SAFE public projects git trees. - safe/jmp/linux-2.6/blob - drivers/gpu/drm/radeon/r100.c

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