2166                                                       (critical_point << RS400_DISP1_CRITICAL_POINT_STOP_SHIFT)));
2167                 }
2168 #endif
2169
2170                 DRM_DEBUG("GRPH_BUFFER_CNTL from to %x\n",
2171                           /*      (unsigned int)info->SavedReg->grph_buffer_cntl, */
2172                           (unsigned int)RREG32(RADEON_GRPH_BUFFER_CNTL));
2173         }
2174
2175         if (mode2) {
2176                 u32 grph2_cntl;
2177                 stop_req = mode2->hdisplay * pixel_bytes2 / 16;
2178
2179                 if (stop_req > max_stop_req)
2180                         stop_req = max_stop_req;
2181
2182                 /*
2183                   Find the drain rate of the display buffer.
2184                 */
2185                 temp_ff.full = rfixed_const((16/pixel_bytes2));
2186                 disp_drain_rate2.full = rfixed_div(pix_clk2, temp_ff);
2187
2188                 grph2_cntl = RREG32(RADEON_GRPH2_BUFFER_CNTL);
2189                 grph2_cntl &= ~(RADEON_GRPH_STOP_REQ_MASK);
2190                 grph2_cntl |= (stop_req << RADEON_GRPH_STOP_REQ_SHIFT);
2191                 grph2_cntl &= ~(RADEON_GRPH_START_REQ_MASK);
2192                 if ((rdev->family == CHIP_R350) &&
2193                     (stop_req > 0x15)) {
2194                         stop_req -= 0x10;
2195                 }
2196                 grph2_cntl |= (stop_req << RADEON_GRPH_START_REQ_SHIFT);
2197                 grph2_cntl |= RADEON_GRPH_BUFFER_SIZE;
2198                 grph2_cntl &= ~(RADEON_GRPH_CRITICAL_CNTL   |
2199                           RADEON_GRPH_CRITICAL_AT_SOF |
2200                           RADEON_GRPH_STOP_CNTL);
2201
2202                 if ((rdev->family == CHIP_RS100) ||
2203                     (rdev->family == CHIP_RS200))
2204                         critical_point2 = 0;
2205                 else {
2206                         temp = (rdev->mc.vram_width * rdev->mc.vram_is_ddr + 1)/128;
2207                         temp_ff.full = rfixed_const(temp);
2208                         temp_ff.full = rfixed_mul(mclk_ff, temp_ff);
2209                         if (sclk_ff.full < temp_ff.full)
2210                                 temp_ff.full = sclk_ff.full;
2211
2212                         read_return_rate.full = temp_ff.full;
2213
2214                         if (mode1) {
2215                                 temp_ff.full = read_return_rate.full - disp_drain_rate.full;
2216                                 time_disp1_drop_priority.full = rfixed_div(crit_point_ff, temp_ff);
2217                         } else {
2218                                 time_disp1_drop_priority.full = 0;
2219                         }
2220                         crit_point_ff.full = disp_latency.full + time_disp1_drop_priority.full + disp_latency.full;
2221                         crit_point_ff.full = rfixed_mul(crit_point_ff, disp_drain_rate2);
2222                         crit_point_ff.full += rfixed_const_half(0);
2223
2224                         critical_point2 = rfixed_trunc(crit_point_ff);
2225
2226                         if (rdev->disp_priority == 2) {
2227                                 critical_point2 = 0;
2228                         }
2229
2230                         if (max_stop_req - critical_point2 < 4)
2231                                 critical_point2 = 0;
2232
2233                 }
2234
2235                 if (critical_point2 == 0 && rdev->family == CHIP_R300) {
2236                         /* some R300 cards have problem with this set to 0 */
2237                         critical_point2 = 0x10;
2238                 }
2239
2240                 WREG32(RADEON_GRPH2_BUFFER_CNTL, ((grph2_cntl & ~RADEON_GRPH_CRITICAL_POINT_MASK) |
2241                                                   (critical_point2 << RADEON_GRPH_CRITICAL_POINT_SHIFT)));
2242
2243                 if ((rdev->family == CHIP_RS400) ||
2244                     (rdev->family == CHIP_RS480)) {
2245 #if 0
2246                         /* attempt to program RS400 disp2 regs correctly ??? */
2247                         temp = RREG32(RS400_DISP2_REQ_CNTL1);
2248                         temp &= ~(RS400_DISP2_START_REQ_LEVEL_MASK |
2249                                   RS400_DISP2_STOP_REQ_LEVEL_MASK);
2250                         WREG32(RS400_DISP2_REQ_CNTL1, (temp |
2251                                                        (critical_point2 << RS400_DISP1_START_REQ_LEVEL_SHIFT) |
2252                                                        (critical_point2 << RS400_DISP1_STOP_REQ_LEVEL_SHIFT)));
2253                         temp = RREG32(RS400_DISP2_REQ_CNTL2);
2254                         temp &= ~(RS400_DISP2_CRITICAL_POINT_START_MASK |
2255                                   RS400_DISP2_CRITICAL_POINT_STOP_MASK);
2256                         WREG32(RS400_DISP2_REQ_CNTL2, (temp |
2257                                                        (critical_point2 << RS400_DISP2_CRITICAL_POINT_START_SHIFT) |
2258                                                        (critical_point2 << RS400_DISP2_CRITICAL_POINT_STOP_SHIFT)));
2259 #endif
2260                         WREG32(RS400_DISP2_REQ_CNTL1, 0x105DC1CC);
2261                         WREG32(RS400_DISP2_REQ_CNTL2, 0x2749D000);
2262                         WREG32(RS400_DMIF_MEM_CNTL1,  0x29CA71DC);
2263                         WREG32(RS400_DISP1_REQ_CNTL1, 0x28FBC3AC);
2264                 }
2265
2266                 DRM_DEBUG("GRPH2_BUFFER_CNTL from to %x\n",
2267                           (unsigned int)RREG32(RADEON_GRPH2_BUFFER_CNTL));
2268         }
2269 }