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_reg.h"
  33 #include "radeon.h"
  34 #include "radeon_asic.h"
  35 #include "r100d.h"
  36 #include "rs100d.h"
  37 #include "rv200d.h"
  38 #include "rv250d.h"
  39
  40 #include <linux/firmware.h>
  41 #include <linux/platform_device.h>
  42
  43 #include "r100_reg_safe.h"
  44 #include "rn50_reg_safe.h"
  45
  46 /* Firmware Names */
  47 #define FIRMWARE_R100           "radeon/R100_cp.bin"
  48 #define FIRMWARE_R200           "radeon/R200_cp.bin"
  49 #define FIRMWARE_R300           "radeon/R300_cp.bin"
  50 #define FIRMWARE_R420           "radeon/R420_cp.bin"
  51 #define FIRMWARE_RS690          "radeon/RS690_cp.bin"
  52 #define FIRMWARE_RS600          "radeon/RS600_cp.bin"
  53 #define FIRMWARE_R520           "radeon/R520_cp.bin"
  54
  55 MODULE_FIRMWARE(FIRMWARE_R100);
  56 MODULE_FIRMWARE(FIRMWARE_R200);
  57 MODULE_FIRMWARE(FIRMWARE_R300);
  58 MODULE_FIRMWARE(FIRMWARE_R420);
  59 MODULE_FIRMWARE(FIRMWARE_RS690);
  60 MODULE_FIRMWARE(FIRMWARE_RS600);
  61 MODULE_FIRMWARE(FIRMWARE_R520);
  62
  63 #include "r100_track.h"
  64
  65 /* This files gather functions specifics to:
  66  * r100,rv100,rs100,rv200,rs200,r200,rv250,rs300,rv280
  67  */
  68
  69 /* hpd for digital panel detect/disconnect */
  70 bool r100_hpd_sense(struct radeon_device *rdev, enum radeon_hpd_id hpd)
  71 {
  72         bool connected = false;
  73
  74         switch (hpd) {
  75         case RADEON_HPD_1:
  76                 if (RREG32(RADEON_FP_GEN_CNTL) & RADEON_FP_DETECT_SENSE)
  77                         connected = true;
  78                 break;
  79         case RADEON_HPD_2:
  80                 if (RREG32(RADEON_FP2_GEN_CNTL) & RADEON_FP2_DETECT_SENSE)
  81                         connected = true;
  82                 break;
  83         default:
  84                 break;
  85         }
  86         return connected;
  87 }
  88
  89 void r100_hpd_set_polarity(struct radeon_device *rdev,
  90                            enum radeon_hpd_id hpd)
  91 {
  92         u32 tmp;
  93         bool connected = r100_hpd_sense(rdev, hpd);
  94
  95         switch (hpd) {
  96         case RADEON_HPD_1:
  97                 tmp = RREG32(RADEON_FP_GEN_CNTL);
  98                 if (connected)
  99                         tmp &= ~RADEON_FP_DETECT_INT_POL;
 100                 else
 101                         tmp |= RADEON_FP_DETECT_INT_POL;
 102                 WREG32(RADEON_FP_GEN_CNTL, tmp);
 103                 break;
 104         case RADEON_HPD_2:
 105                 tmp = RREG32(RADEON_FP2_GEN_CNTL);
 106                 if (connected)
 107                         tmp &= ~RADEON_FP2_DETECT_INT_POL;
 108                 else
 109                         tmp |= RADEON_FP2_DETECT_INT_POL;
 110                 WREG32(RADEON_FP2_GEN_CNTL, tmp);
 111                 break;
 112         default:
 113                 break;
 114         }
 115 }
 116
 117 void r100_hpd_init(struct radeon_device *rdev)
 118 {
 119         struct drm_device *dev = rdev->ddev;
 120         struct drm_connector *connector;
 121
 122         list_for_each_entry(connector, &dev->mode_config.connector_list, head) {
 123                 struct radeon_connector *radeon_connector = to_radeon_connector(connector);
 124                 switch (radeon_connector->hpd.hpd) {
 125                 case RADEON_HPD_1:
 126                         rdev->irq.hpd[0] = true;
 127                         break;
 128                 case RADEON_HPD_2:
 129                         rdev->irq.hpd[1] = true;
 130                         break;
 131                 default:
 132                         break;
 133                 }
 134         }
 135         if (rdev->irq.installed)
 136                 r100_irq_set(rdev);
 137 }
 138
 139 void r100_hpd_fini(struct radeon_device *rdev)
 140 {
 141         struct drm_device *dev = rdev->ddev;
 142         struct drm_connector *connector;
 143
 144         list_for_each_entry(connector, &dev->mode_config.connector_list, head) {
 145                 struct radeon_connector *radeon_connector = to_radeon_connector(connector);
 146                 switch (radeon_connector->hpd.hpd) {
 147                 case RADEON_HPD_1:
 148                         rdev->irq.hpd[0] = false;
 149                         break;
 150                 case RADEON_HPD_2:
 151                         rdev->irq.hpd[1] = false;
 152                         break;
 153                 default:
 154                         break;
 155                 }
 156         }
 157 }
 158
 159 /*
 160  * PCI GART
 161  */
 162 void r100_pci_gart_tlb_flush(struct radeon_device *rdev)
 163 {
 164         /* TODO: can we do somethings here ? */
 165         /* It seems hw only cache one entry so we should discard this
 166          * entry otherwise if first GPU GART read hit this entry it
 167          * could end up in wrong address. */
 168 }
 169
 170 int r100_pci_gart_init(struct radeon_device *rdev)
 171 {
 172         int r;
 173
 174         if (rdev->gart.table.ram.ptr) {
 175                 WARN(1, "R100 PCI GART already initialized.\n");
 176                 return 0;
 177         }
 178         /* Initialize common gart structure */
 179         r = radeon_gart_init(rdev);
 180         if (r)
 181                 return r;
 182         rdev->gart.table_size = rdev->gart.num_gpu_pages * 4;
 183         rdev->asic->gart_tlb_flush = &r100_pci_gart_tlb_flush;
 184         rdev->asic->gart_set_page = &r100_pci_gart_set_page;
 185         return radeon_gart_table_ram_alloc(rdev);
 186 }
 187
 188 /* required on r1xx, r2xx, r300, r(v)350, r420/r481, rs400/rs480 */
 189 void r100_enable_bm(struct radeon_device *rdev)
 190 {
 191         uint32_t tmp;
 192         /* Enable bus mastering */
 193         tmp = RREG32(RADEON_BUS_CNTL) & ~RADEON_BUS_MASTER_DIS;
 194         WREG32(RADEON_BUS_CNTL, tmp);
 195 }
 196
 197 int r100_pci_gart_enable(struct radeon_device *rdev)
 198 {
 199         uint32_t tmp;
 200
 201         radeon_gart_restore(rdev);
 202         /* discard memory request outside of configured range */
 203         tmp = RREG32(RADEON_AIC_CNTL) | RADEON_DIS_OUT_OF_PCI_GART_ACCESS;
 204         WREG32(RADEON_AIC_CNTL, tmp);
 205         /* set address range for PCI address translate */
 206         WREG32(RADEON_AIC_LO_ADDR, rdev->mc.gtt_start);
 207         WREG32(RADEON_AIC_HI_ADDR, rdev->mc.gtt_end);
 208         /* set PCI GART page-table base address */
 209         WREG32(RADEON_AIC_PT_BASE, rdev->gart.table_addr);
 210         tmp = RREG32(RADEON_AIC_CNTL) | RADEON_PCIGART_TRANSLATE_EN;
 211         WREG32(RADEON_AIC_CNTL, tmp);
 212         r100_pci_gart_tlb_flush(rdev);
 213         rdev->gart.ready = true;
 214         return 0;
 215 }
 216
 217 void r100_pci_gart_disable(struct radeon_device *rdev)
 218 {
 219         uint32_t tmp;
 220
 221         /* discard memory request outside of configured range */
 222         tmp = RREG32(RADEON_AIC_CNTL) | RADEON_DIS_OUT_OF_PCI_GART_ACCESS;
 223         WREG32(RADEON_AIC_CNTL, tmp & ~RADEON_PCIGART_TRANSLATE_EN);
 224         WREG32(RADEON_AIC_LO_ADDR, 0);
 225         WREG32(RADEON_AIC_HI_ADDR, 0);
 226 }
 227
 228 int r100_pci_gart_set_page(struct radeon_device *rdev, int i, uint64_t addr)
 229 {
 230         if (i < 0 || i > rdev->gart.num_gpu_pages) {
 231                 return -EINVAL;
 232         }
 233         rdev->gart.table.ram.ptr[i] = cpu_to_le32(lower_32_bits(addr));
 234         return 0;
 235 }
 236
 237 void r100_pci_gart_fini(struct radeon_device *rdev)
 238 {
 239         radeon_gart_fini(rdev);
 240         r100_pci_gart_disable(rdev);
 241         radeon_gart_table_ram_free(rdev);
 242 }
 243
 244 int r100_irq_set(struct radeon_device *rdev)
 245 {
 246         uint32_t tmp = 0;
 247
 248         if (!rdev->irq.installed) {
 249                 WARN(1, "Can't enable IRQ/MSI because no handler is installed.\n");
 250                 WREG32(R_000040_GEN_INT_CNTL, 0);
 251                 return -EINVAL;
 252         }
 253         if (rdev->irq.sw_int) {
 254                 tmp |= RADEON_SW_INT_ENABLE;
 255         }
 256         if (rdev->irq.crtc_vblank_int[0]) {
 257                 tmp |= RADEON_CRTC_VBLANK_MASK;
 258         }
 259         if (rdev->irq.crtc_vblank_int[1]) {
 260                 tmp |= RADEON_CRTC2_VBLANK_MASK;
 261         }
 262         if (rdev->irq.hpd[0]) {
 263                 tmp |= RADEON_FP_DETECT_MASK;
 264         }
 265         if (rdev->irq.hpd[1]) {
 266                 tmp |= RADEON_FP2_DETECT_MASK;
 267         }
 268         WREG32(RADEON_GEN_INT_CNTL, tmp);
 269         return 0;
 270 }
 271
 272 void r100_irq_disable(struct radeon_device *rdev)
 273 {
 274         u32 tmp;
 275
 276         WREG32(R_000040_GEN_INT_CNTL, 0);
 277         /* Wait and acknowledge irq */
 278         mdelay(1);
 279         tmp = RREG32(R_000044_GEN_INT_STATUS);
 280         WREG32(R_000044_GEN_INT_STATUS, tmp);
 281 }
 282
 283 static inline uint32_t r100_irq_ack(struct radeon_device *rdev)
 284 {
 285         uint32_t irqs = RREG32(RADEON_GEN_INT_STATUS);
 286         uint32_t irq_mask = RADEON_SW_INT_TEST |
 287                 RADEON_CRTC_VBLANK_STAT | RADEON_CRTC2_VBLANK_STAT |
 288                 RADEON_FP_DETECT_STAT | RADEON_FP2_DETECT_STAT;
 289
 290         if (irqs) {
 291                 WREG32(RADEON_GEN_INT_STATUS, irqs);
 292         }
 293         return irqs & irq_mask;
 294 }
 295
 296 int r100_irq_process(struct radeon_device *rdev)
 297 {
 298         uint32_t status, msi_rearm;
 299         bool queue_hotplug = false;
 300
 301         status = r100_irq_ack(rdev);
 302         if (!status) {
 303                 return IRQ_NONE;
 304         }
 305         if (rdev->shutdown) {
 306                 return IRQ_NONE;
 307         }
 308         while (status) {
 309                 /* SW interrupt */
 310                 if (status & RADEON_SW_INT_TEST) {
 311                         radeon_fence_process(rdev);
 312                 }
 313                 /* Vertical blank interrupts */
 314                 if (status & RADEON_CRTC_VBLANK_STAT) {
 315                         drm_handle_vblank(rdev->ddev, 0);
 316                         rdev->pm.vblank_sync = true;
 317                         wake_up(&rdev->irq.vblank_queue);
 318                 }
 319                 if (status & RADEON_CRTC2_VBLANK_STAT) {
 320                         drm_handle_vblank(rdev->ddev, 1);
 321                         rdev->pm.vblank_sync = true;
 322                         wake_up(&rdev->irq.vblank_queue);
 323                 }
 324                 if (status & RADEON_FP_DETECT_STAT) {
 325                         queue_hotplug = true;
 326                         DRM_DEBUG("HPD1\n");
 327                 }
 328                 if (status & RADEON_FP2_DETECT_STAT) {
 329                         queue_hotplug = true;
 330                         DRM_DEBUG("HPD2\n");
 331                 }
 332                 status = r100_irq_ack(rdev);
 333         }
 334         if (queue_hotplug)
 335                 queue_work(rdev->wq, &rdev->hotplug_work);
 336         if (rdev->msi_enabled) {
 337                 switch (rdev->family) {
 338                 case CHIP_RS400:
 339                 case CHIP_RS480:
 340                         msi_rearm = RREG32(RADEON_AIC_CNTL) & ~RS400_MSI_REARM;
 341                         WREG32(RADEON_AIC_CNTL, msi_rearm);
 342                         WREG32(RADEON_AIC_CNTL, msi_rearm | RS400_MSI_REARM);
 343                         break;
 344                 default:
 345                         msi_rearm = RREG32(RADEON_MSI_REARM_EN) & ~RV370_MSI_REARM_EN;
 346                         WREG32(RADEON_MSI_REARM_EN, msi_rearm);
 347                         WREG32(RADEON_MSI_REARM_EN, msi_rearm | RV370_MSI_REARM_EN);
 348                         break;
 349                 }
 350         }
 351         return IRQ_HANDLED;
 352 }
 353
 354 u32 r100_get_vblank_counter(struct radeon_device *rdev, int crtc)
 355 {
 356         if (crtc == 0)
 357                 return RREG32(RADEON_CRTC_CRNT_FRAME);
 358         else
 359                 return RREG32(RADEON_CRTC2_CRNT_FRAME);
 360 }
 361
 362 /* Who ever call radeon_fence_emit should call ring_lock and ask
 363  * for enough space (today caller are ib schedule and buffer move) */
 364 void r100_fence_ring_emit(struct radeon_device *rdev,
 365                           struct radeon_fence *fence)
 366 {
 367         /* We have to make sure that caches are flushed before
 368          * CPU might read something from VRAM. */
 369         radeon_ring_write(rdev, PACKET0(RADEON_RB3D_DSTCACHE_CTLSTAT, 0));
 370         radeon_ring_write(rdev, RADEON_RB3D_DC_FLUSH_ALL);
 371         radeon_ring_write(rdev, PACKET0(RADEON_RB3D_ZCACHE_CTLSTAT, 0));
 372         radeon_ring_write(rdev, RADEON_RB3D_ZC_FLUSH_ALL);
 373         /* Wait until IDLE & CLEAN */
 374         radeon_ring_write(rdev, PACKET0(RADEON_WAIT_UNTIL, 0));
 375         radeon_ring_write(rdev, RADEON_WAIT_2D_IDLECLEAN | RADEON_WAIT_3D_IDLECLEAN);
 376         radeon_ring_write(rdev, PACKET0(RADEON_HOST_PATH_CNTL, 0));
 377         radeon_ring_write(rdev, rdev->config.r100.hdp_cntl |
 378                                 RADEON_HDP_READ_BUFFER_INVALIDATE);
 379         radeon_ring_write(rdev, PACKET0(RADEON_HOST_PATH_CNTL, 0));
 380         radeon_ring_write(rdev, rdev->config.r100.hdp_cntl);
 381         /* Emit fence sequence & fire IRQ */
 382         radeon_ring_write(rdev, PACKET0(rdev->fence_drv.scratch_reg, 0));
 383         radeon_ring_write(rdev, fence->seq);
 384         radeon_ring_write(rdev, PACKET0(RADEON_GEN_INT_STATUS, 0));
 385         radeon_ring_write(rdev, RADEON_SW_INT_FIRE);
 386 }
 387
 388 int r100_wb_init(struct radeon_device *rdev)
 389 {
 390         int r;
 391
 392         if (rdev->wb.wb_obj == NULL) {
 393                 r = radeon_bo_create(rdev, NULL, RADEON_GPU_PAGE_SIZE, true,
 394                                         RADEON_GEM_DOMAIN_GTT,
 395                                         &rdev->wb.wb_obj);
 396                 if (r) {
 397                         dev_err(rdev->dev, "(%d) create WB buffer failed\n", r);
 398                         return r;
 399                 }
 400                 r = radeon_bo_reserve(rdev->wb.wb_obj, false);
 401                 if (unlikely(r != 0))
 402                         return r;
 403                 r = radeon_bo_pin(rdev->wb.wb_obj, RADEON_GEM_DOMAIN_GTT,
 404                                         &rdev->wb.gpu_addr);
 405                 if (r) {
 406                         dev_err(rdev->dev, "(%d) pin WB buffer failed\n", r);
 407                         radeon_bo_unreserve(rdev->wb.wb_obj);
 408                         return r;
 409                 }
 410                 r = radeon_bo_kmap(rdev->wb.wb_obj, (void **)&rdev->wb.wb);
 411                 radeon_bo_unreserve(rdev->wb.wb_obj);
 412                 if (r) {
 413                         dev_err(rdev->dev, "(%d) map WB buffer failed\n", r);
 414                         return r;
 415                 }
 416         }
 417         WREG32(R_000774_SCRATCH_ADDR, rdev->wb.gpu_addr);
 418         WREG32(R_00070C_CP_RB_RPTR_ADDR,
 419                 S_00070C_RB_RPTR_ADDR((rdev->wb.gpu_addr + 1024) >> 2));
 420         WREG32(R_000770_SCRATCH_UMSK, 0xff);
 421         return 0;
 422 }
 423
 424 void r100_wb_disable(struct radeon_device *rdev)
 425 {
 426         WREG32(R_000770_SCRATCH_UMSK, 0);
 427 }
 428
 429 void r100_wb_fini(struct radeon_device *rdev)
 430 {
 431         int r;
 432
 433         r100_wb_disable(rdev);
 434         if (rdev->wb.wb_obj) {
 435                 r = radeon_bo_reserve(rdev->wb.wb_obj, false);
 436                 if (unlikely(r != 0)) {
 437                         dev_err(rdev->dev, "(%d) can't finish WB\n", r);
 438                         return;
 439                 }
 440                 radeon_bo_kunmap(rdev->wb.wb_obj);
 441                 radeon_bo_unpin(rdev->wb.wb_obj);
 442                 radeon_bo_unreserve(rdev->wb.wb_obj);
 443                 radeon_bo_unref(&rdev->wb.wb_obj);
 444                 rdev->wb.wb = NULL;
 445                 rdev->wb.wb_obj = NULL;
 446         }
 447 }
 448
 449 int r100_copy_blit(struct radeon_device *rdev,
 450                    uint64_t src_offset,
 451                    uint64_t dst_offset,
 452                    unsigned num_pages,
 453                    struct radeon_fence *fence)
 454 {
 455         uint32_t cur_pages;
 456         uint32_t stride_bytes = PAGE_SIZE;
 457         uint32_t pitch;
 458         uint32_t stride_pixels;
 459         unsigned ndw;
 460         int num_loops;
 461         int r = 0;
 462
 463         /* radeon limited to 16k stride */
 464         stride_bytes &= 0x3fff;
 465         /* radeon pitch is /64 */
 466         pitch = stride_bytes / 64;
 467         stride_pixels = stride_bytes / 4;
 468         num_loops = DIV_ROUND_UP(num_pages, 8191);
 469
 470         /* Ask for enough room for blit + flush + fence */
 471         ndw = 64 + (10 * num_loops);
 472         r = radeon_ring_lock(rdev, ndw);
 473         if (r) {
 474                 DRM_ERROR("radeon: moving bo (%d) asking for %u dw.\n", r, ndw);
 475                 return -EINVAL;
 476         }
 477         while (num_pages > 0) {
 478                 cur_pages = num_pages;
 479                 if (cur_pages > 8191) {
 480                         cur_pages = 8191;
 481                 }
 482                 num_pages -= cur_pages;
 483
 484                 /* pages are in Y direction - height
 485                    page width in X direction - width */
 486                 radeon_ring_write(rdev, PACKET3(PACKET3_BITBLT_MULTI, 8));
 487                 radeon_ring_write(rdev,
 488                                   RADEON_GMC_SRC_PITCH_OFFSET_CNTL |
 489                                   RADEON_GMC_DST_PITCH_OFFSET_CNTL |
 490                                   RADEON_GMC_SRC_CLIPPING |
 491                                   RADEON_GMC_DST_CLIPPING |
 492                                   RADEON_GMC_BRUSH_NONE |
 493                                   (RADEON_COLOR_FORMAT_ARGB8888 << 8) |
 494                                   RADEON_GMC_SRC_DATATYPE_COLOR |
 495                                   RADEON_ROP3_S |
 496                                   RADEON_DP_SRC_SOURCE_MEMORY |
 497                                   RADEON_GMC_CLR_CMP_CNTL_DIS |
 498                                   RADEON_GMC_WR_MSK_DIS);
 499                 radeon_ring_write(rdev, (pitch << 22) | (src_offset >> 10));
 500                 radeon_ring_write(rdev, (pitch << 22) | (dst_offset >> 10));
 501                 radeon_ring_write(rdev, (0x1fff) | (0x1fff << 16));
 502                 radeon_ring_write(rdev, 0);
 503                 radeon_ring_write(rdev, (0x1fff) | (0x1fff << 16));
 504                 radeon_ring_write(rdev, num_pages);
 505                 radeon_ring_write(rdev, num_pages);
 506                 radeon_ring_write(rdev, cur_pages | (stride_pixels << 16));
 507         }
 508         radeon_ring_write(rdev, PACKET0(RADEON_DSTCACHE_CTLSTAT, 0));
 509         radeon_ring_write(rdev, RADEON_RB2D_DC_FLUSH_ALL);
 510         radeon_ring_write(rdev, PACKET0(RADEON_WAIT_UNTIL, 0));
 511         radeon_ring_write(rdev,
 512                           RADEON_WAIT_2D_IDLECLEAN |
 513                           RADEON_WAIT_HOST_IDLECLEAN |
 514                           RADEON_WAIT_DMA_GUI_IDLE);
 515         if (fence) {
 516                 r = radeon_fence_emit(rdev, fence);
 517         }
 518         radeon_ring_unlock_commit(rdev);
 519         return r;
 520 }
 521
 522 static int r100_cp_wait_for_idle(struct radeon_device *rdev)
 523 {
 524         unsigned i;
 525         u32 tmp;
 526
 527         for (i = 0; i < rdev->usec_timeout; i++) {
 528                 tmp = RREG32(R_000E40_RBBM_STATUS);
 529                 if (!G_000E40_CP_CMDSTRM_BUSY(tmp)) {
 530                         return 0;
 531                 }
 532                 udelay(1);
 533         }
 534         return -1;
 535 }
 536
 537 void r100_ring_start(struct radeon_device *rdev)
 538 {
 539         int r;
 540
 541         r = radeon_ring_lock(rdev, 2);
 542         if (r) {
 543                 return;
 544         }
 545         radeon_ring_write(rdev, PACKET0(RADEON_ISYNC_CNTL, 0));
 546         radeon_ring_write(rdev,
 547                           RADEON_ISYNC_ANY2D_IDLE3D |
 548                           RADEON_ISYNC_ANY3D_IDLE2D |
 549                           RADEON_ISYNC_WAIT_IDLEGUI |
 550                           RADEON_ISYNC_CPSCRATCH_IDLEGUI);
 551         radeon_ring_unlock_commit(rdev);
 552 }
 553
 554
 555 /* Load the microcode for the CP */
 556 static int r100_cp_init_microcode(struct radeon_device *rdev)
 557 {
 558         struct platform_device *pdev;
 559         const char *fw_name = NULL;
 560         int err;
 561
 562         DRM_DEBUG("\n");
 563
 564         pdev = platform_device_register_simple("radeon_cp", 0, NULL, 0);
 565         err = IS_ERR(pdev);
 566         if (err) {
 567                 printk(KERN_ERR "radeon_cp: Failed to register firmware\n");
 568                 return -EINVAL;
 569         }
 570         if ((rdev->family == CHIP_R100) || (rdev->family == CHIP_RV100) ||
 571             (rdev->family == CHIP_RV200) || (rdev->family == CHIP_RS100) ||
 572             (rdev->family == CHIP_RS200)) {
 573                 DRM_INFO("Loading R100 Microcode\n");
 574                 fw_name = FIRMWARE_R100;
 575         } else if ((rdev->family == CHIP_R200) ||
 576                    (rdev->family == CHIP_RV250) ||
 577                    (rdev->family == CHIP_RV280) ||
 578                    (rdev->family == CHIP_RS300)) {
 579                 DRM_INFO("Loading R200 Microcode\n");
 580                 fw_name = FIRMWARE_R200;
 581         } else if ((rdev->family == CHIP_R300) ||
 582                    (rdev->family == CHIP_R350) ||
 583                    (rdev->family == CHIP_RV350) ||
 584                    (rdev->family == CHIP_RV380) ||
 585                    (rdev->family == CHIP_RS400) ||
 586                    (rdev->family == CHIP_RS480)) {
 587                 DRM_INFO("Loading R300 Microcode\n");
 588                 fw_name = FIRMWARE_R300;
 589         } else if ((rdev->family == CHIP_R420) ||
 590                    (rdev->family == CHIP_R423) ||
 591                    (rdev->family == CHIP_RV410)) {
 592                 DRM_INFO("Loading R400 Microcode\n");
 593                 fw_name = FIRMWARE_R420;
 594         } else if ((rdev->family == CHIP_RS690) ||
 595                    (rdev->family == CHIP_RS740)) {
 596                 DRM_INFO("Loading RS690/RS740 Microcode\n");
 597                 fw_name = FIRMWARE_RS690;
 598         } else if (rdev->family == CHIP_RS600) {
 599                 DRM_INFO("Loading RS600 Microcode\n");
 600                 fw_name = FIRMWARE_RS600;
 601         } else if ((rdev->family == CHIP_RV515) ||
 602                    (rdev->family == CHIP_R520) ||
 603                    (rdev->family == CHIP_RV530) ||
 604                    (rdev->family == CHIP_R580) ||
 605                    (rdev->family == CHIP_RV560) ||
 606                    (rdev->family == CHIP_RV570)) {
 607                 DRM_INFO("Loading R500 Microcode\n");
 608                 fw_name = FIRMWARE_R520;
 609         }
 610
 611         err = request_firmware(&rdev->me_fw, fw_name, &pdev->dev);
 612         platform_device_unregister(pdev);
 613         if (err) {
 614                 printk(KERN_ERR "radeon_cp: Failed to load firmware \"%s\"\n",
 615                        fw_name);
 616         } else if (rdev->me_fw->size % 8) {
 617                 printk(KERN_ERR
 618                        "radeon_cp: Bogus length %zu in firmware \"%s\"\n",
 619                        rdev->me_fw->size, fw_name);
 620                 err = -EINVAL;
 621                 release_firmware(rdev->me_fw);
 622                 rdev->me_fw = NULL;
 623         }
 624         return err;
 625 }
 626
 627 static void r100_cp_load_microcode(struct radeon_device *rdev)
 628 {
 629         const __be32 *fw_data;
 630         int i, size;
 631
 632         if (r100_gui_wait_for_idle(rdev)) {
 633                 printk(KERN_WARNING "Failed to wait GUI idle while "
 634                        "programming pipes. Bad things might happen.\n");
 635         }
 636
 637         if (rdev->me_fw) {
 638                 size = rdev->me_fw->size / 4;
 639                 fw_data = (const __be32 *)&rdev->me_fw->data[0];
 640                 WREG32(RADEON_CP_ME_RAM_ADDR, 0);
 641                 for (i = 0; i < size; i += 2) {
 642                         WREG32(RADEON_CP_ME_RAM_DATAH,
 643                                be32_to_cpup(&fw_data[i]));
 644                         WREG32(RADEON_CP_ME_RAM_DATAL,
 645                                be32_to_cpup(&fw_data[i + 1]));
 646                 }
 647         }
 648 }
 649
 650 int r100_cp_init(struct radeon_device *rdev, unsigned ring_size)
 651 {
 652         unsigned rb_bufsz;
 653         unsigned rb_blksz;
 654         unsigned max_fetch;
 655         unsigned pre_write_timer;
 656         unsigned pre_write_limit;
 657         unsigned indirect2_start;
 658         unsigned indirect1_start;
 659         uint32_t tmp;
 660         int r;
 661
 662         if (r100_debugfs_cp_init(rdev)) {
 663                 DRM_ERROR("Failed to register debugfs file for CP !\n");
 664         }
 665         /* Reset CP */
 666         tmp = RREG32(RADEON_CP_CSQ_STAT);
 667         if ((tmp & (1 << 31))) {
 668                 DRM_INFO("radeon: cp busy (0x%08X) resetting\n", tmp);
 669                 WREG32(RADEON_CP_CSQ_MODE, 0);
 670                 WREG32(RADEON_CP_CSQ_CNTL, 0);
 671                 WREG32(RADEON_RBBM_SOFT_RESET, RADEON_SOFT_RESET_CP);
 672                 tmp = RREG32(RADEON_RBBM_SOFT_RESET);
 673                 mdelay(2);
 674                 WREG32(RADEON_RBBM_SOFT_RESET, 0);
 675                 tmp = RREG32(RADEON_RBBM_SOFT_RESET);
 676                 mdelay(2);
 677                 tmp = RREG32(RADEON_CP_CSQ_STAT);
 678                 if ((tmp & (1 << 31))) {
 679                         DRM_INFO("radeon: cp reset failed (0x%08X)\n", tmp);
 680                 }
 681         } else {
 682                 DRM_INFO("radeon: cp idle (0x%08X)\n", tmp);
 683         }
 684
 685         if (!rdev->me_fw) {
 686                 r = r100_cp_init_microcode(rdev);
 687                 if (r) {
 688                         DRM_ERROR("Failed to load firmware!\n");
 689                         return r;
 690                 }
 691         }
 692
 693         /* Align ring size */
 694         rb_bufsz = drm_order(ring_size / 8);
 695         ring_size = (1 << (rb_bufsz + 1)) * 4;
 696         r100_cp_load_microcode(rdev);
 697         r = radeon_ring_init(rdev, ring_size);
 698         if (r) {
 699                 return r;
 700         }
 701         /* Each time the cp read 1024 bytes (16 dword/quadword) update
 702          * the rptr copy in system ram */
 703         rb_blksz = 9;
 704         /* cp will read 128bytes at a time (4 dwords) */
 705         max_fetch = 1;
 706         rdev->cp.align_mask = 16 - 1;
 707         /* Write to CP_RB_WPTR will be delayed for pre_write_timer clocks */
 708         pre_write_timer = 64;
 709         /* Force CP_RB_WPTR write if written more than one time before the
 710          * delay expire
 711          */
 712         pre_write_limit = 0;
 713         /* Setup the cp cache like this (cache size is 96 dwords) :
 714          *      RING            0  to 15
 715          *      INDIRECT1       16 to 79
 716          *      INDIRECT2       80 to 95
 717          * So ring cache size is 16dwords (> (2 * max_fetch = 2 * 4dwords))
 718          *    indirect1 cache size is 64dwords (> (2 * max_fetch = 2 * 4dwords))
 719          *    indirect2 cache size is 16dwords (> (2 * max_fetch = 2 * 4dwords))
 720          * Idea being that most of the gpu cmd will be through indirect1 buffer
 721          * so it gets the bigger cache.
 722          */
 723         indirect2_start = 80;
 724         indirect1_start = 16;
 725         /* cp setup */
 726         WREG32(0x718, pre_write_timer | (pre_write_limit << 28));
 727         tmp = (REG_SET(RADEON_RB_BUFSZ, rb_bufsz) |
 728                REG_SET(RADEON_RB_BLKSZ, rb_blksz) |
 729                REG_SET(RADEON_MAX_FETCH, max_fetch) |
 730                RADEON_RB_NO_UPDATE);
 731 #ifdef __BIG_ENDIAN
 732         tmp |= RADEON_BUF_SWAP_32BIT;
 733 #endif
 734         WREG32(RADEON_CP_RB_CNTL, tmp);
 735
 736         /* Set ring address */
 737         DRM_INFO("radeon: ring at 0x%016lX\n", (unsigned long)rdev->cp.gpu_addr);
 738         WREG32(RADEON_CP_RB_BASE, rdev->cp.gpu_addr);
 739         /* Force read & write ptr to 0 */
 740         WREG32(RADEON_CP_RB_CNTL, tmp | RADEON_RB_RPTR_WR_ENA);
 741         WREG32(RADEON_CP_RB_RPTR_WR, 0);
 742         WREG32(RADEON_CP_RB_WPTR, 0);
 743         WREG32(RADEON_CP_RB_CNTL, tmp);
 744         udelay(10);
 745         rdev->cp.rptr = RREG32(RADEON_CP_RB_RPTR);
 746         rdev->cp.wptr = RREG32(RADEON_CP_RB_WPTR);
 747         /* protect against crazy HW on resume */
 748         rdev->cp.wptr &= rdev->cp.ptr_mask;
 749         /* Set cp mode to bus mastering & enable cp*/
 750         WREG32(RADEON_CP_CSQ_MODE,
 751                REG_SET(RADEON_INDIRECT2_START, indirect2_start) |
 752                REG_SET(RADEON_INDIRECT1_START, indirect1_start));
 753         WREG32(0x718, 0);
 754         WREG32(0x744, 0x00004D4D);
 755         WREG32(RADEON_CP_CSQ_CNTL, RADEON_CSQ_PRIBM_INDBM);
 756         radeon_ring_start(rdev);
 757         r = radeon_ring_test(rdev);
 758         if (r) {
 759                 DRM_ERROR("radeon: cp isn't working (%d).\n", r);
 760                 return r;
 761         }
 762         rdev->cp.ready = true;
 763         return 0;
 764 }
 765
 766 void r100_cp_fini(struct radeon_device *rdev)
 767 {
 768         if (r100_cp_wait_for_idle(rdev)) {
 769                 DRM_ERROR("Wait for CP idle timeout, shutting down CP.\n");
 770         }
 771         /* Disable ring */
 772         r100_cp_disable(rdev);
 773         radeon_ring_fini(rdev);
 774         DRM_INFO("radeon: cp finalized\n");
 775 }
 776
 777 void r100_cp_disable(struct radeon_device *rdev)
 778 {
 779         /* Disable ring */
 780         rdev->cp.ready = false;
 781         WREG32(RADEON_CP_CSQ_MODE, 0);
 782         WREG32(RADEON_CP_CSQ_CNTL, 0);
 783         if (r100_gui_wait_for_idle(rdev)) {
 784                 printk(KERN_WARNING "Failed to wait GUI idle while "
 785                        "programming pipes. Bad things might happen.\n");
 786         }
 787 }
 788
 789 int r100_cp_reset(struct radeon_device *rdev)
 790 {
 791         uint32_t tmp;
 792         bool reinit_cp;
 793         int i;
 794
 795         reinit_cp = rdev->cp.ready;
 796         rdev->cp.ready = false;
 797         WREG32(RADEON_CP_CSQ_MODE, 0);
 798         WREG32(RADEON_CP_CSQ_CNTL, 0);
 799         WREG32(RADEON_RBBM_SOFT_RESET, RADEON_SOFT_RESET_CP);
 800         (void)RREG32(RADEON_RBBM_SOFT_RESET);
 801         udelay(200);
 802         WREG32(RADEON_RBBM_SOFT_RESET, 0);
 803         /* Wait to prevent race in RBBM_STATUS */
 804         mdelay(1);
 805         for (i = 0; i < rdev->usec_timeout; i++) {
 806                 tmp = RREG32(RADEON_RBBM_STATUS);
 807                 if (!(tmp & (1 << 16))) {
 808                         DRM_INFO("CP reset succeed (RBBM_STATUS=0x%08X)\n",
 809                                  tmp);
 810                         if (reinit_cp) {
 811                                 return r100_cp_init(rdev, rdev->cp.ring_size);
 812                         }
 813                         return 0;
 814                 }
 815                 DRM_UDELAY(1);
 816         }
 817         tmp = RREG32(RADEON_RBBM_STATUS);
 818         DRM_ERROR("Failed to reset CP (RBBM_STATUS=0x%08X)!\n", tmp);
 819         return -1;
 820 }
 821
 822 void r100_cp_commit(struct radeon_device *rdev)
 823 {
 824         WREG32(RADEON_CP_RB_WPTR, rdev->cp.wptr);
 825         (void)RREG32(RADEON_CP_RB_WPTR);
 826 }
 827
 828
 829 /*
 830  * CS functions
 831  */
 832 int r100_cs_parse_packet0(struct radeon_cs_parser *p,
 833                           struct radeon_cs_packet *pkt,
 834                           const unsigned *auth, unsigned n,
 835                           radeon_packet0_check_t check)
 836 {
 837         unsigned reg;
 838         unsigned i, j, m;
 839         unsigned idx;
 840         int r;
 841
 842         idx = pkt->idx + 1;
 843         reg = pkt->reg;
 844         /* Check that register fall into register range
 845          * determined by the number of entry (n) in the
 846          * safe register bitmap.
 847          */
 848         if (pkt->one_reg_wr) {
 849                 if ((reg >> 7) > n) {
 850                         return -EINVAL;
 851                 }
 852         } else {
 853                 if (((reg + (pkt->count << 2)) >> 7) > n) {
 854                         return -EINVAL;
 855                 }
 856         }
 857         for (i = 0; i <= pkt->count; i++, idx++) {
 858                 j = (reg >> 7);
 859                 m = 1 << ((reg >> 2) & 31);
 860                 if (auth[j] & m) {
 861                         r = check(p, pkt, idx, reg);
 862                         if (r) {
 863                                 return r;
 864                         }
 865                 }
 866                 if (pkt->one_reg_wr) {
 867                         if (!(auth[j] & m)) {
 868                                 break;
 869                         }
 870                 } else {
 871                         reg += 4;
 872                 }
 873         }
 874         return 0;
 875 }
 876
 877 void r100_cs_dump_packet(struct radeon_cs_parser *p,
 878                          struct radeon_cs_packet *pkt)
 879 {
 880         volatile uint32_t *ib;
 881         unsigned i;
 882         unsigned idx;
 883
 884         ib = p->ib->ptr;
 885         idx = pkt->idx;
 886         for (i = 0; i <= (pkt->count + 1); i++, idx++) {
 887                 DRM_INFO("ib[%d]=0x%08X\n", idx, ib[idx]);
 888         }
 889 }
 890
 891 /**
 892  * r100_cs_packet_parse() - parse cp packet and point ib index to next packet
 893  * @parser:     parser structure holding parsing context.
 894  * @pkt:        where to store packet informations
 895  *
 896  * Assume that chunk_ib_index is properly set. Will return -EINVAL
 897  * if packet is bigger than remaining ib size. or if packets is unknown.
 898  **/
 899 int r100_cs_packet_parse(struct radeon_cs_parser *p,
 900                          struct radeon_cs_packet *pkt,
 901                          unsigned idx)
 902 {
 903         struct radeon_cs_chunk *ib_chunk = &p->chunks[p->chunk_ib_idx];
 904         uint32_t header;
 905
 906         if (idx >= ib_chunk->length_dw) {
 907                 DRM_ERROR("Can not parse packet at %d after CS end %d !\n",
 908                           idx, ib_chunk->length_dw);
 909                 return -EINVAL;
 910         }
 911         header = radeon_get_ib_value(p, idx);
 912         pkt->idx = idx;
 913         pkt->type = CP_PACKET_GET_TYPE(header);
 914         pkt->count = CP_PACKET_GET_COUNT(header);
 915         switch (pkt->type) {
 916         case PACKET_TYPE0:
 917                 pkt->reg = CP_PACKET0_GET_REG(header);
 918                 pkt->one_reg_wr = CP_PACKET0_GET_ONE_REG_WR(header);
 919                 break;
 920         case PACKET_TYPE3:
 921                 pkt->opcode = CP_PACKET3_GET_OPCODE(header);
 922                 break;
 923         case PACKET_TYPE2:
 924                 pkt->count = -1;
 925                 break;
 926         default:
 927                 DRM_ERROR("Unknown packet type %d at %d !\n", pkt->type, idx);
 928                 return -EINVAL;
 929         }
 930         if ((pkt->count + 1 + pkt->idx) >= ib_chunk->length_dw) {
 931                 DRM_ERROR("Packet (%d:%d:%d) end after CS buffer (%d) !\n",
 932                           pkt->idx, pkt->type, pkt->count, ib_chunk->length_dw);
 933                 return -EINVAL;
 934         }
 935         return 0;
 936 }
 937
 938 /**
 939  * r100_cs_packet_next_vline() - parse userspace VLINE packet
 940  * @parser:             parser structure holding parsing context.
 941  *
 942  * Userspace sends a special sequence for VLINE waits.
 943  * PACKET0 - VLINE_START_END + value
 944  * PACKET0 - WAIT_UNTIL +_value
 945  * RELOC (P3) - crtc_id in reloc.
 946  *
 947  * This function parses this and relocates the VLINE START END
 948  * and WAIT UNTIL packets to the correct crtc.
 949  * It also detects a switched off crtc and nulls out the
 950  * wait in that case.
 951  */
 952 int r100_cs_packet_parse_vline(struct radeon_cs_parser *p)
 953 {
 954         struct drm_mode_object *obj;
 955         struct drm_crtc *crtc;
 956         struct radeon_crtc *radeon_crtc;
 957         struct radeon_cs_packet p3reloc, waitreloc;
 958         int crtc_id;
 959         int r;
 960         uint32_t header, h_idx, reg;
 961         volatile uint32_t *ib;
 962
 963         ib = p->ib->ptr;
 964
 965         /* parse the wait until */
 966         r = r100_cs_packet_parse(p, &waitreloc, p->idx);
 967         if (r)
 968                 return r;
 969
 970         /* check its a wait until and only 1 count */
 971         if (waitreloc.reg != RADEON_WAIT_UNTIL ||
 972             waitreloc.count != 0) {
 973                 DRM_ERROR("vline wait had illegal wait until segment\n");
 974                 r = -EINVAL;
 975                 return r;
 976         }
 977
 978         if (radeon_get_ib_value(p, waitreloc.idx + 1) != RADEON_WAIT_CRTC_VLINE) {
 979                 DRM_ERROR("vline wait had illegal wait until\n");
 980                 r = -EINVAL;
 981                 return r;
 982         }
 983
 984         /* jump over the NOP */
 985         r = r100_cs_packet_parse(p, &p3reloc, p->idx + waitreloc.count + 2);
 986         if (r)
 987                 return r;
 988
 989         h_idx = p->idx - 2;
 990         p->idx += waitreloc.count + 2;
 991         p->idx += p3reloc.count + 2;
 992
 993         header = radeon_get_ib_value(p, h_idx);
 994         crtc_id = radeon_get_ib_value(p, h_idx + 5);
 995         reg = CP_PACKET0_GET_REG(header);
 996         mutex_lock(&p->rdev->ddev->mode_config.mutex);
 997         obj = drm_mode_object_find(p->rdev->ddev, crtc_id, DRM_MODE_OBJECT_CRTC);
 998         if (!obj) {
 999                 DRM_ERROR("cannot find crtc %d\n", crtc_id);
1000                 r = -EINVAL;
1001                 goto out;
1002         }
1003         crtc = obj_to_crtc(obj);
1004         radeon_crtc = to_radeon_crtc(crtc);
1005         crtc_id = radeon_crtc->crtc_id;
1006
1007         if (!crtc->enabled) {
1008                 /* if the CRTC isn't enabled - we need to nop out the wait until */
1009                 ib[h_idx + 2] = PACKET2(0);
1010                 ib[h_idx + 3] = PACKET2(0);
1011         } else if (crtc_id == 1) {
1012                 switch (reg) {
1013                 case AVIVO_D1MODE_VLINE_START_END:
1014                         header &= ~R300_CP_PACKET0_REG_MASK;
1015                         header |= AVIVO_D2MODE_VLINE_START_END >> 2;
1016                         break;
1017                 case RADEON_CRTC_GUI_TRIG_VLINE:
1018                         header &= ~R300_CP_PACKET0_REG_MASK;
1019                         header |= RADEON_CRTC2_GUI_TRIG_VLINE >> 2;
1020                         break;
1021                 default:
1022                         DRM_ERROR("unknown crtc reloc\n");
1023                         r = -EINVAL;
1024                         goto out;
1025                 }
1026                 ib[h_idx] = header;
1027                 ib[h_idx + 3] |= RADEON_ENG_DISPLAY_SELECT_CRTC1;
1028         }
1029 out:
1030         mutex_unlock(&p->rdev->ddev->mode_config.mutex);
1031         return r;
1032 }
1033
1034 /**
1035  * r100_cs_packet_next_reloc() - parse next packet which should be reloc packet3
1036  * @parser:             parser structure holding parsing context.
1037  * @data:               pointer to relocation data
1038  * @offset_start:       starting offset
1039  * @offset_mask:        offset mask (to align start offset on)
1040  * @reloc:              reloc informations
1041  *
1042  * Check next packet is relocation packet3, do bo validation and compute
1043  * GPU offset using the provided start.
1044  **/
1045 int r100_cs_packet_next_reloc(struct radeon_cs_parser *p,
1046                               struct radeon_cs_reloc **cs_reloc)
1047 {
1048         struct radeon_cs_chunk *relocs_chunk;
1049         struct radeon_cs_packet p3reloc;
1050         unsigned idx;
1051         int r;
1052
1053         if (p->chunk_relocs_idx == -1) {
1054                 DRM_ERROR("No relocation chunk !\n");
1055                 return -EINVAL;
1056         }
1057         *cs_reloc = NULL;
1058         relocs_chunk = &p->chunks[p->chunk_relocs_idx];
1059         r = r100_cs_packet_parse(p, &p3reloc, p->idx);
1060         if (r) {
1061                 return r;
1062         }
1063         p->idx += p3reloc.count + 2;
1064         if (p3reloc.type != PACKET_TYPE3 || p3reloc.opcode != PACKET3_NOP) {
1065                 DRM_ERROR("No packet3 for relocation for packet at %d.\n",
1066                           p3reloc.idx);
1067                 r100_cs_dump_packet(p, &p3reloc);
1068                 return -EINVAL;
1069         }
1070         idx = radeon_get_ib_value(p, p3reloc.idx + 1);
1071         if (idx >= relocs_chunk->length_dw) {
1072                 DRM_ERROR("Relocs at %d after relocations chunk end %d !\n",
1073                           idx, relocs_chunk->length_dw);
1074                 r100_cs_dump_packet(p, &p3reloc);
1075                 return -EINVAL;
1076         }
1077         /* FIXME: we assume reloc size is 4 dwords */
1078         *cs_reloc = p->relocs_ptr[(idx / 4)];
1079         return 0;
1080 }
1081
1082 static int r100_get_vtx_size(uint32_t vtx_fmt)
1083 {
1084         int vtx_size;
1085         vtx_size = 2;
1086         /* ordered according to bits in spec */
1087         if (vtx_fmt & RADEON_SE_VTX_FMT_W0)
1088                 vtx_size++;
1089         if (vtx_fmt & RADEON_SE_VTX_FMT_FPCOLOR)
1090                 vtx_size += 3;
1091         if (vtx_fmt & RADEON_SE_VTX_FMT_FPALPHA)
1092                 vtx_size++;
1093         if (vtx_fmt & RADEON_SE_VTX_FMT_PKCOLOR)
1094                 vtx_size++;
1095         if (vtx_fmt & RADEON_SE_VTX_FMT_FPSPEC)
1096                 vtx_size += 3;
1097         if (vtx_fmt & RADEON_SE_VTX_FMT_FPFOG)
1098                 vtx_size++;
1099         if (vtx_fmt & RADEON_SE_VTX_FMT_PKSPEC)
1100                 vtx_size++;
1101         if (vtx_fmt & RADEON_SE_VTX_FMT_ST0)
1102                 vtx_size += 2;
1103         if (vtx_fmt & RADEON_SE_VTX_FMT_ST1)
1104                 vtx_size += 2;
1105         if (vtx_fmt & RADEON_SE_VTX_FMT_Q1)
1106                 vtx_size++;
1107         if (vtx_fmt & RADEON_SE_VTX_FMT_ST2)
1108                 vtx_size += 2;
1109         if (vtx_fmt & RADEON_SE_VTX_FMT_Q2)
1110                 vtx_size++;
1111         if (vtx_fmt & RADEON_SE_VTX_FMT_ST3)
1112                 vtx_size += 2;
1113         if (vtx_fmt & RADEON_SE_VTX_FMT_Q3)
1114                 vtx_size++;
1115         if (vtx_fmt & RADEON_SE_VTX_FMT_Q0)
1116                 vtx_size++;
1117         /* blend weight */
1118         if (vtx_fmt & (0x7 << 15))
1119                 vtx_size += (vtx_fmt >> 15) & 0x7;
1120         if (vtx_fmt & RADEON_SE_VTX_FMT_N0)
1121                 vtx_size += 3;
1122         if (vtx_fmt & RADEON_SE_VTX_FMT_XY1)
1123                 vtx_size += 2;
1124         if (vtx_fmt & RADEON_SE_VTX_FMT_Z1)
1125                 vtx_size++;
1126         if (vtx_fmt & RADEON_SE_VTX_FMT_W1)
1127                 vtx_size++;
1128         if (vtx_fmt & RADEON_SE_VTX_FMT_N1)
1129                 vtx_size++;
1130         if (vtx_fmt & RADEON_SE_VTX_FMT_Z)
1131                 vtx_size++;
1132         return vtx_size;
1133 }
1134
1135 static int r100_packet0_check(struct radeon_cs_parser *p,
1136                               struct radeon_cs_packet *pkt,
1137                               unsigned idx, unsigned reg)
1138 {
1139         struct radeon_cs_reloc *reloc;
1140         struct r100_cs_track *track;
1141         volatile uint32_t *ib;
1142         uint32_t tmp;
1143         int r;
1144         int i, face;
1145         u32 tile_flags = 0;
1146         u32 idx_value;
1147
1148         ib = p->ib->ptr;
1149         track = (struct r100_cs_track *)p->track;
1150
1151         idx_value = radeon_get_ib_value(p, idx);
1152
1153         switch (reg) {
1154         case RADEON_CRTC_GUI_TRIG_VLINE:
1155                 r = r100_cs_packet_parse_vline(p);
1156                 if (r) {
1157                         DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1158                                   idx, reg);
1159                         r100_cs_dump_packet(p, pkt);
1160                         return r;
1161                 }
1162                 break;
1163                 /* FIXME: only allow PACKET3 blit? easier to check for out of
1164                  * range access */
1165         case RADEON_DST_PITCH_OFFSET:
1166         case RADEON_SRC_PITCH_OFFSET:
1167                 r = r100_reloc_pitch_offset(p, pkt, idx, reg);
1168                 if (r)
1169                         return r;
1170                 break;
1171         case RADEON_RB3D_DEPTHOFFSET:
1172                 r = r100_cs_packet_next_reloc(p, &reloc);
1173                 if (r) {
1174                         DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1175                                   idx, reg);
1176                         r100_cs_dump_packet(p, pkt);
1177                         return r;
1178                 }
1179                 track->zb.robj = reloc->robj;
1180                 track->zb.offset = idx_value;
1181                 ib[idx] = idx_value + ((u32)reloc->lobj.gpu_offset);
1182                 break;
1183         case RADEON_RB3D_COLOROFFSET:
1184                 r = r100_cs_packet_next_reloc(p, &reloc);
1185                 if (r) {
1186                         DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1187                                   idx, reg);
1188                         r100_cs_dump_packet(p, pkt);
1189                         return r;
1190                 }
1191                 track->cb[0].robj = reloc->robj;
1192                 track->cb[0].offset = idx_value;
1193                 ib[idx] = idx_value + ((u32)reloc->lobj.gpu_offset);
1194                 break;
1195         case RADEON_PP_TXOFFSET_0:
1196         case RADEON_PP_TXOFFSET_1:
1197         case RADEON_PP_TXOFFSET_2:
1198                 i = (reg - RADEON_PP_TXOFFSET_0) / 24;
1199                 r = r100_cs_packet_next_reloc(p, &reloc);
1200                 if (r) {
1201                         DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1202                                   idx, reg);
1203                         r100_cs_dump_packet(p, pkt);
1204                         return r;
1205                 }
1206                 ib[idx] = idx_value + ((u32)reloc->lobj.gpu_offset);
1207                 track->textures[i].robj = reloc->robj;
1208                 break;
1209         case RADEON_PP_CUBIC_OFFSET_T0_0:
1210         case RADEON_PP_CUBIC_OFFSET_T0_1:
1211         case RADEON_PP_CUBIC_OFFSET_T0_2:
1212         case RADEON_PP_CUBIC_OFFSET_T0_3:
1213         case RADEON_PP_CUBIC_OFFSET_T0_4:
1214                 i = (reg - RADEON_PP_CUBIC_OFFSET_T0_0) / 4;
1215                 r = r100_cs_packet_next_reloc(p, &reloc);
1216                 if (r) {
1217                         DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1218                                   idx, reg);
1219                         r100_cs_dump_packet(p, pkt);
1220                         return r;
1221                 }
1222                 track->textures[0].cube_info[i].offset = idx_value;
1223                 ib[idx] = idx_value + ((u32)reloc->lobj.gpu_offset);
1224                 track->textures[0].cube_info[i].robj = reloc->robj;
1225                 break;
1226         case RADEON_PP_CUBIC_OFFSET_T1_0:
1227         case RADEON_PP_CUBIC_OFFSET_T1_1:
1228         case RADEON_PP_CUBIC_OFFSET_T1_2:
1229         case RADEON_PP_CUBIC_OFFSET_T1_3:
1230         case RADEON_PP_CUBIC_OFFSET_T1_4:
1231                 i = (reg - RADEON_PP_CUBIC_OFFSET_T1_0) / 4;
1232                 r = r100_cs_packet_next_reloc(p, &reloc);
1233                 if (r) {
1234                         DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1235                                   idx, reg);
1236                         r100_cs_dump_packet(p, pkt);
1237                         return r;
1238                 }
1239                 track->textures[1].cube_info[i].offset = idx_value;
1240                 ib[idx] = idx_value + ((u32)reloc->lobj.gpu_offset);
1241                 track->textures[1].cube_info[i].robj = reloc->robj;
1242                 break;
1243         case RADEON_PP_CUBIC_OFFSET_T2_0:
1244         case RADEON_PP_CUBIC_OFFSET_T2_1:
1245         case RADEON_PP_CUBIC_OFFSET_T2_2:
1246         case RADEON_PP_CUBIC_OFFSET_T2_3:
1247         case RADEON_PP_CUBIC_OFFSET_T2_4:
1248                 i = (reg - RADEON_PP_CUBIC_OFFSET_T2_0) / 4;
1249                 r = r100_cs_packet_next_reloc(p, &reloc);
1250                 if (r) {
1251                         DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1252                                   idx, reg);
1253                         r100_cs_dump_packet(p, pkt);
1254                         return r;
1255                 }
1256                 track->textures[2].cube_info[i].offset = idx_value;
1257                 ib[idx] = idx_value + ((u32)reloc->lobj.gpu_offset);
1258                 track->textures[2].cube_info[i].robj = reloc->robj;
1259                 break;
1260         case RADEON_RE_WIDTH_HEIGHT:
1261                 track->maxy = ((idx_value >> 16) & 0x7FF);
1262                 break;
1263         case RADEON_RB3D_COLORPITCH:
1264                 r = r100_cs_packet_next_reloc(p, &reloc);
1265                 if (r) {
1266                         DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1267                                   idx, reg);
1268                         r100_cs_dump_packet(p, pkt);
1269                         return r;
1270                 }
1271
1272                 if (reloc->lobj.tiling_flags & RADEON_TILING_MACRO)
1273                         tile_flags |= RADEON_COLOR_TILE_ENABLE;
1274                 if (reloc->lobj.tiling_flags & RADEON_TILING_MICRO)
1275                         tile_flags |= RADEON_COLOR_MICROTILE_ENABLE;
1276
1277                 tmp = idx_value & ~(0x7 << 16);
1278                 tmp |= tile_flags;
1279                 ib[idx] = tmp;
1280
1281                 track->cb[0].pitch = idx_value & RADEON_COLORPITCH_MASK;
1282                 break;
1283         case RADEON_RB3D_DEPTHPITCH:
1284                 track->zb.pitch = idx_value & RADEON_DEPTHPITCH_MASK;
1285                 break;
1286         case RADEON_RB3D_CNTL:
1287                 switch ((idx_value >> RADEON_RB3D_COLOR_FORMAT_SHIFT) & 0x1f) {
1288                 case 7:
1289                 case 8:
1290                 case 9:
1291                 case 11:
1292                 case 12:
1293                         track->cb[0].cpp = 1;
1294                         break;
1295                 case 3:
1296                 case 4:
1297                 case 15:
1298                         track->cb[0].cpp = 2;
1299                         break;
1300                 case 6:
1301                         track->cb[0].cpp = 4;
1302                         break;
1303                 default:
1304                         DRM_ERROR("Invalid color buffer format (%d) !\n",
1305                                   ((idx_value >> RADEON_RB3D_COLOR_FORMAT_SHIFT) & 0x1f));
1306                         return -EINVAL;
1307                 }
1308                 track->z_enabled = !!(idx_value & RADEON_Z_ENABLE);
1309                 break;
1310         case RADEON_RB3D_ZSTENCILCNTL:
1311                 switch (idx_value & 0xf) {
1312                 case 0:
1313                         track->zb.cpp = 2;
1314                         break;
1315                 case 2:
1316                 case 3:
1317                 case 4:
1318                 case 5:
1319                 case 9:
1320                 case 11:
1321                         track->zb.cpp = 4;
1322                         break;
1323                 default:
1324                         break;
1325                 }
1326                 break;
1327         case RADEON_RB3D_ZPASS_ADDR:
1328                 r = r100_cs_packet_next_reloc(p, &reloc);
1329                 if (r) {
1330                         DRM_ERROR("No reloc for ib[%d]=0x%04X\n",
1331                                   idx, reg);
1332                         r100_cs_dump_packet(p, pkt);
1333                         return r;
1334                 }
1335                 ib[idx] = idx_value + ((u32)reloc->lobj.gpu_offset);
1336                 break;
1337         case RADEON_PP_CNTL:
1338                 {
1339                         uint32_t temp = idx_value >> 4;
1340                         for (i = 0; i < track->num_texture; i++)
1341                                 track->textures[i].enabled = !!(temp & (1 << i));
1342                 }
1343                 break;
1344         case RADEON_SE_VF_CNTL:
1345                 track->vap_vf_cntl = idx_value;
1346                 break;
1347         case RADEON_SE_VTX_FMT:
1348                 track->vtx_size = r100_get_vtx_size(idx_value);
1349                 break;
1350         case RADEON_PP_TEX_SIZE_0:
1351         case RADEON_PP_TEX_SIZE_1:
1352         case RADEON_PP_TEX_SIZE_2:
1353                 i = (reg - RADEON_PP_TEX_SIZE_0) / 8;
1354                 track->textures[i].width = (idx_value & RADEON_TEX_USIZE_MASK) + 1;
1355                 track->textures[i].height = ((idx_value & RADEON_TEX_VSIZE_MASK) >> RADEON_TEX_VSIZE_SHIFT) + 1;
1356                 break;
1357         case RADEON_PP_TEX_PITCH_0:
1358         case RADEON_PP_TEX_PITCH_1:
1359         case RADEON_PP_TEX_PITCH_2:
1360                 i = (reg - RADEON_PP_TEX_PITCH_0) / 8;
1361                 track->textures[i].pitch = idx_value + 32;
1362                 break;
1363         case RADEON_PP_TXFILTER_0:
1364         case RADEON_PP_TXFILTER_1:
1365         case RADEON_PP_TXFILTER_2:
1366                 i = (reg - RADEON_PP_TXFILTER_0) / 24;
1367                 track->textures[i].num_levels = ((idx_value & RADEON_MAX_MIP_LEVEL_MASK)
1368                                                  >> RADEON_MAX_MIP_LEVEL_SHIFT);
1369                 tmp = (idx_value >> 23) & 0x7;
1370                 if (tmp == 2 || tmp == 6)
1371                         track->textures[i].roundup_w = false;
1372                 tmp = (idx_value >> 27) & 0x7;
1373                 if (tmp == 2 || tmp == 6)
1374                         track->textures[i].roundup_h = false;
1375                 break;
1376         case RADEON_PP_TXFORMAT_0:
1377         case RADEON_PP_TXFORMAT_1:
1378         case RADEON_PP_TXFORMAT_2:
1379                 i = (reg - RADEON_PP_TXFORMAT_0) / 24;
1380                 if (idx_value & RADEON_TXFORMAT_NON_POWER2) {
1381                         track->textures[i].use_pitch = 1;
1382                 } else {
1383                         track->textures[i].use_pitch = 0;
1384                         track->textures[i].width = 1 << ((idx_value >> RADEON_TXFORMAT_WIDTH_SHIFT) & RADEON_TXFORMAT_WIDTH_MASK);
1385                         track->textures[i].height = 1 << ((idx_value >> RADEON_TXFORMAT_HEIGHT_SHIFT) & RADEON_TXFORMAT_HEIGHT_MASK);
1386                 }
1387                 if (idx_value & RADEON_TXFORMAT_CUBIC_MAP_ENABLE)
1388                         track->textures[i].tex_coord_type = 2;
1389                 switch ((idx_value & RADEON_TXFORMAT_FORMAT_MASK)) {
1390                 case RADEON_TXFORMAT_I8:
1391                 case RADEON_TXFORMAT_RGB332:
1392                 case RADEON_TXFORMAT_Y8:
1393                         track->textures[i].cpp = 1;
1394                         break;
1395                 case RADEON_TXFORMAT_AI88:
1396                 case RADEON_TXFORMAT_ARGB1555:
1397                 case RADEON_TXFORMAT_RGB565:
1398                 case RADEON_TXFORMAT_ARGB4444:
1399                 case RADEON_TXFORMAT_VYUY422:
1400                 case RADEON_TXFORMAT_YVYU422:
1401                 case RADEON_TXFORMAT_SHADOW16:
1402                 case RADEON_TXFORMAT_LDUDV655:
1403                 case RADEON_TXFORMAT_DUDV88:
1404                         track->textures[i].cpp = 2;
1405                         break;
1406                 case RADEON_TXFORMAT_ARGB8888:
1407                 case RADEON_TXFORMAT_RGBA8888:
1408                 case RADEON_TXFORMAT_SHADOW32:
1409                 case RADEON_TXFORMAT_LDUDUV8888:
1410                         track->textures[i].cpp = 4;
1411                         break;
1412                 case RADEON_TXFORMAT_DXT1:
1413                         track->textures[i].cpp = 1;
1414                         track->textures[i].compress_format = R100_TRACK_COMP_DXT1;
1415                         break;
1416                 case RADEON_TXFORMAT_DXT23:
1417                 case RADEON_TXFORMAT_DXT45:
1418                         track->textures[i].cpp = 1;
1419                         track->textures[i].compress_format = R100_TRACK_COMP_DXT35;
1420                         break;
1421                 }
1422                 track->textures[i].cube_info[4].width = 1 << ((idx_value >> 16) & 0xf);
1423                 track->textures[i].cube_info[4].height = 1 << ((idx_value >> 20) & 0xf);
1424                 break;
1425         case RADEON_PP_CUBIC_FACES_0:
1426         case RADEON_PP_CUBIC_FACES_1:
1427         case RADEON_PP_CUBIC_FACES_2:
1428                 tmp = idx_value;
1429                 i = (reg - RADEON_PP_CUBIC_FACES_0) / 4;
1430                 for (face = 0; face < 4; face++) {
1431                         track->textures[i].cube_info[face].width = 1 << ((tmp >> (face * 8)) & 0xf);
1432                         track->textures[i].cube_info[face].height = 1 << ((tmp >> ((face * 8) + 4)) & 0xf);
1433                 }
1434                 break;
1435         default:
1436                 printk(KERN_ERR "Forbidden register 0x%04X in cs at %d\n",
1437                        reg, idx);
1438                 return -EINVAL;
1439         }
1440         return 0;
1441 }
1442
1443 int r100_cs_track_check_pkt3_indx_buffer(struct radeon_cs_parser *p,
1444                                          struct radeon_cs_packet *pkt,
1445                                          struct radeon_bo *robj)
1446 {
1447         unsigned idx;
1448         u32 value;
1449         idx = pkt->idx + 1;
1450         value = radeon_get_ib_value(p, idx + 2);
1451         if ((value + 1) > radeon_bo_size(robj)) {
1452                 DRM_ERROR("[drm] Buffer too small for PACKET3 INDX_BUFFER "
1453                           "(need %u have %lu) !\n",
1454                           value + 1,
1455                           radeon_bo_size(robj));
1456                 return -EINVAL;
1457         }
1458         return 0;
1459 }
1460
1461 static int r100_packet3_check(struct radeon_cs_parser *p,
1462                               struct radeon_cs_packet *pkt)
1463 {
1464         struct radeon_cs_reloc *reloc;
1465         struct r100_cs_track *track;
1466         unsigned idx;
1467         volatile uint32_t *ib;
1468         int r;
1469
1470         ib = p->ib->ptr;
1471         idx = pkt->idx + 1;
1472         track = (struct r100_cs_track *)p->track;
1473         switch (pkt->opcode) {
1474         case PACKET3_3D_LOAD_VBPNTR:
1475                 r = r100_packet3_load_vbpntr(p, pkt, idx);
1476                 if (r)
1477                         return r;
1478                 break;
1479         case PACKET3_INDX_BUFFER:
1480                 r = r100_cs_packet_next_reloc(p, &reloc);
1481                 if (r) {
1482                         DRM_ERROR("No reloc for packet3 %d\n", pkt->opcode);
1483                         r100_cs_dump_packet(p, pkt);
1484                         return r;
1485                 }
1486                 ib[idx+1] = radeon_get_ib_value(p, idx+1) + ((u32)reloc->lobj.gpu_offset);
1487                 r = r100_cs_track_check_pkt3_indx_buffer(p, pkt, reloc->robj);
1488                 if (r) {
1489                         return r;
1490                 }
1491                 break;
1492         case 0x23:
1493                 /* 3D_RNDR_GEN_INDX_PRIM on r100/r200 */
1494                 r = r100_cs_packet_next_reloc(p, &reloc);
1495                 if (r) {
1496                         DRM_ERROR("No reloc for packet3 %d\n", pkt->opcode);
1497                         r100_cs_dump_packet(p, pkt);
1498                         return r;
1499                 }
1500                 ib[idx] = radeon_get_ib_value(p, idx) + ((u32)reloc->lobj.gpu_offset);
1501                 track->num_arrays = 1;
1502                 track->vtx_size = r100_get_vtx_size(radeon_get_ib_value(p, idx + 2));
1503
1504                 track->arrays[0].robj = reloc->robj;
1505                 track->arrays[0].esize = track->vtx_size;
1506
1507                 track->max_indx = radeon_get_ib_value(p, idx+1);
1508
1509                 track->vap_vf_cntl = radeon_get_ib_value(p, idx+3);
1510                 track->immd_dwords = pkt->count - 1;
1511                 r = r100_cs_track_check(p->rdev, track);
1512                 if (r)
1513                         return r;
1514                 break;
1515         case PACKET3_3D_DRAW_IMMD:
1516                 if (((radeon_get_ib_value(p, idx + 1) >> 4) & 0x3) != 3) {
1517                         DRM_ERROR("PRIM_WALK must be 3 for IMMD draw\n");
1518                         return -EINVAL;
1519                 }
1520                 track->vtx_size = r100_get_vtx_size(radeon_get_ib_value(p, idx + 0));
1521                 track->vap_vf_cntl = radeon_get_ib_value(p, idx + 1);
1522                 track->immd_dwords = pkt->count - 1;
1523                 r = r100_cs_track_check(p->rdev, track);
1524                 if (r)
1525                         return r;
1526                 break;
1527                 /* triggers drawing using in-packet vertex data */
1528         case PACKET3_3D_DRAW_IMMD_2:
1529                 if (((radeon_get_ib_value(p, idx) >> 4) & 0x3) != 3) {
1530                         DRM_ERROR("PRIM_WALK must be 3 for IMMD draw\n");
1531                         return -EINVAL;
1532                 }
1533                 track->vap_vf_cntl = radeon_get_ib_value(p, idx);
1534                 track->immd_dwords = pkt->count;
1535                 r = r100_cs_track_check(p->rdev, track);
1536                 if (r)
1537                         return r;
1538                 break;
1539                 /* triggers drawing using in-packet vertex data */
1540         case PACKET3_3D_DRAW_VBUF_2:
1541                 track->vap_vf_cntl = radeon_get_ib_value(p, idx);
1542                 r = r100_cs_track_check(p->rdev, track);
1543                 if (r)
1544                         return r;
1545                 break;
1546                 /* triggers drawing of vertex buffers setup elsewhere */
1547         case PACKET3_3D_DRAW_INDX_2:
1548                 track->vap_vf_cntl = radeon_get_ib_value(p, idx);
1549                 r = r100_cs_track_check(p->rdev, track);
1550                 if (r)
1551                         return r;
1552                 break;
1553                 /* triggers drawing using indices to vertex buffer */
1554         case PACKET3_3D_DRAW_VBUF:
1555                 track->vap_vf_cntl = radeon_get_ib_value(p, idx + 1);
1556                 r = r100_cs_track_check(p->rdev, track);
1557                 if (r)
1558                         return r;
1559                 break;
1560                 /* triggers drawing of vertex buffers setup elsewhere */
1561         case PACKET3_3D_DRAW_INDX:
1562                 track->vap_vf_cntl = radeon_get_ib_value(p, idx + 1);
1563                 r = r100_cs_track_check(p->rdev, track);
1564                 if (r)
1565                         return r;
1566                 break;
1567                 /* triggers drawing using indices to vertex buffer */
1568         case PACKET3_NOP:
1569                 break;
1570         default:
1571                 DRM_ERROR("Packet3 opcode %x not supported\n", pkt->opcode);
1572                 return -EINVAL;
1573         }
1574         return 0;
1575 }
1576
1577 int r100_cs_parse(struct radeon_cs_parser *p)
1578 {
1579         struct radeon_cs_packet pkt;
1580         struct r100_cs_track *track;
1581         int r;
1582
1583         track = kzalloc(sizeof(*track), GFP_KERNEL);
1584         r100_cs_track_clear(p->rdev, track);
1585         p->track = track;
1586         do {
1587                 r = r100_cs_packet_parse(p, &pkt, p->idx);
1588                 if (r) {
1589                         return r;
1590                 }
1591                 p->idx += pkt.count + 2;
1592                 switch (pkt.type) {
1593                         case PACKET_TYPE0:
1594                                 if (p->rdev->family >= CHIP_R200)
1595                                         r = r100_cs_parse_packet0(p, &pkt,
1596                                                                   p->rdev->config.r100.reg_safe_bm,
1597                                                                   p->rdev->config.r100.reg_safe_bm_size,
1598                                                                   &r200_packet0_check);
1599                                 else
1600                                         r = r100_cs_parse_packet0(p, &pkt,
1601                                                                   p->rdev->config.r100.reg_safe_bm,
1602                                                                   p->rdev->config.r100.reg_safe_bm_size,
1603                                                                   &r100_packet0_check);
1604                                 break;
1605                         case PACKET_TYPE2:
1606                                 break;
1607                         case PACKET_TYPE3:
1608                                 r = r100_packet3_check(p, &pkt);
1609                                 break;
1610                         default:
1611                                 DRM_ERROR("Unknown packet type %d !\n",
1612                                           pkt.type);
1613                                 return -EINVAL;
1614                 }
1615                 if (r) {
1616                         return r;
1617                 }
1618         } while (p->idx < p->chunks[p->chunk_ib_idx].length_dw);
1619         return 0;
1620 }
1621
1622
1623 /*
1624  * Global GPU functions
1625  */
1626 void r100_errata(struct radeon_device *rdev)
1627 {
1628         rdev->pll_errata = 0;
1629
1630         if (rdev->family == CHIP_RV200 || rdev->family == CHIP_RS200) {
1631                 rdev->pll_errata |= CHIP_ERRATA_PLL_DUMMYREADS;
1632         }
1633
1634         if (rdev->family == CHIP_RV100 ||
1635             rdev->family == CHIP_RS100 ||
1636             rdev->family == CHIP_RS200) {
1637                 rdev->pll_errata |= CHIP_ERRATA_PLL_DELAY;
1638         }
1639 }
1640
1641 /* Wait for vertical sync on primary CRTC */
1642 void r100_gpu_wait_for_vsync(struct radeon_device *rdev)
1643 {
1644         uint32_t crtc_gen_cntl, tmp;
1645         int i;
1646
1647         crtc_gen_cntl = RREG32(RADEON_CRTC_GEN_CNTL);
1648         if ((crtc_gen_cntl & RADEON_CRTC_DISP_REQ_EN_B) ||
1649             !(crtc_gen_cntl & RADEON_CRTC_EN)) {
1650                 return;
1651         }
1652         /* Clear the CRTC_VBLANK_SAVE bit */
1653         WREG32(RADEON_CRTC_STATUS, RADEON_CRTC_VBLANK_SAVE_CLEAR);
1654         for (i = 0; i < rdev->usec_timeout; i++) {
1655                 tmp = RREG32(RADEON_CRTC_STATUS);
1656                 if (tmp & RADEON_CRTC_VBLANK_SAVE) {
1657                         return;
1658                 }
1659                 DRM_UDELAY(1);
1660         }
1661 }
1662
1663 /* Wait for vertical sync on secondary CRTC */
1664 void r100_gpu_wait_for_vsync2(struct radeon_device *rdev)
1665 {
1666         uint32_t crtc2_gen_cntl, tmp;
1667         int i;
1668
1669         crtc2_gen_cntl = RREG32(RADEON_CRTC2_GEN_CNTL);
1670         if ((crtc2_gen_cntl & RADEON_CRTC2_DISP_REQ_EN_B) ||
1671             !(crtc2_gen_cntl & RADEON_CRTC2_EN))
1672                 return;
1673
1674         /* Clear the CRTC_VBLANK_SAVE bit */
1675         WREG32(RADEON_CRTC2_STATUS, RADEON_CRTC2_VBLANK_SAVE_CLEAR);
1676         for (i = 0; i < rdev->usec_timeout; i++) {
1677                 tmp = RREG32(RADEON_CRTC2_STATUS);
1678                 if (tmp & RADEON_CRTC2_VBLANK_SAVE) {
1679                         return;
1680                 }
1681                 DRM_UDELAY(1);
1682         }
1683 }
1684
1685 int r100_rbbm_fifo_wait_for_entry(struct radeon_device *rdev, unsigned n)
1686 {
1687         unsigned i;
1688         uint32_t tmp;
1689
1690         for (i = 0; i < rdev->usec_timeout; i++) {
1691                 tmp = RREG32(RADEON_RBBM_STATUS) & RADEON_RBBM_FIFOCNT_MASK;
1692                 if (tmp >= n) {
1693                         return 0;
1694                 }
1695                 DRM_UDELAY(1);
1696         }
1697         return -1;
1698 }
1699
1700 int r100_gui_wait_for_idle(struct radeon_device *rdev)
1701 {
1702         unsigned i;
1703         uint32_t tmp;
1704
1705         if (r100_rbbm_fifo_wait_for_entry(rdev, 64)) {
1706                 printk(KERN_WARNING "radeon: wait for empty RBBM fifo failed !"
1707                        " Bad things might happen.\n");
1708         }
1709         for (i = 0; i < rdev->usec_timeout; i++) {
1710                 tmp = RREG32(RADEON_RBBM_STATUS);
1711                 if (!(tmp & RADEON_RBBM_ACTIVE)) {
1712                         return 0;
1713                 }
1714                 DRM_UDELAY(1);
1715         }
1716         return -1;
1717 }
1718
1719 int r100_mc_wait_for_idle(struct radeon_device *rdev)
1720 {
1721         unsigned i;
1722         uint32_t tmp;
1723
1724         for (i = 0; i < rdev->usec_timeout; i++) {
1725                 /* read MC_STATUS */
1726                 tmp = RREG32(RADEON_MC_STATUS);
1727                 if (tmp & RADEON_MC_IDLE) {
1728                         return 0;
1729                 }
1730                 DRM_UDELAY(1);
1731         }
1732         return -1;
1733 }
1734
1735 void r100_gpu_init(struct radeon_device *rdev)
1736 {
1737         /* TODO: anythings to do here ? pipes ? */
1738         r100_hdp_reset(rdev);
1739 }
1740
1741 void r100_hdp_reset(struct radeon_device *rdev)
1742 {
1743         uint32_t tmp;
1744
1745         tmp = RREG32(RADEON_HOST_PATH_CNTL) & RADEON_HDP_APER_CNTL;
1746         tmp |= (7 << 28);
1747         WREG32(RADEON_HOST_PATH_CNTL, tmp | RADEON_HDP_SOFT_RESET | RADEON_HDP_READ_BUFFER_INVALIDATE);
1748         (void)RREG32(RADEON_HOST_PATH_CNTL);
1749         udelay(200);
1750         WREG32(RADEON_RBBM_SOFT_RESET, 0);
1751         WREG32(RADEON_HOST_PATH_CNTL, tmp);
1752         (void)RREG32(RADEON_HOST_PATH_CNTL);
1753 }
1754
1755 int r100_rb2d_reset(struct radeon_device *rdev)
1756 {
1757         uint32_t tmp;
1758         int i;
1759
1760         WREG32(RADEON_RBBM_SOFT_RESET, RADEON_SOFT_RESET_E2);
1761         (void)RREG32(RADEON_RBBM_SOFT_RESET);
1762         udelay(200);
1763         WREG32(RADEON_RBBM_SOFT_RESET, 0);
1764         /* Wait to prevent race in RBBM_STATUS */
1765         mdelay(1);
1766         for (i = 0; i < rdev->usec_timeout; i++) {
1767                 tmp = RREG32(RADEON_RBBM_STATUS);
1768                 if (!(tmp & (1 << 26))) {
1769                         DRM_INFO("RB2D reset succeed (RBBM_STATUS=0x%08X)\n",
1770                                  tmp);
1771                         return 0;
1772                 }
1773                 DRM_UDELAY(1);
1774         }
1775         tmp = RREG32(RADEON_RBBM_STATUS);
1776         DRM_ERROR("Failed to reset RB2D (RBBM_STATUS=0x%08X)!\n", tmp);
1777         return -1;
1778 }
1779
1780 int r100_gpu_reset(struct radeon_device *rdev)
1781 {
1782         uint32_t status;
1783
1784         /* reset order likely matter */
1785         status = RREG32(RADEON_RBBM_STATUS);
1786         /* reset HDP */
1787         r100_hdp_reset(rdev);
1788         /* reset rb2d */
1789         if (status & ((1 << 17) | (1 << 18) | (1 << 27))) {
1790                 r100_rb2d_reset(rdev);
1791         }
1792         /* TODO: reset 3D engine */
1793         /* reset CP */
1794         status = RREG32(RADEON_RBBM_STATUS);
1795         if (status & (1 << 16)) {
1796                 r100_cp_reset(rdev);
1797         }
1798         /* Check if GPU is idle */
1799         status = RREG32(RADEON_RBBM_STATUS);
1800         if (status & RADEON_RBBM_ACTIVE) {
1801                 DRM_ERROR("Failed to reset GPU (RBBM_STATUS=0x%08X)\n", status);
1802                 return -1;
1803         }
1804         DRM_INFO("GPU reset succeed (RBBM_STATUS=0x%08X)\n", status);
1805         return 0;
1806 }
1807
1808 void r100_set_common_regs(struct radeon_device *rdev)
1809 {
1810         struct drm_device *dev = rdev->ddev;
1811         bool force_dac2 = false;
1812         u32 tmp;
1813
1814         /* set these so they don't interfere with anything */
1815         WREG32(RADEON_OV0_SCALE_CNTL, 0);
1816         WREG32(RADEON_SUBPIC_CNTL, 0);
1817         WREG32(RADEON_VIPH_CONTROL, 0);
1818         WREG32(RADEON_I2C_CNTL_1, 0);
1819         WREG32(RADEON_DVI_I2C_CNTL_1, 0);
1820         WREG32(RADEON_CAP0_TRIG_CNTL, 0);
1821         WREG32(RADEON_CAP1_TRIG_CNTL, 0);
1822
1823         /* always set up dac2 on rn50 and some rv100 as lots
1824          * of servers seem to wire it up to a VGA port but
1825          * don't report it in the bios connector
1826          * table.
1827          */
1828         switch (dev->pdev->device) {
1829                 /* RN50 */
1830         case 0x515e:
1831         case 0x5969:
1832                 force_dac2 = true;
1833                 break;
1834                 /* RV100*/
1835         case 0x5159:
1836         case 0x515a:
1837                 /* DELL triple head servers */
1838                 if ((dev->pdev->subsystem_vendor == 0x1028 /* DELL */) &&
1839                     ((dev->pdev->subsystem_device == 0x016c) ||
1840                      (dev->pdev->subsystem_device == 0x016d) ||
1841                      (dev->pdev->subsystem_device == 0x016e) ||
1842                      (dev->pdev->subsystem_device == 0x016f) ||
1843                      (dev->pdev->subsystem_device == 0x0170) ||
1844                      (dev->pdev->subsystem_device == 0x017d) ||
1845                      (dev->pdev->subsystem_device == 0x017e) ||
1846                      (dev->pdev->subsystem_device == 0x0183) ||
1847                      (dev->pdev->subsystem_device == 0x018a) ||
1848                      (dev->pdev->subsystem_device == 0x019a)))
1849                         force_dac2 = true;
1850                 break;
1851         }
1852
1853         if (force_dac2) {
1854                 u32 disp_hw_debug = RREG32(RADEON_DISP_HW_DEBUG);
1855                 u32 tv_dac_cntl = RREG32(RADEON_TV_DAC_CNTL);
1856                 u32 dac2_cntl = RREG32(RADEON_DAC_CNTL2);
1857
1858                 /* For CRT on DAC2, don't turn it on if BIOS didn't
1859                    enable it, even it's detected.
1860                 */
1861
1862                 /* force it to crtc0 */
1863                 dac2_cntl &= ~RADEON_DAC2_DAC_CLK_SEL;
1864                 dac2_cntl |= RADEON_DAC2_DAC2_CLK_SEL;
1865                 disp_hw_debug |= RADEON_CRT2_DISP1_SEL;
1866
1867                 /* set up the TV DAC */
1868                 tv_dac_cntl &= ~(RADEON_TV_DAC_PEDESTAL |
1869                                  RADEON_TV_DAC_STD_MASK |
1870                                  RADEON_TV_DAC_RDACPD |
1871                                  RADEON_TV_DAC_GDACPD |
1872                                  RADEON_TV_DAC_BDACPD |
1873                                  RADEON_TV_DAC_BGADJ_MASK |
1874                                  RADEON_TV_DAC_DACADJ_MASK);
1875                 tv_dac_cntl |= (RADEON_TV_DAC_NBLANK |
1876                                 RADEON_TV_DAC_NHOLD |
1877                                 RADEON_TV_DAC_STD_PS2 |
1878                                 (0x58 << 16));
1879
1880                 WREG32(RADEON_TV_DAC_CNTL, tv_dac_cntl);
1881                 WREG32(RADEON_DISP_HW_DEBUG, disp_hw_debug);
1882                 WREG32(RADEON_DAC_CNTL2, dac2_cntl);
1883         }
1884
1885         /* switch PM block to ACPI mode */
1886         tmp = RREG32_PLL(RADEON_PLL_PWRMGT_CNTL);
1887         tmp &= ~RADEON_PM_MODE_SEL;
1888         WREG32_PLL(RADEON_PLL_PWRMGT_CNTL, tmp);
1889
1890 }
1891
1892 /*
1893  * VRAM info
1894  */
1895 static void r100_vram_get_type(struct radeon_device *rdev)
1896 {
1897         uint32_t tmp;
1898
1899         rdev->mc.vram_is_ddr = false;
1900         if (rdev->flags & RADEON_IS_IGP)
1901                 rdev->mc.vram_is_ddr = true;
1902         else if (RREG32(RADEON_MEM_SDRAM_MODE_REG) & RADEON_MEM_CFG_TYPE_DDR)
1903                 rdev->mc.vram_is_ddr = true;
1904         if ((rdev->family == CHIP_RV100) ||
1905             (rdev->family == CHIP_RS100) ||
1906             (rdev->family == CHIP_RS200)) {
1907                 tmp = RREG32(RADEON_MEM_CNTL);
1908                 if (tmp & RV100_HALF_MODE) {
1909                         rdev->mc.vram_width = 32;
1910                 } else {
1911                         rdev->mc.vram_width = 64;
1912                 }
1913                 if (rdev->flags & RADEON_SINGLE_CRTC) {
1914                         rdev->mc.vram_width /= 4;
1915                         rdev->mc.vram_is_ddr = true;
1916                 }
1917         } else if (rdev->family <= CHIP_RV280) {
1918                 tmp = RREG32(RADEON_MEM_CNTL);
1919                 if (tmp & RADEON_MEM_NUM_CHANNELS_MASK) {
1920                         rdev->mc.vram_width = 128;
1921                 } else {
1922                         rdev->mc.vram_width = 64;
1923                 }
1924         } else {
1925                 /* newer IGPs */
1926                 rdev->mc.vram_width = 128;
1927         }
1928 }
1929
1930 static u32 r100_get_accessible_vram(struct radeon_device *rdev)
1931 {
1932         u32 aper_size;
1933         u8 byte;
1934
1935         aper_size = RREG32(RADEON_CONFIG_APER_SIZE);
1936
1937         /* Set HDP_APER_CNTL only on cards that are known not to be broken,
1938          * that is has the 2nd generation multifunction PCI interface
1939          */
1940         if (rdev->family == CHIP_RV280 ||
1941             rdev->family >= CHIP_RV350) {
1942                 WREG32_P(RADEON_HOST_PATH_CNTL, RADEON_HDP_APER_CNTL,
1943                        ~RADEON_HDP_APER_CNTL);
1944                 DRM_INFO("Generation 2 PCI interface, using max accessible memory\n");
1945                 return aper_size * 2;
1946         }
1947
1948         /* Older cards have all sorts of funny issues to deal with. First
1949          * check if it's a multifunction card by reading the PCI config
1950          * header type... Limit those to one aperture size
1951          */
1952         pci_read_config_byte(rdev->pdev, 0xe, &byte);
1953         if (byte & 0x80) {
1954                 DRM_INFO("Generation 1 PCI interface in multifunction mode\n");
1955                 DRM_INFO("Limiting VRAM to one aperture\n");
1956                 return aper_size;
1957         }
1958
1959         /* Single function older card. We read HDP_APER_CNTL to see how the BIOS
1960          * have set it up. We don't write this as it's broken on some ASICs but
1961          * we expect the BIOS to have done the right thing (might be too optimistic...)
1962          */
1963         if (RREG32(RADEON_HOST_PATH_CNTL) & RADEON_HDP_APER_CNTL)
1964                 return aper_size * 2;
1965         return aper_size;
1966 }
1967
1968 void r100_vram_init_sizes(struct radeon_device *rdev)
1969 {
1970         u64 config_aper_size;
1971
1972         /* work out accessible VRAM */
1973         rdev->mc.aper_base = drm_get_resource_start(rdev->ddev, 0);
1974         rdev->mc.aper_size = drm_get_resource_len(rdev->ddev, 0);
1975         rdev->mc.visible_vram_size = r100_get_accessible_vram(rdev);
1976         /* FIXME we don't use the second aperture yet when we could use it */
1977         if (rdev->mc.visible_vram_size > rdev->mc.aper_size)
1978                 rdev->mc.visible_vram_size = rdev->mc.aper_size;
1979         config_aper_size = RREG32(RADEON_CONFIG_APER_SIZE);
1980         if (rdev->flags & RADEON_IS_IGP) {
1981                 uint32_t tom;
1982                 /* read NB_TOM to get the amount of ram stolen for the GPU */
1983                 tom = RREG32(RADEON_NB_TOM);
1984                 rdev->mc.real_vram_size = (((tom >> 16) - (tom & 0xffff) + 1) << 16);
1985                 WREG32(RADEON_CONFIG_MEMSIZE, rdev->mc.real_vram_size);
1986                 rdev->mc.mc_vram_size = rdev->mc.real_vram_size;
1987         } else {
1988                 rdev->mc.real_vram_size = RREG32(RADEON_CONFIG_MEMSIZE);
1989                 /* Some production boards of m6 will report 0
1990                  * if it's 8 MB
1991                  */
1992                 if (rdev->mc.real_vram_size == 0) {
1993                         rdev->mc.real_vram_size = 8192 * 1024;
1994                         WREG32(RADEON_CONFIG_MEMSIZE, rdev->mc.real_vram_size);
1995                 }
1996                 /* Fix for RN50, M6, M7 with 8/16/32(??) MBs of VRAM -
1997                  * Novell bug 204882 + along with lots of ubuntu ones
1998                  */
1999                 if (config_aper_size > rdev->mc.real_vram_size)
2000                         rdev->mc.mc_vram_size = config_aper_size;
2001                 else
2002                         rdev->mc.mc_vram_size = rdev->mc.real_vram_size;
2003         }
2004         /* FIXME remove this once we support unmappable VRAM */
2005         if (rdev->mc.mc_vram_size > rdev->mc.aper_size) {
2006                 rdev->mc.mc_vram_size = rdev->mc.aper_size;
2007                 rdev->mc.real_vram_size = rdev->mc.aper_size;
2008         }
2009 }
2010
2011 void r100_vga_set_state(struct radeon_device *rdev, bool state)
2012 {
2013         uint32_t temp;
2014
2015         temp = RREG32(RADEON_CONFIG_CNTL);
2016         if (state == false) {
2017                 temp &= ~(1<<8);
2018                 temp |= (1<<9);
2019         } else {
2020                 temp &= ~(1<<9);
2021         }
2022         WREG32(RADEON_CONFIG_CNTL, temp);
2023 }
2024
2025 void r100_mc_init(struct radeon_device *rdev)
2026 {
2027         u64 base;
2028
2029         r100_vram_get_type(rdev);
2030         r100_vram_init_sizes(rdev);
2031         base = rdev->mc.aper_base;
2032         if (rdev->flags & RADEON_IS_IGP)
2033                 base = (RREG32(RADEON_NB_TOM) & 0xffff) << 16;
2034         radeon_vram_location(rdev, &rdev->mc, base);
2035         if (!(rdev->flags & RADEON_IS_AGP))
2036                 radeon_gtt_location(rdev, &rdev->mc);
2037         radeon_update_bandwidth_info(rdev);
2038 }
2039
2040
2041 /*
2042  * Indirect registers accessor
2043  */
2044 void r100_pll_errata_after_index(struct radeon_device *rdev)
2045 {
2046         if (!(rdev->pll_errata & CHIP_ERRATA_PLL_DUMMYREADS)) {
2047                 return;
2048         }
2049         (void)RREG32(RADEON_CLOCK_CNTL_DATA);
2050         (void)RREG32(RADEON_CRTC_GEN_CNTL);
2051 }
2052
2053 static void r100_pll_errata_after_data(struct radeon_device *rdev)
2054 {
2055         /* This workarounds is necessary on RV100, RS100 and RS200 chips
2056          * or the chip could hang on a subsequent access
2057          */
2058         if (rdev->pll_errata & CHIP_ERRATA_PLL_DELAY) {
2059                 udelay(5000);
2060         }
2061
2062         /* This function is required to workaround a hardware bug in some (all?)
2063          * revisions of the R300.  This workaround should be called after every
2064          * CLOCK_CNTL_INDEX register access.  If not, register reads afterward
2065          * may not be correct.
2066          */
2067         if (rdev->pll_errata & CHIP_ERRATA_R300_CG) {
2068                 uint32_t save, tmp;
2069
2070                 save = RREG32(RADEON_CLOCK_CNTL_INDEX);
2071                 tmp = save & ~(0x3f | RADEON_PLL_WR_EN);
2072                 WREG32(RADEON_CLOCK_CNTL_INDEX, tmp);
2073                 tmp = RREG32(RADEON_CLOCK_CNTL_DATA);
2074                 WREG32(RADEON_CLOCK_CNTL_INDEX, save);
2075         }
2076 }
2077
2078 uint32_t r100_pll_rreg(struct radeon_device *rdev, uint32_t reg)
2079 {
2080         uint32_t data;
2081
2082         WREG8(RADEON_CLOCK_CNTL_INDEX, reg & 0x3f);
2083         r100_pll_errata_after_index(rdev);
2084         data = RREG32(RADEON_CLOCK_CNTL_DATA);
2085         r100_pll_errata_after_data(rdev);
2086         return data;
2087 }
2088
2089 void r100_pll_wreg(struct radeon_device *rdev, uint32_t reg, uint32_t v)
2090 {
2091         WREG8(RADEON_CLOCK_CNTL_INDEX, ((reg & 0x3f) | RADEON_PLL_WR_EN));
2092         r100_pll_errata_after_index(rdev);
2093         WREG32(RADEON_CLOCK_CNTL_DATA, v);
2094         r100_pll_errata_after_data(rdev);
2095 }
2096
2097 void r100_set_safe_registers(struct radeon_device *rdev)
2098 {
2099         if (ASIC_IS_RN50(rdev)) {
2100                 rdev->config.r100.reg_safe_bm = rn50_reg_safe_bm;
2101                 rdev->config.r100.reg_safe_bm_size = ARRAY_SIZE(rn50_reg_safe_bm);
2102         } else if (rdev->family < CHIP_R200) {
2103                 rdev->config.r100.reg_safe_bm = r100_reg_safe_bm;
2104                 rdev->config.r100.reg_safe_bm_size = ARRAY_SIZE(r100_reg_safe_bm);
2105         } else {
2106                 r200_set_safe_registers(rdev);
2107         }
2108 }
2109
2110 /*
2111  * Debugfs info
2112  */
2113 #if defined(CONFIG_DEBUG_FS)
2114 static int r100_debugfs_rbbm_info(struct seq_file *m, void *data)
2115 {
2116         struct drm_info_node *node = (struct drm_info_node *) m->private;
2117         struct drm_device *dev = node->minor->dev;
2118         struct radeon_device *rdev = dev->dev_private;
2119         uint32_t reg, value;
2120         unsigned i;
2121
2122         seq_printf(m, "RBBM_STATUS 0x%08x\n", RREG32(RADEON_RBBM_STATUS));
2123         seq_printf(m, "RBBM_CMDFIFO_STAT 0x%08x\n", RREG32(0xE7C));
2124         seq_printf(m, "CP_STAT 0x%08x\n", RREG32(RADEON_CP_STAT));
2125         for (i = 0; i < 64; i++) {
2126                 WREG32(RADEON_RBBM_CMDFIFO_ADDR, i | 0x100);
2127                 reg = (RREG32(RADEON_RBBM_CMDFIFO_DATA) - 1) >> 2;
2128                 WREG32(RADEON_RBBM_CMDFIFO_ADDR, i);
2129                 value = RREG32(RADEON_RBBM_CMDFIFO_DATA);
2130                 seq_printf(m, "[0x%03X] 0x%04X=0x%08X\n", i, reg, value);
2131         }
2132         return 0;
2133 }
2134
2135 static int r100_debugfs_cp_ring_info(struct seq_file *m, void *data)
2136 {
2137         struct drm_info_node *node = (struct drm_info_node *) m->private;
2138         struct drm_device *dev = node->minor->dev;
2139         struct radeon_device *rdev = dev->dev_private;
2140         uint32_t rdp, wdp;
2141         unsigned count, i, j;
2142
2143         radeon_ring_free_size(rdev);
2144         rdp = RREG32(RADEON_CP_RB_RPTR);
2145         wdp = RREG32(RADEON_CP_RB_WPTR);
2146         count = (rdp + rdev->cp.ring_size - wdp) & rdev->cp.ptr_mask;
2147         seq_printf(m, "CP_STAT 0x%08x\n", RREG32(RADEON_CP_STAT));
2148         seq_printf(m, "CP_RB_WPTR 0x%08x\n", wdp);
2149         seq_printf(m, "CP_RB_RPTR 0x%08x\n", rdp);
2150         seq_printf(m, "%u free dwords in ring\n", rdev->cp.ring_free_dw);
2151         seq_printf(m, "%u dwords in ring\n", count);
2152         for (j = 0; j <= count; j++) {
2153                 i = (rdp + j) & rdev->cp.ptr_mask;
2154                 seq_printf(m, "r[%04d]=0x%08x\n", i, rdev->cp.ring[i]);
2155         }
2156         return 0;
2157 }
2158
2159
2160 static int r100_debugfs_cp_csq_fifo(struct seq_file *m, void *data)
2161 {
2162         struct drm_info_node *node = (struct drm_info_node *) m->private;
2163         struct drm_device *dev = node->minor->dev;
2164         struct radeon_device *rdev = dev->dev_private;
2165         uint32_t csq_stat, csq2_stat, tmp;
2166         unsigned r_rptr, r_wptr, ib1_rptr, ib1_wptr, ib2_rptr, ib2_wptr;
2167         unsigned i;
2168
2169         seq_printf(m, "CP_STAT 0x%08x\n", RREG32(RADEON_CP_STAT));
2170         seq_printf(m, "CP_CSQ_MODE 0x%08x\n", RREG32(RADEON_CP_CSQ_MODE));
2171         csq_stat = RREG32(RADEON_CP_CSQ_STAT);
2172         csq2_stat = RREG32(RADEON_CP_CSQ2_STAT);
2173         r_rptr = (csq_stat >> 0) & 0x3ff;
2174         r_wptr = (csq_stat >> 10) & 0x3ff;
2175         ib1_rptr = (csq_stat >> 20) & 0x3ff;
2176         ib1_wptr = (csq2_stat >> 0) & 0x3ff;
2177         ib2_rptr = (csq2_stat >> 10) & 0x3ff;
2178         ib2_wptr = (csq2_stat >> 20) & 0x3ff;
2179         seq_printf(m, "CP_CSQ_STAT 0x%08x\n", csq_stat);
2180         seq_printf(m, "CP_CSQ2_STAT 0x%08x\n", csq2_stat);
2181         seq_printf(m, "Ring rptr %u\n", r_rptr);
2182         seq_printf(m, "Ring wptr %u\n", r_wptr);
2183         seq_printf(m, "Indirect1 rptr %u\n", ib1_rptr);
2184         seq_printf(m, "Indirect1 wptr %u\n", ib1_wptr);
2185         seq_printf(m, "Indirect2 rptr %u\n", ib2_rptr);
2186         seq_printf(m, "Indirect2 wptr %u\n", ib2_wptr);
2187         /* FIXME: 0, 128, 640 depends on fifo setup see cp_init_kms
2188          * 128 = indirect1_start * 8 & 640 = indirect2_start * 8 */
2189         seq_printf(m, "Ring fifo:\n");
2190         for (i = 0; i < 256; i++) {
2191                 WREG32(RADEON_CP_CSQ_ADDR, i << 2);
2192                 tmp = RREG32(RADEON_CP_CSQ_DATA);
2193                 seq_printf(m, "rfifo[%04d]=0x%08X\n", i, tmp);
2194         }
2195         seq_printf(m, "Indirect1 fifo:\n");
2196         for (i = 256; i <= 512; i++) {
2197                 WREG32(RADEON_CP_CSQ_ADDR, i << 2);
2198                 tmp = RREG32(RADEON_CP_CSQ_DATA);
2199                 seq_printf(m, "ib1fifo[%04d]=0x%08X\n", i, tmp);
2200         }
2201         seq_printf(m, "Indirect2 fifo:\n");
2202         for (i = 640; i < ib1_wptr; i++) {
2203                 WREG32(RADEON_CP_CSQ_ADDR, i << 2);
2204                 tmp = RREG32(RADEON_CP_CSQ_DATA);
2205                 seq_printf(m, "ib2fifo[%04d]=0x%08X\n", i, tmp);
2206         }
2207         return 0;
2208 }
2209
2210 static int r100_debugfs_mc_info(struct seq_file *m, void *data)
2211 {
2212         struct drm_info_node *node = (struct drm_info_node *) m->private;
2213         struct drm_device *dev = node->minor->dev;
2214         struct radeon_device *rdev = dev->dev_private;
2215         uint32_t tmp;
2216
2217         tmp = RREG32(RADEON_CONFIG_MEMSIZE);
2218         seq_printf(m, "CONFIG_MEMSIZE 0x%08x\n", tmp);
2219         tmp = RREG32(RADEON_MC_FB_LOCATION);
2220         seq_printf(m, "MC_FB_LOCATION 0x%08x\n", tmp);
2221         tmp = RREG32(RADEON_BUS_CNTL);
2222         seq_printf(m, "BUS_CNTL 0x%08x\n", tmp);
2223         tmp = RREG32(RADEON_MC_AGP_LOCATION);
2224         seq_printf(m, "MC_AGP_LOCATION 0x%08x\n", tmp);
2225         tmp = RREG32(RADEON_AGP_BASE);
2226         seq_printf(m, "AGP_BASE 0x%08x\n", tmp);
2227         tmp = RREG32(RADEON_HOST_PATH_CNTL);
2228         seq_printf(m, "HOST_PATH_CNTL 0x%08x\n", tmp);
2229         tmp = RREG32(0x01D0);
2230         seq_printf(m, "AIC_CTRL 0x%08x\n", tmp);
2231         tmp = RREG32(RADEON_AIC_LO_ADDR);
2232         seq_printf(m, "AIC_LO_ADDR 0x%08x\n", tmp);
2233         tmp = RREG32(RADEON_AIC_HI_ADDR);
2234         seq_printf(m, "AIC_HI_ADDR 0x%08x\n", tmp);
2235         tmp = RREG32(0x01E4);
2236         seq_printf(m, "AIC_TLB_ADDR 0x%08x\n", tmp);
2237         return 0;
2238 }
2239
2240 static struct drm_info_list r100_debugfs_rbbm_list[] = {
2241         {"r100_rbbm_info", r100_debugfs_rbbm_info, 0, NULL},
2242 };
2243
2244 static struct drm_info_list r100_debugfs_cp_list[] = {
2245         {"r100_cp_ring_info", r100_debugfs_cp_ring_info, 0, NULL},
2246         {"r100_cp_csq_fifo", r100_debugfs_cp_csq_fifo, 0, NULL},
2247 };
2248
2249 static struct drm_info_list r100_debugfs_mc_info_list[] = {
2250         {"r100_mc_info", r100_debugfs_mc_info, 0, NULL},
2251 };
2252 #endif
2253
2254 int r100_debugfs_rbbm_init(struct radeon_device *rdev)
2255 {
2256 #if defined(CONFIG_DEBUG_FS)
2257         return radeon_debugfs_add_files(rdev, r100_debugfs_rbbm_list, 1);
2258 #else
2259         return 0;
2260 #endif
2261 }
2262
2263 int r100_debugfs_cp_init(struct radeon_device *rdev)
2264 {
2265 #if defined(CONFIG_DEBUG_FS)
2266         return radeon_debugfs_add_files(rdev, r100_debugfs_cp_list, 2);
2267 #else
2268         return 0;
2269 #endif
2270 }
2271
2272 int r100_debugfs_mc_info_init(struct radeon_device *rdev)
2273 {
2274 #if defined(CONFIG_DEBUG_FS)
2275         return radeon_debugfs_add_files(rdev, r100_debugfs_mc_info_list, 1);
2276 #else
2277         return 0;
2278 #endif
2279 }
2280
2281 int r100_set_surface_reg(struct radeon_device *rdev, int reg,
2282                          uint32_t tiling_flags, uint32_t pitch,
2283                          uint32_t offset, uint32_t obj_size)
2284 {
2285         int surf_index = reg * 16;
2286         int flags = 0;
2287
2288         /* r100/r200 divide by 16 */
2289         if (rdev->family < CHIP_R300)
2290                 flags = pitch / 16;
2291         else
2292                 flags = pitch / 8;
2293
2294         if (rdev->family <= CHIP_RS200) {
2295                 if ((tiling_flags & (RADEON_TILING_MACRO|RADEON_TILING_MICRO))
2296                                  == (RADEON_TILING_MACRO|RADEON_TILING_MICRO))
2297                         flags |= RADEON_SURF_TILE_COLOR_BOTH;
2298                 if (tiling_flags & RADEON_TILING_MACRO)
2299                         flags |= RADEON_SURF_TILE_COLOR_MACRO;
2300         } else if (rdev->family <= CHIP_RV280) {
2301                 if (tiling_flags & (RADEON_TILING_MACRO))
2302                         flags |= R200_SURF_TILE_COLOR_MACRO;
2303                 if (tiling_flags & RADEON_TILING_MICRO)
2304                         flags |= R200_SURF_TILE_COLOR_MICRO;
2305         } else {
2306                 if (tiling_flags & RADEON_TILING_MACRO)
2307                         flags |= R300_SURF_TILE_MACRO;
2308                 if (tiling_flags & RADEON_TILING_MICRO)
2309                         flags |= R300_SURF_TILE_MICRO;
2310         }
2311
2312         if (tiling_flags & RADEON_TILING_SWAP_16BIT)
2313                 flags |= RADEON_SURF_AP0_SWP_16BPP | RADEON_SURF_AP1_SWP_16BPP;
2314         if (tiling_flags & RADEON_TILING_SWAP_32BIT)
2315                 flags |= RADEON_SURF_AP0_SWP_32BPP | RADEON_SURF_AP1_SWP_32BPP;
2316
2317         DRM_DEBUG("writing surface %d %d %x %x\n", reg, flags, offset, offset+obj_size-1);
2318         WREG32(RADEON_SURFACE0_INFO + surf_index, flags);
2319         WREG32(RADEON_SURFACE0_LOWER_BOUND + surf_index, offset);
2320         WREG32(RADEON_SURFACE0_UPPER_BOUND + surf_index, offset + obj_size - 1);
2321         return 0;
2322 }
2323
2324 void r100_clear_surface_reg(struct radeon_device *rdev, int reg)
2325 {
2326         int surf_index = reg * 16;
2327         WREG32(RADEON_SURFACE0_INFO + surf_index, 0);
2328 }
2329
2330 void r100_bandwidth_update(struct radeon_device *rdev)
2331 {
2332         fixed20_12 trcd_ff, trp_ff, tras_ff, trbs_ff, tcas_ff;
2333         fixed20_12 sclk_ff, mclk_ff, sclk_eff_ff, sclk_delay_ff;
2334         fixed20_12 peak_disp_bw, mem_bw, pix_clk, pix_clk2, temp_ff, crit_point_ff;
2335         uint32_t temp, data, mem_trcd, mem_trp, mem_tras;
2336         fixed20_12 memtcas_ff[8] = {
2337                 fixed_init(1),
2338                 fixed_init(2),
2339                 fixed_init(3),
2340                 fixed_init(0),
2341                 fixed_init_half(1),
2342                 fixed_init_half(2),
2343                 fixed_init(0),
2344         };
2345         fixed20_12 memtcas_rs480_ff[8] = {
2346                 fixed_init(0),
2347                 fixed_init(1),
2348                 fixed_init(2),
2349                 fixed_init(3),
2350                 fixed_init(0),
2351                 fixed_init_half(1),
2352                 fixed_init_half(2),
2353                 fixed_init_half(3),
2354         };
2355         fixed20_12 memtcas2_ff[8] = {
2356                 fixed_init(0),
2357                 fixed_init(1),
2358                 fixed_init(2),
2359                 fixed_init(3),
2360                 fixed_init(4),
2361                 fixed_init(5),
2362                 fixed_init(6),
2363                 fixed_init(7),
2364         };
2365         fixed20_12 memtrbs[8] = {
2366                 fixed_init(1),
2367                 fixed_init_half(1),
2368                 fixed_init(2),
2369                 fixed_init_half(2),
2370                 fixed_init(3),
2371                 fixed_init_half(3),
2372                 fixed_init(4),
2373                 fixed_init_half(4)
2374         };
2375         fixed20_12 memtrbs_r4xx[8] = {
2376                 fixed_init(4),
2377                 fixed_init(5),
2378                 fixed_init(6),
2379                 fixed_init(7),
2380                 fixed_init(8),
2381                 fixed_init(9),
2382                 fixed_init(10),
2383                 fixed_init(11)
2384         };
2385         fixed20_12 min_mem_eff;
2386         fixed20_12 mc_latency_sclk, mc_latency_mclk, k1;
2387         fixed20_12 cur_latency_mclk, cur_latency_sclk;
2388         fixed20_12 disp_latency, disp_latency_overhead, disp_drain_rate,
2389                 disp_drain_rate2, read_return_rate;
2390         fixed20_12 time_disp1_drop_priority;
2391         int c;
2392         int cur_size = 16;       /* in octawords */
2393         int critical_point = 0, critical_point2;
2394 /*      uint32_t read_return_rate, time_disp1_drop_priority; */
2395         int stop_req, max_stop_req;
2396         struct drm_display_mode *mode1 = NULL;
2397         struct drm_display_mode *mode2 = NULL;
2398         uint32_t pixel_bytes1 = 0;
2399         uint32_t pixel_bytes2 = 0;
2400
2401         radeon_update_display_priority(rdev);
2402
2403         if (rdev->mode_info.crtcs[0]->base.enabled) {
2404                 mode1 = &rdev->mode_info.crtcs[0]->base.mode;
2405                 pixel_bytes1 = rdev->mode_info.crtcs[0]->base.fb->bits_per_pixel / 8;
2406         }
2407         if (!(rdev->flags & RADEON_SINGLE_CRTC)) {
2408                 if (rdev->mode_info.crtcs[1]->base.enabled) {
2409                         mode2 = &rdev->mode_info.crtcs[1]->base.mode;
2410                         pixel_bytes2 = rdev->mode_info.crtcs[1]->base.fb->bits_per_pixel / 8;
2411                 }
2412         }
2413
2414         min_mem_eff.full = rfixed_const_8(0);
2415         /* get modes */
2416         if ((rdev->disp_priority == 2) && ASIC_IS_R300(rdev)) {
2417                 uint32_t mc_init_misc_lat_timer = RREG32(R300_MC_INIT_MISC_LAT_TIMER);
2418                 mc_init_misc_lat_timer &= ~(R300_MC_DISP1R_INIT_LAT_MASK << R300_MC_DISP1R_INIT_LAT_SHIFT);
2419                 mc_init_misc_lat_timer &= ~(R300_MC_DISP0R_INIT_LAT_MASK << R300_MC_DISP0R_INIT_LAT_SHIFT);
2420                 /* check crtc enables */
2421                 if (mode2)
2422                         mc_init_misc_lat_timer |= (1 << R300_MC_DISP1R_INIT_LAT_SHIFT);
2423                 if (mode1)
2424                         mc_init_misc_lat_timer |= (1 << R300_MC_DISP0R_INIT_LAT_SHIFT);
2425                 WREG32(R300_MC_INIT_MISC_LAT_TIMER, mc_init_misc_lat_timer);
2426         }
2427
2428         /*
2429          * determine is there is enough bw for current mode
2430          */
2431         sclk_ff = rdev->pm.sclk;
2432         mclk_ff = rdev->pm.mclk;
2433
2434         temp = (rdev->mc.vram_width / 8) * (rdev->mc.vram_is_ddr ? 2 : 1);
2435         temp_ff.full = rfixed_const(temp);
2436         mem_bw.full = rfixed_mul(mclk_ff, temp_ff);
2437
2438         pix_clk.full = 0;
2439         pix_clk2.full = 0;
2440         peak_disp_bw.full = 0;
2441         if (mode1) {
2442                 temp_ff.full = rfixed_const(1000);
2443                 pix_clk.full = rfixed_const(mode1->clock); /* convert to fixed point */
2444                 pix_clk.full = rfixed_div(pix_clk, temp_ff);
2445                 temp_ff.full = rfixed_const(pixel_bytes1);
2446                 peak_disp_bw.full += rfixed_mul(pix_clk, temp_ff);
2447         }
2448         if (mode2) {
2449                 temp_ff.full = rfixed_const(1000);
2450                 pix_clk2.full = rfixed_const(mode2->clock); /* convert to fixed point */
2451                 pix_clk2.full = rfixed_div(pix_clk2, temp_ff);
2452                 temp_ff.full = rfixed_const(pixel_bytes2);
2453                 peak_disp_bw.full += rfixed_mul(pix_clk2, temp_ff);
2454         }
2455
2456         mem_bw.full = rfixed_mul(mem_bw, min_mem_eff);
2457         if (peak_disp_bw.full >= mem_bw.full) {
2458                 DRM_ERROR("You may not have enough display bandwidth for current mode\n"
2459                           "If you have flickering problem, try to lower resolution, refresh rate, or color depth\n");
2460         }
2461
2462         /*  Get values from the EXT_MEM_CNTL register...converting its contents. */
2463         temp = RREG32(RADEON_MEM_TIMING_CNTL);
2464         if ((rdev->family == CHIP_RV100) || (rdev->flags & RADEON_IS_IGP)) { /* RV100, M6, IGPs */
2465                 mem_trcd = ((temp >> 2) & 0x3) + 1;
2466                 mem_trp  = ((temp & 0x3)) + 1;
2467                 mem_tras = ((temp & 0x70) >> 4) + 1;
2468         } else if (rdev->family == CHIP_R300 ||
2469                    rdev->family == CHIP_R350) { /* r300, r350 */
2470                 mem_trcd = (temp & 0x7) + 1;
2471                 mem_trp = ((temp >> 8) & 0x7) + 1;
2472                 mem_tras = ((temp >> 11) & 0xf) + 4;
2473         } else if (rdev->family == CHIP_RV350 ||
2474                    rdev->family <= CHIP_RV380) {
2475                 /* rv3x0 */
2476                 mem_trcd = (temp & 0x7) + 3;
2477                 mem_trp = ((temp >> 8) & 0x7) + 3;
2478                 mem_tras = ((temp >> 11) & 0xf) + 6;
2479         } else if (rdev->family == CHIP_R420 ||
2480                    rdev->family == CHIP_R423 ||
2481                    rdev->family == CHIP_RV410) {
2482                 /* r4xx */
2483                 mem_trcd = (temp & 0xf) + 3;
2484                 if (mem_trcd > 15)
2485                         mem_trcd = 15;
2486                 mem_trp = ((temp >> 8) & 0xf) + 3;
2487                 if (mem_trp > 15)
2488                         mem_trp = 15;
2489                 mem_tras = ((temp >> 12) & 0x1f) + 6;
2490                 if (mem_tras > 31)
2491                         mem_tras = 31;
2492         } else { /* RV200, R200 */
2493                 mem_trcd = (temp & 0x7) + 1;
2494                 mem_trp = ((temp >> 8) & 0x7) + 1;
2495                 mem_tras = ((temp >> 12) & 0xf) + 4;
2496         }
2497         /* convert to FF */
2498         trcd_ff.full = rfixed_const(mem_trcd);
2499         trp_ff.full = rfixed_const(mem_trp);
2500         tras_ff.full = rfixed_const(mem_tras);
2501
2502         /* Get values from the MEM_SDRAM_MODE_REG register...converting its */
2503         temp = RREG32(RADEON_MEM_SDRAM_MODE_REG);
2504         data = (temp & (7 << 20)) >> 20;
2505         if ((rdev->family == CHIP_RV100) || rdev->flags & RADEON_IS_IGP) {
2506                 if (rdev->family == CHIP_RS480) /* don't think rs400 */
2507                         tcas_ff = memtcas_rs480_ff[data];
2508                 else
2509                         tcas_ff = memtcas_ff[data];
2510         } else
2511                 tcas_ff = memtcas2_ff[data];
2512
2513         if (rdev->family == CHIP_RS400 ||
2514             rdev->family == CHIP_RS480) {
2515                 /* extra cas latency stored in bits 23-25 0-4 clocks */
2516                 data = (temp >> 23) & 0x7;
2517                 if (data < 5)
2518                         tcas_ff.full += rfixed_const(data);
2519         }
2520
2521         if (ASIC_IS_R300(rdev) && !(rdev->flags & RADEON_IS_IGP)) {
2522                 /* on the R300, Tcas is included in Trbs.
2523                  */
2524                 temp = RREG32(RADEON_MEM_CNTL);
2525                 data = (R300_MEM_NUM_CHANNELS_MASK & temp);
2526                 if (data == 1) {
2527                         if (R300_MEM_USE_CD_CH_ONLY & temp) {
2528                                 temp = RREG32(R300_MC_IND_INDEX);
2529                                 temp &= ~R300_MC_IND_ADDR_MASK;
2530                                 temp |= R300_MC_READ_CNTL_CD_mcind;
2531                                 WREG32(R300_MC_IND_INDEX, temp);
2532                                 temp = RREG32(R300_MC_IND_DATA);
2533                                 data = (R300_MEM_RBS_POSITION_C_MASK & temp);
2534                         } else {
2535                                 temp = RREG32(R300_MC_READ_CNTL_AB);
2536                                 data = (R300_MEM_RBS_POSITION_A_MASK & temp);
2537                         }
2538                 } else {
2539                         temp = RREG32(R300_MC_READ_CNTL_AB);
2540                         data = (R300_MEM_RBS_POSITION_A_MASK & temp);
2541                 }
2542                 if (rdev->family == CHIP_RV410 ||
2543                     rdev->family == CHIP_R420 ||
2544                     rdev->family == CHIP_R423)
2545                         trbs_ff = memtrbs_r4xx[data];
2546                 else
2547                         trbs_ff = memtrbs[data];
2548                 tcas_ff.full += trbs_ff.full;
2549         }
2550
2551         sclk_eff_ff.full = sclk_ff.full;
2552
2553         if (rdev->flags & RADEON_IS_AGP) {
2554                 fixed20_12 agpmode_ff;
2555                 agpmode_ff.full = rfixed_const(radeon_agpmode);
2556                 temp_ff.full = rfixed_const_666(16);
2557                 sclk_eff_ff.full -= rfixed_mul(agpmode_ff, temp_ff);
2558         }
2559         /* TODO PCIE lanes may affect this - agpmode == 16?? */
2560
2561         if (ASIC_IS_R300(rdev)) {
2562                 sclk_delay_ff.full = rfixed_const(250);
2563         } else {
2564                 if ((rdev->family == CHIP_RV100) ||
2565                     rdev->flags & RADEON_IS_IGP) {
2566                         if (rdev->mc.vram_is_ddr)
2567                                 sclk_delay_ff.full = rfixed_const(41);
2568                         else
2569                                 sclk_delay_ff.full = rfixed_const(33);
2570                 } else {
2571                         if (rdev->mc.vram_width == 128)
2572                                 sclk_delay_ff.full = rfixed_const(57);
2573                         else
2574                                 sclk_delay_ff.full = rfixed_const(41);
2575                 }
2576         }
2577
2578         mc_latency_sclk.full = rfixed_div(sclk_delay_ff, sclk_eff_ff);
2579
2580         if (rdev->mc.vram_is_ddr) {
2581                 if (rdev->mc.vram_width == 32) {
2582                         k1.full = rfixed_const(40);
2583                         c  = 3;
2584                 } else {
2585                         k1.full = rfixed_const(20);
2586                         c  = 1;
2587                 }
2588         } else {
2589                 k1.full = rfixed_const(40);
2590                 c  = 3;
2591         }
2592
2593         temp_ff.full = rfixed_const(2);
2594         mc_latency_mclk.full = rfixed_mul(trcd_ff, temp_ff);
2595         temp_ff.full = rfixed_const(c);
2596         mc_latency_mclk.full += rfixed_mul(tcas_ff, temp_ff);
2597         temp_ff.full = rfixed_const(4);
2598         mc_latency_mclk.full += rfixed_mul(tras_ff, temp_ff);
2599         mc_latency_mclk.full += rfixed_mul(trp_ff, temp_ff);
2600         mc_latency_mclk.full += k1.full;
2601
2602         mc_latency_mclk.full = rfixed_div(mc_latency_mclk, mclk_ff);
2603         mc_latency_mclk.full += rfixed_div(temp_ff, sclk_eff_ff);
2604
2605         /*
2606           HW cursor time assuming worst case of full size colour cursor.
2607         */
2608         temp_ff.full = rfixed_const((2 * (cur_size - (rdev->mc.vram_is_ddr + 1))));
2609         temp_ff.full += trcd_ff.full;
2610         if (temp_ff.full < tras_ff.full)
2611                 temp_ff.full = tras_ff.full;
2612         cur_latency_mclk.full = rfixed_div(temp_ff, mclk_ff);
2613
2614         temp_ff.full = rfixed_const(cur_size);
2615         cur_latency_sclk.full = rfixed_div(temp_ff, sclk_eff_ff);
2616         /*
2617           Find the total latency for the display data.
2618         */
2619         disp_latency_overhead.full = rfixed_const(8);
2620         disp_latency_overhead.full = rfixed_div(disp_latency_overhead, sclk_ff);
2621         mc_latency_mclk.full += disp_latency_overhead.full + cur_latency_mclk.full;
2622         mc_latency_sclk.full += disp_latency_overhead.full + cur_latency_sclk.full;
2623
2624         if (mc_latency_mclk.full > mc_latency_sclk.full)
2625                 disp_latency.full = mc_latency_mclk.full;
2626         else
2627                 disp_latency.full = mc_latency_sclk.full;
2628
2629         /* setup Max GRPH_STOP_REQ default value */
2630         if (ASIC_IS_RV100(rdev))
2631                 max_stop_req = 0x5c;
2632         else
2633                 max_stop_req = 0x7c;
2634
2635         if (mode1) {
2636                 /*  CRTC1
2637                     Set GRPH_BUFFER_CNTL register using h/w defined optimal values.
2638                     GRPH_STOP_REQ <= MIN[ 0x7C, (CRTC_H_DISP + 1) * (bit depth) / 0x10 ]
2639                 */
2640                 stop_req = mode1->hdisplay * pixel_bytes1 / 16;
2641
2642                 if (stop_req > max_stop_req)
2643                         stop_req = max_stop_req;
2644
2645                 /*
2646                   Find the drain rate of the display buffer.
2647                 */
2648                 temp_ff.full = rfixed_const((16/pixel_bytes1));
2649                 disp_drain_rate.full = rfixed_div(pix_clk, temp_ff);
2650
2651                 /*
2652                   Find the critical point of the display buffer.
2653                 */
2654                 crit_point_ff.full = rfixed_mul(disp_drain_rate, disp_latency);
2655                 crit_point_ff.full += rfixed_const_half(0);
2656
2657                 critical_point = rfixed_trunc(crit_point_ff);
2658
2659                 if (rdev->disp_priority == 2) {
2660                         critical_point = 0;
2661                 }
2662
2663                 /*
2664                   The critical point should never be above max_stop_req-4.  Setting
2665                   GRPH_CRITICAL_CNTL = 0 will thus force high priority all the time.
2666                 */
2667                 if (max_stop_req - critical_point < 4)
2668                         critical_point = 0;
2669
2670                 if (critical_point == 0 && mode2 && rdev->family == CHIP_R300) {
2671                         /* some R300 cards have problem with this set to 0, when CRTC2 is enabled.*/
2672                         critical_point = 0x10;
2673                 }
2674
2675                 temp = RREG32(RADEON_GRPH_BUFFER_CNTL);
2676                 temp &= ~(RADEON_GRPH_STOP_REQ_MASK);
2677                 temp |= (stop_req << RADEON_GRPH_STOP_REQ_SHIFT);
2678                 temp &= ~(RADEON_GRPH_START_REQ_MASK);
2679                 if ((rdev->family == CHIP_R350) &&
2680                     (stop_req > 0x15)) {
2681                         stop_req -= 0x10;
2682                 }
2683                 temp |= (stop_req << RADEON_GRPH_START_REQ_SHIFT);
2684                 temp |= RADEON_GRPH_BUFFER_SIZE;
2685                 temp &= ~(RADEON_GRPH_CRITICAL_CNTL   |
2686                           RADEON_GRPH_CRITICAL_AT_SOF |
2687                           RADEON_GRPH_STOP_CNTL);
2688                 /*
2689                   Write the result into the register.
2690                 */
2691                 WREG32(RADEON_GRPH_BUFFER_CNTL, ((temp & ~RADEON_GRPH_CRITICAL_POINT_MASK) |
2692                                                        (critical_point << RADEON_GRPH_CRITICAL_POINT_SHIFT)));
2693
2694 #if 0
2695                 if ((rdev->family == CHIP_RS400) ||
2696                     (rdev->family == CHIP_RS480)) {
2697                         /* attempt to program RS400 disp regs correctly ??? */
2698                         temp = RREG32(RS400_DISP1_REG_CNTL);
2699                         temp &= ~(RS400_DISP1_START_REQ_LEVEL_MASK |
2700                                   RS400_DISP1_STOP_REQ_LEVEL_MASK);
2701                         WREG32(RS400_DISP1_REQ_CNTL1, (temp |
2702                                                        (critical_point << RS400_DISP1_START_REQ_LEVEL_SHIFT) |
2703                                                        (critical_point << RS400_DISP1_STOP_REQ_LEVEL_SHIFT)));
2704                         temp = RREG32(RS400_DMIF_MEM_CNTL1);
2705                         temp &= ~(RS400_DISP1_CRITICAL_POINT_START_MASK |
2706                                   RS400_DISP1_CRITICAL_POINT_STOP_MASK);
2707                         WREG32(RS400_DMIF_MEM_CNTL1, (temp |
2708                                                       (critical_point << RS400_DISP1_CRITICAL_POINT_START_SHIFT) |
2709                                                       (critical_point << RS400_DISP1_CRITICAL_POINT_STOP_SHIFT)));
2710                 }
2711 #endif
2712
2713                 DRM_DEBUG("GRPH_BUFFER_CNTL from to %x\n",
2714                           /*      (unsigned int)info->SavedReg->grph_buffer_cntl, */
2715                           (unsigned int)RREG32(RADEON_GRPH_BUFFER_CNTL));
2716         }
2717
2718         if (mode2) {
2719                 u32 grph2_cntl;
2720                 stop_req = mode2->hdisplay * pixel_bytes2 / 16;
2721
2722                 if (stop_req > max_stop_req)
2723                         stop_req = max_stop_req;
2724
2725                 /*
2726                   Find the drain rate of the display buffer.
2727                 */
2728                 temp_ff.full = rfixed_const((16/pixel_bytes2));
2729                 disp_drain_rate2.full = rfixed_div(pix_clk2, temp_ff);
2730
2731                 grph2_cntl = RREG32(RADEON_GRPH2_BUFFER_CNTL);
2732                 grph2_cntl &= ~(RADEON_GRPH_STOP_REQ_MASK);
2733                 grph2_cntl |= (stop_req << RADEON_GRPH_STOP_REQ_SHIFT);
2734                 grph2_cntl &= ~(RADEON_GRPH_START_REQ_MASK);
2735                 if ((rdev->family == CHIP_R350) &&
2736                     (stop_req > 0x15)) {
2737                         stop_req -= 0x10;
2738                 }
2739                 grph2_cntl |= (stop_req << RADEON_GRPH_START_REQ_SHIFT);
2740                 grph2_cntl |= RADEON_GRPH_BUFFER_SIZE;
2741                 grph2_cntl &= ~(RADEON_GRPH_CRITICAL_CNTL   |
2742                           RADEON_GRPH_CRITICAL_AT_SOF |
2743                           RADEON_GRPH_STOP_CNTL);
2744
2745                 if ((rdev->family == CHIP_RS100) ||
2746                     (rdev->family == CHIP_RS200))
2747                         critical_point2 = 0;
2748                 else {
2749                         temp = (rdev->mc.vram_width * rdev->mc.vram_is_ddr + 1)/128;
2750                         temp_ff.full = rfixed_const(temp);
2751                         temp_ff.full = rfixed_mul(mclk_ff, temp_ff);
2752                         if (sclk_ff.full < temp_ff.full)
2753                                 temp_ff.full = sclk_ff.full;
2754
2755                         read_return_rate.full = temp_ff.full;
2756
2757                         if (mode1) {
2758                                 temp_ff.full = read_return_rate.full - disp_drain_rate.full;
2759                                 time_disp1_drop_priority.full = rfixed_div(crit_point_ff, temp_ff);
2760                         } else {
2761                                 time_disp1_drop_priority.full = 0;
2762                         }
2763                         crit_point_ff.full = disp_latency.full + time_disp1_drop_priority.full + disp_latency.full;
2764                         crit_point_ff.full = rfixed_mul(crit_point_ff, disp_drain_rate2);
2765                         crit_point_ff.full += rfixed_const_half(0);
2766
2767                         critical_point2 = rfixed_trunc(crit_point_ff);
2768
2769                         if (rdev->disp_priority == 2) {
2770                                 critical_point2 = 0;
2771                         }
2772
2773                         if (max_stop_req - critical_point2 < 4)
2774                                 critical_point2 = 0;
2775
2776                 }
2777
2778                 if (critical_point2 == 0 && rdev->family == CHIP_R300) {
2779                         /* some R300 cards have problem with this set to 0 */
2780                         critical_point2 = 0x10;
2781                 }
2782
2783                 WREG32(RADEON_GRPH2_BUFFER_CNTL, ((grph2_cntl & ~RADEON_GRPH_CRITICAL_POINT_MASK) |
2784                                                   (critical_point2 << RADEON_GRPH_CRITICAL_POINT_SHIFT)));
2785
2786                 if ((rdev->family == CHIP_RS400) ||
2787                     (rdev->family == CHIP_RS480)) {
2788 #if 0
2789                         /* attempt to program RS400 disp2 regs correctly ??? */
2790                         temp = RREG32(RS400_DISP2_REQ_CNTL1);
2791                         temp &= ~(RS400_DISP2_START_REQ_LEVEL_MASK |
2792                                   RS400_DISP2_STOP_REQ_LEVEL_MASK);
2793                         WREG32(RS400_DISP2_REQ_CNTL1, (temp |
2794                                                        (critical_point2 << RS400_DISP1_START_REQ_LEVEL_SHIFT) |
2795                                                        (critical_point2 << RS400_DISP1_STOP_REQ_LEVEL_SHIFT)));
2796                         temp = RREG32(RS400_DISP2_REQ_CNTL2);
2797                         temp &= ~(RS400_DISP2_CRITICAL_POINT_START_MASK |
2798                                   RS400_DISP2_CRITICAL_POINT_STOP_MASK);
2799                         WREG32(RS400_DISP2_REQ_CNTL2, (temp |
2800                                                        (critical_point2 << RS400_DISP2_CRITICAL_POINT_START_SHIFT) |
2801                                                        (critical_point2 << RS400_DISP2_CRITICAL_POINT_STOP_SHIFT)));
2802 #endif
2803                         WREG32(RS400_DISP2_REQ_CNTL1, 0x105DC1CC);
2804                         WREG32(RS400_DISP2_REQ_CNTL2, 0x2749D000);
2805                         WREG32(RS400_DMIF_MEM_CNTL1,  0x29CA71DC);
2806                         WREG32(RS400_DISP1_REQ_CNTL1, 0x28FBC3AC);
2807                 }
2808
2809                 DRM_DEBUG("GRPH2_BUFFER_CNTL from to %x\n",
2810                           (unsigned int)RREG32(RADEON_GRPH2_BUFFER_CNTL));
2811         }
2812 }
2813
2814 static inline void r100_cs_track_texture_print(struct r100_cs_track_texture *t)
2815 {
2816         DRM_ERROR("pitch                      %d\n", t->pitch);
2817         DRM_ERROR("use_pitch                  %d\n", t->use_pitch);
2818         DRM_ERROR("width                      %d\n", t->width);
2819         DRM_ERROR("width_11                   %d\n", t->width_11);
2820         DRM_ERROR("height                     %d\n", t->height);
2821         DRM_ERROR("height_11                  %d\n", t->height_11);
2822         DRM_ERROR("num levels                 %d\n", t->num_levels);
2823         DRM_ERROR("depth                      %d\n", t->txdepth);
2824         DRM_ERROR("bpp                        %d\n", t->cpp);
2825         DRM_ERROR("coordinate type            %d\n", t->tex_coord_type);
2826         DRM_ERROR("width round to power of 2  %d\n", t->roundup_w);
2827         DRM_ERROR("height round to power of 2 %d\n", t->roundup_h);
2828         DRM_ERROR("compress format            %d\n", t->compress_format);
2829 }
2830
2831 static int r100_cs_track_cube(struct radeon_device *rdev,
2832                               struct r100_cs_track *track, unsigned idx)
2833 {
2834         unsigned face, w, h;
2835         struct radeon_bo *cube_robj;
2836         unsigned long size;
2837
2838         for (face = 0; face < 5; face++) {
2839                 cube_robj = track->textures[idx].cube_info[face].robj;
2840                 w = track->textures[idx].cube_info[face].width;
2841                 h = track->textures[idx].cube_info[face].height;
2842
2843                 size = w * h;
2844                 size *= track->textures[idx].cpp;
2845
2846                 size += track->textures[idx].cube_info[face].offset;
2847
2848                 if (size > radeon_bo_size(cube_robj)) {
2849                         DRM_ERROR("Cube texture offset greater than object size %lu %lu\n",
2850                                   size, radeon_bo_size(cube_robj));
2851                         r100_cs_track_texture_print(&track->textures[idx]);
2852                         return -1;
2853                 }
2854         }
2855         return 0;
2856 }
2857
2858 static int r100_track_compress_size(int compress_format, int w, int h)
2859 {
2860         int block_width, block_height, block_bytes;
2861         int wblocks, hblocks;
2862         int min_wblocks;
2863         int sz;
2864
2865         block_width = 4;
2866         block_height = 4;
2867
2868         switch (compress_format) {
2869         case R100_TRACK_COMP_DXT1:
2870                 block_bytes = 8;
2871                 min_wblocks = 4;
2872                 break;
2873         default:
2874         case R100_TRACK_COMP_DXT35:
2875                 block_bytes = 16;
2876                 min_wblocks = 2;
2877                 break;
2878         }
2879
2880         hblocks = (h + block_height - 1) / block_height;
2881         wblocks = (w + block_width - 1) / block_width;
2882         if (wblocks < min_wblocks)
2883                 wblocks = min_wblocks;
2884         sz = wblocks * hblocks * block_bytes;
2885         return sz;
2886 }
2887
2888 static int r100_cs_track_texture_check(struct radeon_device *rdev,
2889                                        struct r100_cs_track *track)
2890 {
2891         struct radeon_bo *robj;
2892         unsigned long size;
2893         unsigned u, i, w, h;
2894         int ret;
2895
2896         for (u = 0; u < track->num_texture; u++) {
2897                 if (!track->textures[u].enabled)
2898                         continue;
2899                 robj = track->textures[u].robj;
2900                 if (robj == NULL) {
2901                         DRM_ERROR("No texture bound to unit %u\n", u);
2902                         return -EINVAL;
2903                 }
2904                 size = 0;
2905                 for (i = 0; i <= track->textures[u].num_levels; i++) {
2906                         if (track->textures[u].use_pitch) {
2907                                 if (rdev->family < CHIP_R300)
2908                                         w = (track->textures[u].pitch / track->textures[u].cpp) / (1 << i);
2909                                 else
2910                                         w = track->textures[u].pitch / (1 << i);
2911                         } else {
2912                                 w = track->textures[u].width;
2913                                 if (rdev->family >= CHIP_RV515)
2914                                         w |= track->textures[u].width_11;
2915                                 w = w / (1 << i);
2916                                 if (track->textures[u].roundup_w)
2917                                         w = roundup_pow_of_two(w);
2918                         }
2919                         h = track->textures[u].height;
2920                         if (rdev->family >= CHIP_RV515)
2921                                 h |= track->textures[u].height_11;
2922                         h = h / (1 << i);
2923                         if (track->textures[u].roundup_h)
2924                                 h = roundup_pow_of_two(h);
2925                         if (track->textures[u].compress_format) {
2926
2927                                 size += r100_track_compress_size(track->textures[u].compress_format, w, h);
2928                                 /* compressed textures are block based */
2929                         } else
2930                                 size += w * h;
2931                 }
2932                 size *= track->textures[u].cpp;
2933
2934                 switch (track->textures[u].tex_coord_type) {
2935                 case 0:
2936                         break;
2937                 case 1:
2938                         size *= (1 << track->textures[u].txdepth);
2939                         break;
2940                 case 2:
2941                         if (track->separate_cube) {
2942                                 ret = r100_cs_track_cube(rdev, track, u);
2943                                 if (ret)
2944                                         return ret;
2945                         } else
2946                                 size *= 6;
2947                         break;
2948                 default:
2949                         DRM_ERROR("Invalid texture coordinate type %u for unit "
2950                                   "%u\n", track->textures[u].tex_coord_type, u);
2951                         return -EINVAL;
2952                 }
2953                 if (size > radeon_bo_size(robj)) {
2954                         DRM_ERROR("Texture of unit %u needs %lu bytes but is "
2955                                   "%lu\n", u, size, radeon_bo_size(robj));
2956                         r100_cs_track_texture_print(&track->textures[u]);
2957                         return -EINVAL;
2958                 }
2959         }
2960         return 0;
2961 }
2962
2963 int r100_cs_track_check(struct radeon_device *rdev, struct r100_cs_track *track)
2964 {
2965         unsigned i;
2966         unsigned long size;
2967         unsigned prim_walk;
2968         unsigned nverts;
2969
2970         for (i = 0; i < track->num_cb; i++) {
2971                 if (track->cb[i].robj == NULL) {
2972                         if (!(track->fastfill || track->color_channel_mask ||
2973                               track->blend_read_enable)) {
2974                                 continue;
2975                         }
2976                         DRM_ERROR("[drm] No buffer for color buffer %d !\n", i);
2977                         return -EINVAL;
2978                 }
2979                 size = track->cb[i].pitch * track->cb[i].cpp * track->maxy;
2980                 size += track->cb[i].offset;
2981                 if (size > radeon_bo_size(track->cb[i].robj)) {
2982                         DRM_ERROR("[drm] Buffer too small for color buffer %d "
2983                                   "(need %lu have %lu) !\n", i, size,
2984                                   radeon_bo_size(track->cb[i].robj));
2985                         DRM_ERROR("[drm] color buffer %d (%u %u %u %u)\n",
2986                                   i, track->cb[i].pitch, track->cb[i].cpp,
2987                                   track->cb[i].offset, track->maxy);
2988                         return -EINVAL;
2989                 }
2990         }
2991         if (track->z_enabled) {
2992                 if (track->zb.robj == NULL) {
2993                         DRM_ERROR("[drm] No buffer for z buffer !\n");
2994                         return -EINVAL;
2995                 }
2996                 size = track->zb.pitch * track->zb.cpp * track->maxy;
2997                 size += track->zb.offset;
2998                 if (size > radeon_bo_size(track->zb.robj)) {
2999                         DRM_ERROR("[drm] Buffer too small for z buffer "
3000                                   "(need %lu have %lu) !\n", size,
3001                                   radeon_bo_size(track->zb.robj));
3002                         DRM_ERROR("[drm] zbuffer (%u %u %u %u)\n",
3003                                   track->zb.pitch, track->zb.cpp,
3004                                   track->zb.offset, track->maxy);
3005                         return -EINVAL;
3006                 }
3007         }
3008         prim_walk = (track->vap_vf_cntl >> 4) & 0x3;
3009         nverts = (track->vap_vf_cntl >> 16) & 0xFFFF;
3010         switch (prim_walk) {
3011         case 1:
3012                 for (i = 0; i < track->num_arrays; i++) {
3013                         size = track->arrays[i].esize * track->max_indx * 4;
3014                         if (track->arrays[i].robj == NULL) {
3015                                 DRM_ERROR("(PW %u) Vertex array %u no buffer "
3016                                           "bound\n", prim_walk, i);
3017                                 return -EINVAL;
3018                         }
3019                         if (size > radeon_bo_size(track->arrays[i].robj)) {
3020                                 dev_err(rdev->dev, "(PW %u) Vertex array %u "
3021                                         "need %lu dwords have %lu dwords\n",
3022                                         prim_walk, i, size >> 2,
3023                                         radeon_bo_size(track->arrays[i].robj)
3024                                         >> 2);
3025                                 DRM_ERROR("Max indices %u\n", track->max_indx);
3026                                 return -EINVAL;
3027                         }
3028                 }
3029                 break;
3030         case 2:
3031                 for (i = 0; i < track->num_arrays; i++) {
3032                         size = track->arrays[i].esize * (nverts - 1) * 4;
3033                         if (track->arrays[i].robj == NULL) {
3034                                 DRM_ERROR("(PW %u) Vertex array %u no buffer "
3035                                           "bound\n", prim_walk, i);
3036                                 return -EINVAL;
3037                         }
3038                         if (size > radeon_bo_size(track->arrays[i].robj)) {
3039                                 dev_err(rdev->dev, "(PW %u) Vertex array %u "
3040                                         "need %lu dwords have %lu dwords\n",
3041                                         prim_walk, i, size >> 2,
3042                                         radeon_bo_size(track->arrays[i].robj)
3043                                         >> 2);
3044                                 return -EINVAL;
3045                         }
3046                 }
3047                 break;
3048         case 3:
3049                 size = track->vtx_size * nverts;
3050                 if (size != track->immd_dwords) {
3051                         DRM_ERROR("IMMD draw %u dwors but needs %lu dwords\n",
3052                                   track->immd_dwords, size);
3053                         DRM_ERROR("VAP_VF_CNTL.NUM_VERTICES %u, VTX_SIZE %u\n",
3054                                   nverts, track->vtx_size);
3055                         return -EINVAL;
3056                 }
3057                 break;
3058         default:
3059                 DRM_ERROR("[drm] Invalid primitive walk %d for VAP_VF_CNTL\n",
3060                           prim_walk);
3061                 return -EINVAL;
3062         }
3063         return r100_cs_track_texture_check(rdev, track);
3064 }
3065
3066 void r100_cs_track_clear(struct radeon_device *rdev, struct r100_cs_track *track)
3067 {
3068         unsigned i, face;
3069
3070         if (rdev->family < CHIP_R300) {
3071                 track->num_cb = 1;
3072                 if (rdev->family <= CHIP_RS200)
3073                         track->num_texture = 3;
3074                 else
3075                         track->num_texture = 6;
3076                 track->maxy = 2048;
3077                 track->separate_cube = 1;
3078         } else {
3079                 track->num_cb = 4;
3080                 track->num_texture = 16;
3081                 track->maxy = 4096;
3082                 track->separate_cube = 0;
3083         }
3084
3085         for (i = 0; i < track->num_cb; i++) {
3086                 track->cb[i].robj = NULL;
3087                 track->cb[i].pitch = 8192;
3088                 track->cb[i].cpp = 16;
3089                 track->cb[i].offset = 0;
3090         }
3091         track->z_enabled = true;
3092         track->zb.robj = NULL;
3093         track->zb.pitch = 8192;
3094         track->zb.cpp = 4;
3095         track->zb.offset = 0;
3096         track->vtx_size = 0x7F;
3097         track->immd_dwords = 0xFFFFFFFFUL;
3098         track->num_arrays = 11;
3099         track->max_indx = 0x00FFFFFFUL;
3100         for (i = 0; i < track->num_arrays; i++) {
3101                 track->arrays[i].robj = NULL;
3102                 track->arrays[i].esize = 0x7F;
3103         }
3104         for (i = 0; i < track->num_texture; i++) {
3105                 track->textures[i].compress_format = R100_TRACK_COMP_NONE;
3106                 track->textures[i].pitch = 16536;
3107                 track->textures[i].width = 16536;
3108                 track->textures[i].height = 16536;
3109                 track->textures[i].width_11 = 1 << 11;
3110                 track->textures[i].height_11 = 1 << 11;
3111                 track->textures[i].num_levels = 12;
3112                 if (rdev->family <= CHIP_RS200) {
3113                         track->textures[i].tex_coord_type = 0;
3114                         track->textures[i].txdepth = 0;
3115                 } else {
3116                         track->textures[i].txdepth = 16;
3117                         track->textures[i].tex_coord_type = 1;
3118                 }
3119                 track->textures[i].cpp = 64;
3120                 track->textures[i].robj = NULL;
3121                 /* CS IB emission code makes sure texture unit are disabled */
3122                 track->textures[i].enabled = false;
3123                 track->textures[i].roundup_w = true;
3124                 track->textures[i].roundup_h = true;
3125                 if (track->separate_cube)
3126                         for (face = 0; face < 5; face++) {
3127                                 track->textures[i].cube_info[face].robj = NULL;
3128                                 track->textures[i].cube_info[face].width = 16536;
3129                                 track->textures[i].cube_info[face].height = 16536;
3130                                 track->textures[i].cube_info[face].offset = 0;
3131                         }
3132         }
3133 }
3134
3135 int r100_ring_test(struct radeon_device *rdev)
3136 {
3137         uint32_t scratch;
3138         uint32_t tmp = 0;
3139         unsigned i;
3140         int r;
3141
3142         r = radeon_scratch_get(rdev, &scratch);
3143         if (r) {
3144                 DRM_ERROR("radeon: cp failed to get scratch reg (%d).\n", r);
3145                 return r;
3146         }
3147         WREG32(scratch, 0xCAFEDEAD);
3148         r = radeon_ring_lock(rdev, 2);
3149         if (r) {
3150                 DRM_ERROR("radeon: cp failed to lock ring (%d).\n", r);
3151                 radeon_scratch_free(rdev, scratch);
3152                 return r;
3153         }
3154         radeon_ring_write(rdev, PACKET0(scratch, 0));
3155         radeon_ring_write(rdev, 0xDEADBEEF);
3156         radeon_ring_unlock_commit(rdev);
3157         for (i = 0; i < rdev->usec_timeout; i++) {
3158                 tmp = RREG32(scratch);
3159                 if (tmp == 0xDEADBEEF) {
3160                         break;
3161                 }
3162                 DRM_UDELAY(1);
3163         }
3164         if (i < rdev->usec_timeout) {
3165                 DRM_INFO("ring test succeeded in %d usecs\n", i);
3166         } else {
3167                 DRM_ERROR("radeon: ring test failed (sracth(0x%04X)=0x%08X)\n",
3168                           scratch, tmp);
3169                 r = -EINVAL;
3170         }
3171         radeon_scratch_free(rdev, scratch);
3172         return r;
3173 }
3174
3175 void r100_ring_ib_execute(struct radeon_device *rdev, struct radeon_ib *ib)
3176 {
3177         radeon_ring_write(rdev, PACKET0(RADEON_CP_IB_BASE, 1));
3178         radeon_ring_write(rdev, ib->gpu_addr);
3179         radeon_ring_write(rdev, ib->length_dw);
3180 }
3181
3182 int r100_ib_test(struct radeon_device *rdev)
3183 {
3184         struct radeon_ib *ib;
3185         uint32_t scratch;
3186         uint32_t tmp = 0;
3187         unsigned i;
3188         int r;
3189
3190         r = radeon_scratch_get(rdev, &scratch);
3191         if (r) {
3192                 DRM_ERROR("radeon: failed to get scratch reg (%d).\n", r);
3193                 return r;
3194         }
3195         WREG32(scratch, 0xCAFEDEAD);
3196         r = radeon_ib_get(rdev, &ib);
3197         if (r) {
3198                 return r;
3199         }
3200         ib->ptr[0] = PACKET0(scratch, 0);
3201         ib->ptr[1] = 0xDEADBEEF;
3202         ib->ptr[2] = PACKET2(0);
3203         ib->ptr[3] = PACKET2(0);
3204         ib->ptr[4] = PACKET2(0);
3205         ib->ptr[5] = PACKET2(0);
3206         ib->ptr[6] = PACKET2(0);
3207         ib->ptr[7] = PACKET2(0);
3208         ib->length_dw = 8;
3209         r = radeon_ib_schedule(rdev, ib);
3210         if (r) {
3211                 radeon_scratch_free(rdev, scratch);
3212                 radeon_ib_free(rdev, &ib);
3213                 return r;
3214         }
3215         r = radeon_fence_wait(ib->fence, false);
3216         if (r) {
3217                 return r;
3218         }
3219         for (i = 0; i < rdev->usec_timeout; i++) {
3220                 tmp = RREG32(scratch);
3221                 if (tmp == 0xDEADBEEF) {
3222                         break;
3223                 }
3224                 DRM_UDELAY(1);
3225         }
3226         if (i < rdev->usec_timeout) {
3227                 DRM_INFO("ib test succeeded in %u usecs\n", i);
3228         } else {
3229                 DRM_ERROR("radeon: ib test failed (sracth(0x%04X)=0x%08X)\n",
3230                           scratch, tmp);
3231                 r = -EINVAL;
3232         }
3233         radeon_scratch_free(rdev, scratch);
3234         radeon_ib_free(rdev, &ib);
3235         return r;
3236 }
3237
3238 void r100_ib_fini(struct radeon_device *rdev)
3239 {
3240         radeon_ib_pool_fini(rdev);
3241 }
3242
3243 int r100_ib_init(struct radeon_device *rdev)
3244 {
3245         int r;
3246
3247         r = radeon_ib_pool_init(rdev);
3248         if (r) {
3249                 dev_err(rdev->dev, "failled initializing IB pool (%d).\n", r);
3250                 r100_ib_fini(rdev);
3251                 return r;
3252         }
3253         r = r100_ib_test(rdev);
3254         if (r) {
3255                 dev_err(rdev->dev, "failled testing IB (%d).\n", r);
3256                 r100_ib_fini(rdev);
3257                 return r;
3258         }
3259         return 0;
3260 }
3261
3262 void r100_mc_stop(struct radeon_device *rdev, struct r100_mc_save *save)
3263 {
3264         /* Shutdown CP we shouldn't need to do that but better be safe than
3265          * sorry
3266          */
3267         rdev->cp.ready = false;
3268         WREG32(R_000740_CP_CSQ_CNTL, 0);
3269
3270         /* Save few CRTC registers */
3271         save->GENMO_WT = RREG8(R_0003C2_GENMO_WT);
3272         save->CRTC_EXT_CNTL = RREG32(R_000054_CRTC_EXT_CNTL);
3273         save->CRTC_GEN_CNTL = RREG32(R_000050_CRTC_GEN_CNTL);
3274         save->CUR_OFFSET = RREG32(R_000260_CUR_OFFSET);
3275         if (!(rdev->flags & RADEON_SINGLE_CRTC)) {
3276                 save->CRTC2_GEN_CNTL = RREG32(R_0003F8_CRTC2_GEN_CNTL);
3277                 save->CUR2_OFFSET = RREG32(R_000360_CUR2_OFFSET);
3278         }
3279
3280         /* Disable VGA aperture access */
3281         WREG8(R_0003C2_GENMO_WT, C_0003C2_VGA_RAM_EN & save->GENMO_WT);
3282         /* Disable cursor, overlay, crtc */
3283         WREG32(R_000260_CUR_OFFSET, save->CUR_OFFSET | S_000260_CUR_LOCK(1));
3284         WREG32(R_000054_CRTC_EXT_CNTL, save->CRTC_EXT_CNTL |
3285                                         S_000054_CRTC_DISPLAY_DIS(1));
3286         WREG32(R_000050_CRTC_GEN_CNTL,
3287                         (C_000050_CRTC_CUR_EN & save->CRTC_GEN_CNTL) |
3288                         S_000050_CRTC_DISP_REQ_EN_B(1));
3289         WREG32(R_000420_OV0_SCALE_CNTL,
3290                 C_000420_OV0_OVERLAY_EN & RREG32(R_000420_OV0_SCALE_CNTL));
3291         WREG32(R_000260_CUR_OFFSET, C_000260_CUR_LOCK & save->CUR_OFFSET);
3292         if (!(rdev->flags & RADEON_SINGLE_CRTC)) {
3293                 WREG32(R_000360_CUR2_OFFSET, save->CUR2_OFFSET |
3294                                                 S_000360_CUR2_LOCK(1));
3295                 WREG32(R_0003F8_CRTC2_GEN_CNTL,
3296                         (C_0003F8_CRTC2_CUR_EN & save->CRTC2_GEN_CNTL) |
3297                         S_0003F8_CRTC2_DISPLAY_DIS(1) |
3298                         S_0003F8_CRTC2_DISP_REQ_EN_B(1));
3299                 WREG32(R_000360_CUR2_OFFSET,
3300                         C_000360_CUR2_LOCK & save->CUR2_OFFSET);
3301         }
3302 }
3303
3304 void r100_mc_resume(struct radeon_device *rdev, struct r100_mc_save *save)
3305 {
3306         /* Update base address for crtc */
3307         WREG32(R_00023C_DISPLAY_BASE_ADDR, rdev->mc.vram_start);
3308         if (!(rdev->flags & RADEON_SINGLE_CRTC)) {
3309                 WREG32(R_00033C_CRTC2_DISPLAY_BASE_ADDR, rdev->mc.vram_start);
3310         }
3311         /* Restore CRTC registers */
3312         WREG8(R_0003C2_GENMO_WT, save->GENMO_WT);
3313         WREG32(R_000054_CRTC_EXT_CNTL, save->CRTC_EXT_CNTL);
3314         WREG32(R_000050_CRTC_GEN_CNTL, save->CRTC_GEN_CNTL);
3315         if (!(rdev->flags & RADEON_SINGLE_CRTC)) {
3316                 WREG32(R_0003F8_CRTC2_GEN_CNTL, save->CRTC2_GEN_CNTL);
3317         }
3318 }
3319
3320 void r100_vga_render_disable(struct radeon_device *rdev)
3321 {
3322         u32 tmp;
3323
3324         tmp = RREG8(R_0003C2_GENMO_WT);
3325         WREG8(R_0003C2_GENMO_WT, C_0003C2_VGA_RAM_EN & tmp);
3326 }
3327
3328 static void r100_debugfs(struct radeon_device *rdev)
3329 {
3330         int r;
3331
3332         r = r100_debugfs_mc_info_init(rdev);
3333         if (r)
3334                 dev_warn(rdev->dev, "Failed to create r100_mc debugfs file.\n");
3335 }
3336
3337 static void r100_mc_program(struct radeon_device *rdev)
3338 {
3339         struct r100_mc_save save;
3340
3341         /* Stops all mc clients */
3342         r100_mc_stop(rdev, &save);
3343         if (rdev->flags & RADEON_IS_AGP) {
3344                 WREG32(R_00014C_MC_AGP_LOCATION,
3345                         S_00014C_MC_AGP_START(rdev->mc.gtt_start >> 16) |
3346                         S_00014C_MC_AGP_TOP(rdev->mc.gtt_end >> 16));
3347                 WREG32(R_000170_AGP_BASE, lower_32_bits(rdev->mc.agp_base));
3348                 if (rdev->family > CHIP_RV200)
3349                         WREG32(R_00015C_AGP_BASE_2,
3350                                 upper_32_bits(rdev->mc.agp_base) & 0xff);
3351         } else {
3352                 WREG32(R_00014C_MC_AGP_LOCATION, 0x0FFFFFFF);
3353                 WREG32(R_000170_AGP_BASE, 0);
3354                 if (rdev->family > CHIP_RV200)
3355                         WREG32(R_00015C_AGP_BASE_2, 0);
3356         }
3357         /* Wait for mc idle */
3358         if (r100_mc_wait_for_idle(rdev))
3359                 dev_warn(rdev->dev, "Wait for MC idle timeout.\n");
3360         /* Program MC, should be a 32bits limited address space */
3361         WREG32(R_000148_MC_FB_LOCATION,
3362                 S_000148_MC_FB_START(rdev->mc.vram_start >> 16) |
3363                 S_000148_MC_FB_TOP(rdev->mc.vram_end >> 16));
3364         r100_mc_resume(rdev, &save);
3365 }
3366
3367 void r100_clock_startup(struct radeon_device *rdev)
3368 {
3369         u32 tmp;
3370
3371         if (radeon_dynclks != -1 && radeon_dynclks)
3372                 radeon_legacy_set_clock_gating(rdev, 1);
3373         /* We need to force on some of the block */
3374         tmp = RREG32_PLL(R_00000D_SCLK_CNTL);
3375         tmp |= S_00000D_FORCE_CP(1) | S_00000D_FORCE_VIP(1);
3376         if ((rdev->family == CHIP_RV250) || (rdev->family == CHIP_RV280))
3377                 tmp |= S_00000D_FORCE_DISP1(1) | S_00000D_FORCE_DISP2(1);
3378         WREG32_PLL(R_00000D_SCLK_CNTL, tmp);
3379 }
3380
3381 static int r100_startup(struct radeon_device *rdev)
3382 {
3383         int r;
3384
3385         /* set common regs */
3386         r100_set_common_regs(rdev);
3387         /* program mc */
3388         r100_mc_program(rdev);
3389         /* Resume clock */
3390         r100_clock_startup(rdev);
3391         /* Initialize GPU configuration (# pipes, ...) */
3392         r100_gpu_init(rdev);
3393         /* Initialize GART (initialize after TTM so we can allocate
3394          * memory through TTM but finalize after TTM) */
3395         r100_enable_bm(rdev);
3396         if (rdev->flags & RADEON_IS_PCI) {
3397                 r = r100_pci_gart_enable(rdev);
3398                 if (r)
3399                         return r;
3400         }
3401         /* Enable IRQ */
3402         r100_irq_set(rdev);
3403         rdev->config.r100.hdp_cntl = RREG32(RADEON_HOST_PATH_CNTL);
3404         /* 1M ring buffer */
3405         r = r100_cp_init(rdev, 1024 * 1024);
3406         if (r) {
3407                 dev_err(rdev->dev, "failled initializing CP (%d).\n", r);
3408                 return r;
3409         }
3410         r = r100_wb_init(rdev);
3411         if (r)
3412                 dev_err(rdev->dev, "failled initializing WB (%d).\n", r);
3413         r = r100_ib_init(rdev);
3414         if (r) {
3415                 dev_err(rdev->dev, "failled initializing IB (%d).\n", r);
3416                 return r;
3417         }
3418         return 0;
3419 }
3420
3421 int r100_resume(struct radeon_device *rdev)
3422 {
3423         /* Make sur GART are not working */
3424         if (rdev->flags & RADEON_IS_PCI)
3425                 r100_pci_gart_disable(rdev);
3426         /* Resume clock before doing reset */
3427         r100_clock_startup(rdev);
3428         /* Reset gpu before posting otherwise ATOM will enter infinite loop */
3429         if (radeon_gpu_reset(rdev)) {
3430                 dev_warn(rdev->dev, "GPU reset failed ! (0xE40=0x%08X, 0x7C0=0x%08X)\n",
3431                         RREG32(R_000E40_RBBM_STATUS),
3432                         RREG32(R_0007C0_CP_STAT));
3433         }
3434         /* post */
3435         radeon_combios_asic_init(rdev->ddev);
3436         /* Resume clock after posting */
3437         r100_clock_startup(rdev);
3438         /* Initialize surface registers */
3439         radeon_surface_init(rdev);
3440         return r100_startup(rdev);
3441 }
3442
3443 int r100_suspend(struct radeon_device *rdev)
3444 {
3445         r100_cp_disable(rdev);
3446         r100_wb_disable(rdev);
3447         r100_irq_disable(rdev);
3448         if (rdev->flags & RADEON_IS_PCI)
3449                 r100_pci_gart_disable(rdev);
3450         return 0;
3451 }
3452
3453 void r100_fini(struct radeon_device *rdev)
3454 {
3455         radeon_pm_fini(rdev);
3456         r100_cp_fini(rdev);
3457         r100_wb_fini(rdev);
3458         r100_ib_fini(rdev);
3459         radeon_gem_fini(rdev);
3460         if (rdev->flags & RADEON_IS_PCI)
3461                 r100_pci_gart_fini(rdev);
3462         radeon_agp_fini(rdev);
3463         radeon_irq_kms_fini(rdev);
3464         radeon_fence_driver_fini(rdev);
3465         radeon_bo_fini(rdev);
3466         radeon_atombios_fini(rdev);
3467         kfree(rdev->bios);
3468         rdev->bios = NULL;
3469 }
3470
3471 int r100_init(struct radeon_device *rdev)
3472 {
3473         int r;
3474
3475         /* Register debugfs file specific to this group of asics */
3476         r100_debugfs(rdev);
3477         /* Disable VGA */
3478         r100_vga_render_disable(rdev);
3479         /* Initialize scratch registers */
3480         radeon_scratch_init(rdev);
3481         /* Initialize surface registers */
3482         radeon_surface_init(rdev);
3483         /* TODO: disable VGA need to use VGA request */
3484         /* BIOS*/
3485         if (!radeon_get_bios(rdev)) {
3486                 if (ASIC_IS_AVIVO(rdev))
3487                         return -EINVAL;
3488         }
3489         if (rdev->is_atom_bios) {
3490                 dev_err(rdev->dev, "Expecting combios for RS400/RS480 GPU\n");
3491                 return -EINVAL;
3492         } else {
3493                 r = radeon_combios_init(rdev);
3494                 if (r)
3495                         return r;
3496         }
3497         /* Reset gpu before posting otherwise ATOM will enter infinite loop */
3498         if (radeon_gpu_reset(rdev)) {
3499                 dev_warn(rdev->dev,
3500                         "GPU reset failed ! (0xE40=0x%08X, 0x7C0=0x%08X)\n",
3501                         RREG32(R_000E40_RBBM_STATUS),
3502                         RREG32(R_0007C0_CP_STAT));
3503         }
3504         /* check if cards are posted or not */
3505         if (radeon_boot_test_post_card(rdev) == false)
3506                 return -EINVAL;
3507         /* Set asic errata */
3508         r100_errata(rdev);
3509         /* Initialize clocks */
3510         radeon_get_clock_info(rdev->ddev);
3511         /* Initialize power management */
3512         radeon_pm_init(rdev);
3513         /* initialize AGP */
3514         if (rdev->flags & RADEON_IS_AGP) {
3515                 r = radeon_agp_init(rdev);
3516                 if (r) {
3517                         radeon_agp_disable(rdev);
3518                 }
3519         }
3520         /* initialize VRAM */
3521         r100_mc_init(rdev);
3522         /* Fence driver */
3523         r = radeon_fence_driver_init(rdev);
3524         if (r)
3525                 return r;
3526         r = radeon_irq_kms_init(rdev);
3527         if (r)
3528                 return r;
3529         /* Memory manager */
3530         r = radeon_bo_init(rdev);
3531         if (r)
3532                 return r;
3533         if (rdev->flags & RADEON_IS_PCI) {
3534                 r = r100_pci_gart_init(rdev);
3535                 if (r)
3536                         return r;
3537         }
3538         r100_set_safe_registers(rdev);
3539         rdev->accel_working = true;
3540         r = r100_startup(rdev);
3541         if (r) {
3542                 /* Somethings want wront with the accel init stop accel */
3543                 dev_err(rdev->dev, "Disabling GPU acceleration\n");
3544                 r100_cp_fini(rdev);
3545                 r100_wb_fini(rdev);
3546                 r100_ib_fini(rdev);
3547                 radeon_irq_kms_fini(rdev);
3548                 if (rdev->flags & RADEON_IS_PCI)
3549                         r100_pci_gart_fini(rdev);
3550                 rdev->accel_working = false;
3551         }
3552         return 0;
3553 }