SAFE public projects git trees. - safe/jmp/linux-2.6/blob - drivers/gpu/drm/radeon/r600.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 <linux/firmware.h>
  30 #include <linux/platform_device.h>
  31 #include "drmP.h"
  32 #include "radeon_drm.h"
  33 #include "radeon.h"
  34 #include "radeon_mode.h"
  35 #include "r600d.h"
  36 #include "atom.h"
  37 #include "avivod.h"
  38
  39 #define PFP_UCODE_SIZE 576
  40 #define PM4_UCODE_SIZE 1792
  41 #define RLC_UCODE_SIZE 768
  42 #define R700_PFP_UCODE_SIZE 848
  43 #define R700_PM4_UCODE_SIZE 1360
  44 #define R700_RLC_UCODE_SIZE 1024
  45
  46 /* Firmware Names */
  47 MODULE_FIRMWARE("radeon/R600_pfp.bin");
  48 MODULE_FIRMWARE("radeon/R600_me.bin");
  49 MODULE_FIRMWARE("radeon/RV610_pfp.bin");
  50 MODULE_FIRMWARE("radeon/RV610_me.bin");
  51 MODULE_FIRMWARE("radeon/RV630_pfp.bin");
  52 MODULE_FIRMWARE("radeon/RV630_me.bin");
  53 MODULE_FIRMWARE("radeon/RV620_pfp.bin");
  54 MODULE_FIRMWARE("radeon/RV620_me.bin");
  55 MODULE_FIRMWARE("radeon/RV635_pfp.bin");
  56 MODULE_FIRMWARE("radeon/RV635_me.bin");
  57 MODULE_FIRMWARE("radeon/RV670_pfp.bin");
  58 MODULE_FIRMWARE("radeon/RV670_me.bin");
  59 MODULE_FIRMWARE("radeon/RS780_pfp.bin");
  60 MODULE_FIRMWARE("radeon/RS780_me.bin");
  61 MODULE_FIRMWARE("radeon/RV770_pfp.bin");
  62 MODULE_FIRMWARE("radeon/RV770_me.bin");
  63 MODULE_FIRMWARE("radeon/RV730_pfp.bin");
  64 MODULE_FIRMWARE("radeon/RV730_me.bin");
  65 MODULE_FIRMWARE("radeon/RV710_pfp.bin");
  66 MODULE_FIRMWARE("radeon/RV710_me.bin");
  67 MODULE_FIRMWARE("radeon/R600_rlc.bin");
  68 MODULE_FIRMWARE("radeon/R700_rlc.bin");
  69
  70 int r600_debugfs_mc_info_init(struct radeon_device *rdev);
  71
  72 /* r600,rv610,rv630,rv620,rv635,rv670 */
  73 int r600_mc_wait_for_idle(struct radeon_device *rdev);
  74 void r600_gpu_init(struct radeon_device *rdev);
  75 void r600_fini(struct radeon_device *rdev);
  76
  77 /* hpd for digital panel detect/disconnect */
  78 bool r600_hpd_sense(struct radeon_device *rdev, enum radeon_hpd_id hpd)
  79 {
  80         bool connected = false;
  81
  82         if (ASIC_IS_DCE3(rdev)) {
  83                 switch (hpd) {
  84                 case RADEON_HPD_1:
  85                         if (RREG32(DC_HPD1_INT_STATUS) & DC_HPDx_SENSE)
  86                                 connected = true;
  87                         break;
  88                 case RADEON_HPD_2:
  89                         if (RREG32(DC_HPD2_INT_STATUS) & DC_HPDx_SENSE)
  90                                 connected = true;
  91                         break;
  92                 case RADEON_HPD_3:
  93                         if (RREG32(DC_HPD3_INT_STATUS) & DC_HPDx_SENSE)
  94                                 connected = true;
  95                         break;
  96                 case RADEON_HPD_4:
  97                         if (RREG32(DC_HPD4_INT_STATUS) & DC_HPDx_SENSE)
  98                                 connected = true;
  99                         break;
 100                         /* DCE 3.2 */
 101                 case RADEON_HPD_5:
 102                         if (RREG32(DC_HPD5_INT_STATUS) & DC_HPDx_SENSE)
 103                                 connected = true;
 104                         break;
 105                 case RADEON_HPD_6:
 106                         if (RREG32(DC_HPD6_INT_STATUS) & DC_HPDx_SENSE)
 107                                 connected = true;
 108                         break;
 109                 default:
 110                         break;
 111                 }
 112         } else {
 113                 switch (hpd) {
 114                 case RADEON_HPD_1:
 115                         if (RREG32(DC_HOT_PLUG_DETECT1_INT_STATUS) & DC_HOT_PLUG_DETECTx_SENSE)
 116                                 connected = true;
 117                         break;
 118                 case RADEON_HPD_2:
 119                         if (RREG32(DC_HOT_PLUG_DETECT2_INT_STATUS) & DC_HOT_PLUG_DETECTx_SENSE)
 120                                 connected = true;
 121                         break;
 122                 case RADEON_HPD_3:
 123                         if (RREG32(DC_HOT_PLUG_DETECT3_INT_STATUS) & DC_HOT_PLUG_DETECTx_SENSE)
 124                                 connected = true;
 125                         break;
 126                 default:
 127                         break;
 128                 }
 129         }
 130         return connected;
 131 }
 132
 133 void r600_hpd_set_polarity(struct radeon_device *rdev,
 134                            enum radeon_hpd_id hpd)
 135 {
 136         u32 tmp;
 137         bool connected = r600_hpd_sense(rdev, hpd);
 138
 139         if (ASIC_IS_DCE3(rdev)) {
 140                 switch (hpd) {
 141                 case RADEON_HPD_1:
 142                         tmp = RREG32(DC_HPD1_INT_CONTROL);
 143                         if (connected)
 144                                 tmp &= ~DC_HPDx_INT_POLARITY;
 145                         else
 146                                 tmp |= DC_HPDx_INT_POLARITY;
 147                         WREG32(DC_HPD1_INT_CONTROL, tmp);
 148                         break;
 149                 case RADEON_HPD_2:
 150                         tmp = RREG32(DC_HPD2_INT_CONTROL);
 151                         if (connected)
 152                                 tmp &= ~DC_HPDx_INT_POLARITY;
 153                         else
 154                                 tmp |= DC_HPDx_INT_POLARITY;
 155                         WREG32(DC_HPD2_INT_CONTROL, tmp);
 156                         break;
 157                 case RADEON_HPD_3:
 158                         tmp = RREG32(DC_HPD3_INT_CONTROL);
 159                         if (connected)
 160                                 tmp &= ~DC_HPDx_INT_POLARITY;
 161                         else
 162                                 tmp |= DC_HPDx_INT_POLARITY;
 163                         WREG32(DC_HPD3_INT_CONTROL, tmp);
 164                         break;
 165                 case RADEON_HPD_4:
 166                         tmp = RREG32(DC_HPD4_INT_CONTROL);
 167                         if (connected)
 168                                 tmp &= ~DC_HPDx_INT_POLARITY;
 169                         else
 170                                 tmp |= DC_HPDx_INT_POLARITY;
 171                         WREG32(DC_HPD4_INT_CONTROL, tmp);
 172                         break;
 173                 case RADEON_HPD_5:
 174                         tmp = RREG32(DC_HPD5_INT_CONTROL);
 175                         if (connected)
 176                                 tmp &= ~DC_HPDx_INT_POLARITY;
 177                         else
 178                                 tmp |= DC_HPDx_INT_POLARITY;
 179                         WREG32(DC_HPD5_INT_CONTROL, tmp);
 180                         break;
 181                         /* DCE 3.2 */
 182                 case RADEON_HPD_6:
 183                         tmp = RREG32(DC_HPD6_INT_CONTROL);
 184                         if (connected)
 185                                 tmp &= ~DC_HPDx_INT_POLARITY;
 186                         else
 187                                 tmp |= DC_HPDx_INT_POLARITY;
 188                         WREG32(DC_HPD6_INT_CONTROL, tmp);
 189                         break;
 190                 default:
 191                         break;
 192                 }
 193         } else {
 194                 switch (hpd) {
 195                 case RADEON_HPD_1:
 196                         tmp = RREG32(DC_HOT_PLUG_DETECT1_INT_CONTROL);
 197                         if (connected)
 198                                 tmp &= ~DC_HOT_PLUG_DETECTx_INT_POLARITY;
 199                         else
 200                                 tmp |= DC_HOT_PLUG_DETECTx_INT_POLARITY;
 201                         WREG32(DC_HOT_PLUG_DETECT1_INT_CONTROL, tmp);
 202                         break;
 203                 case RADEON_HPD_2:
 204                         tmp = RREG32(DC_HOT_PLUG_DETECT2_INT_CONTROL);
 205                         if (connected)
 206                                 tmp &= ~DC_HOT_PLUG_DETECTx_INT_POLARITY;
 207                         else
 208                                 tmp |= DC_HOT_PLUG_DETECTx_INT_POLARITY;
 209                         WREG32(DC_HOT_PLUG_DETECT2_INT_CONTROL, tmp);
 210                         break;
 211                 case RADEON_HPD_3:
 212                         tmp = RREG32(DC_HOT_PLUG_DETECT3_INT_CONTROL);
 213                         if (connected)
 214                                 tmp &= ~DC_HOT_PLUG_DETECTx_INT_POLARITY;
 215                         else
 216                                 tmp |= DC_HOT_PLUG_DETECTx_INT_POLARITY;
 217                         WREG32(DC_HOT_PLUG_DETECT3_INT_CONTROL, tmp);
 218                         break;
 219                 default:
 220                         break;
 221                 }
 222         }
 223 }
 224
 225 void r600_hpd_init(struct radeon_device *rdev)
 226 {
 227         struct drm_device *dev = rdev->ddev;
 228         struct drm_connector *connector;
 229
 230         if (ASIC_IS_DCE3(rdev)) {
 231                 u32 tmp = DC_HPDx_CONNECTION_TIMER(0x9c4) | DC_HPDx_RX_INT_TIMER(0xfa);
 232                 if (ASIC_IS_DCE32(rdev))
 233                         tmp |= DC_HPDx_EN;
 234
 235                 list_for_each_entry(connector, &dev->mode_config.connector_list, head) {
 236                         struct radeon_connector *radeon_connector = to_radeon_connector(connector);
 237                         switch (radeon_connector->hpd.hpd) {
 238                         case RADEON_HPD_1:
 239                                 WREG32(DC_HPD1_CONTROL, tmp);
 240                                 rdev->irq.hpd[0] = true;
 241                                 break;
 242                         case RADEON_HPD_2:
 243                                 WREG32(DC_HPD2_CONTROL, tmp);
 244                                 rdev->irq.hpd[1] = true;
 245                                 break;
 246                         case RADEON_HPD_3:
 247                                 WREG32(DC_HPD3_CONTROL, tmp);
 248                                 rdev->irq.hpd[2] = true;
 249                                 break;
 250                         case RADEON_HPD_4:
 251                                 WREG32(DC_HPD4_CONTROL, tmp);
 252                                 rdev->irq.hpd[3] = true;
 253                                 break;
 254                                 /* DCE 3.2 */
 255                         case RADEON_HPD_5:
 256                                 WREG32(DC_HPD5_CONTROL, tmp);
 257                                 rdev->irq.hpd[4] = true;
 258                                 break;
 259                         case RADEON_HPD_6:
 260                                 WREG32(DC_HPD6_CONTROL, tmp);
 261                                 rdev->irq.hpd[5] = true;
 262                                 break;
 263                         default:
 264                                 break;
 265                         }
 266                 }
 267         } else {
 268                 list_for_each_entry(connector, &dev->mode_config.connector_list, head) {
 269                         struct radeon_connector *radeon_connector = to_radeon_connector(connector);
 270                         switch (radeon_connector->hpd.hpd) {
 271                         case RADEON_HPD_1:
 272                                 WREG32(DC_HOT_PLUG_DETECT1_CONTROL, DC_HOT_PLUG_DETECTx_EN);
 273                                 rdev->irq.hpd[0] = true;
 274                                 break;
 275                         case RADEON_HPD_2:
 276                                 WREG32(DC_HOT_PLUG_DETECT2_CONTROL, DC_HOT_PLUG_DETECTx_EN);
 277                                 rdev->irq.hpd[1] = true;
 278                                 break;
 279                         case RADEON_HPD_3:
 280                                 WREG32(DC_HOT_PLUG_DETECT3_CONTROL, DC_HOT_PLUG_DETECTx_EN);
 281                                 rdev->irq.hpd[2] = true;
 282                                 break;
 283                         default:
 284                                 break;
 285                         }
 286                 }
 287         }
 288         if (rdev->irq.installed)
 289                 r600_irq_set(rdev);
 290 }
 291
 292 void r600_hpd_fini(struct radeon_device *rdev)
 293 {
 294         struct drm_device *dev = rdev->ddev;
 295         struct drm_connector *connector;
 296
 297         if (ASIC_IS_DCE3(rdev)) {
 298                 list_for_each_entry(connector, &dev->mode_config.connector_list, head) {
 299                         struct radeon_connector *radeon_connector = to_radeon_connector(connector);
 300                         switch (radeon_connector->hpd.hpd) {
 301                         case RADEON_HPD_1:
 302                                 WREG32(DC_HPD1_CONTROL, 0);
 303                                 rdev->irq.hpd[0] = false;
 304                                 break;
 305                         case RADEON_HPD_2:
 306                                 WREG32(DC_HPD2_CONTROL, 0);
 307                                 rdev->irq.hpd[1] = false;
 308                                 break;
 309                         case RADEON_HPD_3:
 310                                 WREG32(DC_HPD3_CONTROL, 0);
 311                                 rdev->irq.hpd[2] = false;
 312                                 break;
 313                         case RADEON_HPD_4:
 314                                 WREG32(DC_HPD4_CONTROL, 0);
 315                                 rdev->irq.hpd[3] = false;
 316                                 break;
 317                                 /* DCE 3.2 */
 318                         case RADEON_HPD_5:
 319                                 WREG32(DC_HPD5_CONTROL, 0);
 320                                 rdev->irq.hpd[4] = false;
 321                                 break;
 322                         case RADEON_HPD_6:
 323                                 WREG32(DC_HPD6_CONTROL, 0);
 324                                 rdev->irq.hpd[5] = false;
 325                                 break;
 326                         default:
 327                                 break;
 328                         }
 329                 }
 330         } else {
 331                 list_for_each_entry(connector, &dev->mode_config.connector_list, head) {
 332                         struct radeon_connector *radeon_connector = to_radeon_connector(connector);
 333                         switch (radeon_connector->hpd.hpd) {
 334                         case RADEON_HPD_1:
 335                                 WREG32(DC_HOT_PLUG_DETECT1_CONTROL, 0);
 336                                 rdev->irq.hpd[0] = false;
 337                                 break;
 338                         case RADEON_HPD_2:
 339                                 WREG32(DC_HOT_PLUG_DETECT2_CONTROL, 0);
 340                                 rdev->irq.hpd[1] = false;
 341                                 break;
 342                         case RADEON_HPD_3:
 343                                 WREG32(DC_HOT_PLUG_DETECT3_CONTROL, 0);
 344                                 rdev->irq.hpd[2] = false;
 345                                 break;
 346                         default:
 347                                 break;
 348                         }
 349                 }
 350         }
 351 }
 352
 353 /*
 354  * R600 PCIE GART
 355  */
 356 void r600_pcie_gart_tlb_flush(struct radeon_device *rdev)
 357 {
 358         unsigned i;
 359         u32 tmp;
 360
 361         /* flush hdp cache so updates hit vram */
 362         WREG32(R_005480_HDP_MEM_COHERENCY_FLUSH_CNTL, 0x1);
 363
 364         WREG32(VM_CONTEXT0_INVALIDATION_LOW_ADDR, rdev->mc.gtt_start >> 12);
 365         WREG32(VM_CONTEXT0_INVALIDATION_HIGH_ADDR, (rdev->mc.gtt_end - 1) >> 12);
 366         WREG32(VM_CONTEXT0_REQUEST_RESPONSE, REQUEST_TYPE(1));
 367         for (i = 0; i < rdev->usec_timeout; i++) {
 368                 /* read MC_STATUS */
 369                 tmp = RREG32(VM_CONTEXT0_REQUEST_RESPONSE);
 370                 tmp = (tmp & RESPONSE_TYPE_MASK) >> RESPONSE_TYPE_SHIFT;
 371                 if (tmp == 2) {
 372                         printk(KERN_WARNING "[drm] r600 flush TLB failed\n");
 373                         return;
 374                 }
 375                 if (tmp) {
 376                         return;
 377                 }
 378                 udelay(1);
 379         }
 380 }
 381
 382 int r600_pcie_gart_init(struct radeon_device *rdev)
 383 {
 384         int r;
 385
 386         if (rdev->gart.table.vram.robj) {
 387                 WARN(1, "R600 PCIE GART already initialized.\n");
 388                 return 0;
 389         }
 390         /* Initialize common gart structure */
 391         r = radeon_gart_init(rdev);
 392         if (r)
 393                 return r;
 394         rdev->gart.table_size = rdev->gart.num_gpu_pages * 8;
 395         return radeon_gart_table_vram_alloc(rdev);
 396 }
 397
 398 int r600_pcie_gart_enable(struct radeon_device *rdev)
 399 {
 400         u32 tmp;
 401         int r, i;
 402
 403         if (rdev->gart.table.vram.robj == NULL) {
 404                 dev_err(rdev->dev, "No VRAM object for PCIE GART.\n");
 405                 return -EINVAL;
 406         }
 407         r = radeon_gart_table_vram_pin(rdev);
 408         if (r)
 409                 return r;
 410         radeon_gart_restore(rdev);
 411
 412         /* Setup L2 cache */
 413         WREG32(VM_L2_CNTL, ENABLE_L2_CACHE | ENABLE_L2_FRAGMENT_PROCESSING |
 414                                 ENABLE_L2_PTE_CACHE_LRU_UPDATE_BY_WRITE |
 415                                 EFFECTIVE_L2_QUEUE_SIZE(7));
 416         WREG32(VM_L2_CNTL2, 0);
 417         WREG32(VM_L2_CNTL3, BANK_SELECT_0(0) | BANK_SELECT_1(1));
 418         /* Setup TLB control */
 419         tmp = ENABLE_L1_TLB | ENABLE_L1_FRAGMENT_PROCESSING |
 420                 SYSTEM_ACCESS_MODE_NOT_IN_SYS |
 421                 EFFECTIVE_L1_TLB_SIZE(5) | EFFECTIVE_L1_QUEUE_SIZE(5) |
 422                 ENABLE_WAIT_L2_QUERY;
 423         WREG32(MC_VM_L1_TLB_MCB_RD_SYS_CNTL, tmp);
 424         WREG32(MC_VM_L1_TLB_MCB_WR_SYS_CNTL, tmp);
 425         WREG32(MC_VM_L1_TLB_MCB_RD_HDP_CNTL, tmp | ENABLE_L1_STRICT_ORDERING);
 426         WREG32(MC_VM_L1_TLB_MCB_WR_HDP_CNTL, tmp);
 427         WREG32(MC_VM_L1_TLB_MCD_RD_A_CNTL, tmp);
 428         WREG32(MC_VM_L1_TLB_MCD_WR_A_CNTL, tmp);
 429         WREG32(MC_VM_L1_TLB_MCD_RD_B_CNTL, tmp);
 430         WREG32(MC_VM_L1_TLB_MCD_WR_B_CNTL, tmp);
 431         WREG32(MC_VM_L1_TLB_MCB_RD_GFX_CNTL, tmp);
 432         WREG32(MC_VM_L1_TLB_MCB_WR_GFX_CNTL, tmp);
 433         WREG32(MC_VM_L1_TLB_MCB_RD_PDMA_CNTL, tmp);
 434         WREG32(MC_VM_L1_TLB_MCB_WR_PDMA_CNTL, tmp);
 435         WREG32(MC_VM_L1_TLB_MCB_RD_SEM_CNTL, tmp | ENABLE_SEMAPHORE_MODE);
 436         WREG32(MC_VM_L1_TLB_MCB_WR_SEM_CNTL, tmp | ENABLE_SEMAPHORE_MODE);
 437         WREG32(VM_CONTEXT0_PAGE_TABLE_START_ADDR, rdev->mc.gtt_start >> 12);
 438         WREG32(VM_CONTEXT0_PAGE_TABLE_END_ADDR, rdev->mc.gtt_end >> 12);
 439         WREG32(VM_CONTEXT0_PAGE_TABLE_BASE_ADDR, rdev->gart.table_addr >> 12);
 440         WREG32(VM_CONTEXT0_CNTL, ENABLE_CONTEXT | PAGE_TABLE_DEPTH(0) |
 441                                 RANGE_PROTECTION_FAULT_ENABLE_DEFAULT);
 442         WREG32(VM_CONTEXT0_PROTECTION_FAULT_DEFAULT_ADDR,
 443                         (u32)(rdev->dummy_page.addr >> 12));
 444         for (i = 1; i < 7; i++)
 445                 WREG32(VM_CONTEXT0_CNTL + (i * 4), 0);
 446
 447         r600_pcie_gart_tlb_flush(rdev);
 448         rdev->gart.ready = true;
 449         return 0;
 450 }
 451
 452 void r600_pcie_gart_disable(struct radeon_device *rdev)
 453 {
 454         u32 tmp;
 455         int i, r;
 456
 457         /* Disable all tables */
 458         for (i = 0; i < 7; i++)
 459                 WREG32(VM_CONTEXT0_CNTL + (i * 4), 0);
 460
 461         /* Disable L2 cache */
 462         WREG32(VM_L2_CNTL, ENABLE_L2_FRAGMENT_PROCESSING |
 463                                 EFFECTIVE_L2_QUEUE_SIZE(7));
 464         WREG32(VM_L2_CNTL3, BANK_SELECT_0(0) | BANK_SELECT_1(1));
 465         /* Setup L1 TLB control */
 466         tmp = EFFECTIVE_L1_TLB_SIZE(5) | EFFECTIVE_L1_QUEUE_SIZE(5) |
 467                 ENABLE_WAIT_L2_QUERY;
 468         WREG32(MC_VM_L1_TLB_MCD_RD_A_CNTL, tmp);
 469         WREG32(MC_VM_L1_TLB_MCD_WR_A_CNTL, tmp);
 470         WREG32(MC_VM_L1_TLB_MCD_RD_B_CNTL, tmp);
 471         WREG32(MC_VM_L1_TLB_MCD_WR_B_CNTL, tmp);
 472         WREG32(MC_VM_L1_TLB_MCB_RD_GFX_CNTL, tmp);
 473         WREG32(MC_VM_L1_TLB_MCB_WR_GFX_CNTL, tmp);
 474         WREG32(MC_VM_L1_TLB_MCB_RD_PDMA_CNTL, tmp);
 475         WREG32(MC_VM_L1_TLB_MCB_WR_PDMA_CNTL, tmp);
 476         WREG32(MC_VM_L1_TLB_MCB_RD_SEM_CNTL, tmp);
 477         WREG32(MC_VM_L1_TLB_MCB_WR_SEM_CNTL, tmp);
 478         WREG32(MC_VM_L1_TLB_MCB_RD_SYS_CNTL, tmp);
 479         WREG32(MC_VM_L1_TLB_MCB_WR_SYS_CNTL, tmp);
 480         WREG32(MC_VM_L1_TLB_MCB_RD_HDP_CNTL, tmp);
 481         WREG32(MC_VM_L1_TLB_MCB_WR_HDP_CNTL, tmp);
 482         if (rdev->gart.table.vram.robj) {
 483                 r = radeon_bo_reserve(rdev->gart.table.vram.robj, false);
 484                 if (likely(r == 0)) {
 485                         radeon_bo_kunmap(rdev->gart.table.vram.robj);
 486                         radeon_bo_unpin(rdev->gart.table.vram.robj);
 487                         radeon_bo_unreserve(rdev->gart.table.vram.robj);
 488                 }
 489         }
 490 }
 491
 492 void r600_pcie_gart_fini(struct radeon_device *rdev)
 493 {
 494         r600_pcie_gart_disable(rdev);
 495         radeon_gart_table_vram_free(rdev);
 496         radeon_gart_fini(rdev);
 497 }
 498
 499 void r600_agp_enable(struct radeon_device *rdev)
 500 {
 501         u32 tmp;
 502         int i;
 503
 504         /* Setup L2 cache */
 505         WREG32(VM_L2_CNTL, ENABLE_L2_CACHE | ENABLE_L2_FRAGMENT_PROCESSING |
 506                                 ENABLE_L2_PTE_CACHE_LRU_UPDATE_BY_WRITE |
 507                                 EFFECTIVE_L2_QUEUE_SIZE(7));
 508         WREG32(VM_L2_CNTL2, 0);
 509         WREG32(VM_L2_CNTL3, BANK_SELECT_0(0) | BANK_SELECT_1(1));
 510         /* Setup TLB control */
 511         tmp = ENABLE_L1_TLB | ENABLE_L1_FRAGMENT_PROCESSING |
 512                 SYSTEM_ACCESS_MODE_NOT_IN_SYS |
 513                 EFFECTIVE_L1_TLB_SIZE(5) | EFFECTIVE_L1_QUEUE_SIZE(5) |
 514                 ENABLE_WAIT_L2_QUERY;
 515         WREG32(MC_VM_L1_TLB_MCB_RD_SYS_CNTL, tmp);
 516         WREG32(MC_VM_L1_TLB_MCB_WR_SYS_CNTL, tmp);
 517         WREG32(MC_VM_L1_TLB_MCB_RD_HDP_CNTL, tmp | ENABLE_L1_STRICT_ORDERING);
 518         WREG32(MC_VM_L1_TLB_MCB_WR_HDP_CNTL, tmp);
 519         WREG32(MC_VM_L1_TLB_MCD_RD_A_CNTL, tmp);
 520         WREG32(MC_VM_L1_TLB_MCD_WR_A_CNTL, tmp);
 521         WREG32(MC_VM_L1_TLB_MCD_RD_B_CNTL, tmp);
 522         WREG32(MC_VM_L1_TLB_MCD_WR_B_CNTL, tmp);
 523         WREG32(MC_VM_L1_TLB_MCB_RD_GFX_CNTL, tmp);
 524         WREG32(MC_VM_L1_TLB_MCB_WR_GFX_CNTL, tmp);
 525         WREG32(MC_VM_L1_TLB_MCB_RD_PDMA_CNTL, tmp);
 526         WREG32(MC_VM_L1_TLB_MCB_WR_PDMA_CNTL, tmp);
 527         WREG32(MC_VM_L1_TLB_MCB_RD_SEM_CNTL, tmp | ENABLE_SEMAPHORE_MODE);
 528         WREG32(MC_VM_L1_TLB_MCB_WR_SEM_CNTL, tmp | ENABLE_SEMAPHORE_MODE);
 529         for (i = 0; i < 7; i++)
 530                 WREG32(VM_CONTEXT0_CNTL + (i * 4), 0);
 531 }
 532
 533 int r600_mc_wait_for_idle(struct radeon_device *rdev)
 534 {
 535         unsigned i;
 536         u32 tmp;
 537
 538         for (i = 0; i < rdev->usec_timeout; i++) {
 539                 /* read MC_STATUS */
 540                 tmp = RREG32(R_000E50_SRBM_STATUS) & 0x3F00;
 541                 if (!tmp)
 542                         return 0;
 543                 udelay(1);
 544         }
 545         return -1;
 546 }
 547
 548 static void r600_mc_program(struct radeon_device *rdev)
 549 {
 550         struct rv515_mc_save save;
 551         u32 tmp;
 552         int i, j;
 553
 554         /* Initialize HDP */
 555         for (i = 0, j = 0; i < 32; i++, j += 0x18) {
 556                 WREG32((0x2c14 + j), 0x00000000);
 557                 WREG32((0x2c18 + j), 0x00000000);
 558                 WREG32((0x2c1c + j), 0x00000000);
 559                 WREG32((0x2c20 + j), 0x00000000);
 560                 WREG32((0x2c24 + j), 0x00000000);
 561         }
 562         WREG32(HDP_REG_COHERENCY_FLUSH_CNTL, 0);
 563
 564         rv515_mc_stop(rdev, &save);
 565         if (r600_mc_wait_for_idle(rdev)) {
 566                 dev_warn(rdev->dev, "Wait for MC idle timedout !\n");
 567         }
 568         /* Lockout access through VGA aperture (doesn't exist before R600) */
 569         WREG32(VGA_HDP_CONTROL, VGA_MEMORY_DISABLE);
 570         /* Update configuration */
 571         if (rdev->flags & RADEON_IS_AGP) {
 572                 if (rdev->mc.vram_start < rdev->mc.gtt_start) {
 573                         /* VRAM before AGP */
 574                         WREG32(MC_VM_SYSTEM_APERTURE_LOW_ADDR,
 575                                 rdev->mc.vram_start >> 12);
 576                         WREG32(MC_VM_SYSTEM_APERTURE_HIGH_ADDR,
 577                                 rdev->mc.gtt_end >> 12);
 578                 } else {
 579                         /* VRAM after AGP */
 580                         WREG32(MC_VM_SYSTEM_APERTURE_LOW_ADDR,
 581                                 rdev->mc.gtt_start >> 12);
 582                         WREG32(MC_VM_SYSTEM_APERTURE_HIGH_ADDR,
 583                                 rdev->mc.vram_end >> 12);
 584                 }
 585         } else {
 586                 WREG32(MC_VM_SYSTEM_APERTURE_LOW_ADDR, rdev->mc.vram_start >> 12);
 587                 WREG32(MC_VM_SYSTEM_APERTURE_HIGH_ADDR, rdev->mc.vram_end >> 12);
 588         }
 589         WREG32(MC_VM_SYSTEM_APERTURE_DEFAULT_ADDR, 0);
 590         tmp = ((rdev->mc.vram_end >> 24) & 0xFFFF) << 16;
 591         tmp |= ((rdev->mc.vram_start >> 24) & 0xFFFF);
 592         WREG32(MC_VM_FB_LOCATION, tmp);
 593         WREG32(HDP_NONSURFACE_BASE, (rdev->mc.vram_start >> 8));
 594         WREG32(HDP_NONSURFACE_INFO, (2 << 7));
 595         WREG32(HDP_NONSURFACE_SIZE, rdev->mc.mc_vram_size | 0x3FF);
 596         if (rdev->flags & RADEON_IS_AGP) {
 597                 WREG32(MC_VM_AGP_TOP, rdev->mc.gtt_end >> 22);
 598                 WREG32(MC_VM_AGP_BOT, rdev->mc.gtt_start >> 22);
 599                 WREG32(MC_VM_AGP_BASE, rdev->mc.agp_base >> 22);
 600         } else {
 601                 WREG32(MC_VM_AGP_BASE, 0);
 602                 WREG32(MC_VM_AGP_TOP, 0x0FFFFFFF);
 603                 WREG32(MC_VM_AGP_BOT, 0x0FFFFFFF);
 604         }
 605         if (r600_mc_wait_for_idle(rdev)) {
 606                 dev_warn(rdev->dev, "Wait for MC idle timedout !\n");
 607         }
 608         rv515_mc_resume(rdev, &save);
 609         /* we need to own VRAM, so turn off the VGA renderer here
 610          * to stop it overwriting our objects */
 611         rv515_vga_render_disable(rdev);
 612 }
 613
 614 /**
 615  * r600_vram_gtt_location - try to find VRAM & GTT location
 616  * @rdev: radeon device structure holding all necessary informations
 617  * @mc: memory controller structure holding memory informations
 618  *
 619  * Function will place try to place VRAM at same place as in CPU (PCI)
 620  * address space as some GPU seems to have issue when we reprogram at
 621  * different address space.
 622  *
 623  * If there is not enough space to fit the unvisible VRAM after the
 624  * aperture then we limit the VRAM size to the aperture.
 625  *
 626  * If we are using AGP then place VRAM adjacent to AGP aperture are we need
 627  * them to be in one from GPU point of view so that we can program GPU to
 628  * catch access outside them (weird GPU policy see ??).
 629  *
 630  * This function will never fails, worst case are limiting VRAM or GTT.
 631  *
 632  * Note: GTT start, end, size should be initialized before calling this
 633  * function on AGP platform.
 634  */
 635 void r600_vram_gtt_location(struct radeon_device *rdev, struct radeon_mc *mc)
 636 {
 637         u64 size_bf, size_af;
 638
 639         if (mc->mc_vram_size > 0xE0000000) {
 640                 /* leave room for at least 512M GTT */
 641                 dev_warn(rdev->dev, "limiting VRAM\n");
 642                 mc->real_vram_size = 0xE0000000;
 643                 mc->mc_vram_size = 0xE0000000;
 644         }
 645         if (rdev->flags & RADEON_IS_AGP) {
 646                 size_bf = mc->gtt_start;
 647                 size_af = 0xFFFFFFFF - mc->gtt_end + 1;
 648                 if (size_bf > size_af) {
 649                         if (mc->mc_vram_size > size_bf) {
 650                                 dev_warn(rdev->dev, "limiting VRAM\n");
 651                                 mc->real_vram_size = size_bf;
 652                                 mc->mc_vram_size = size_bf;
 653                         }
 654                         mc->vram_start = mc->gtt_start - mc->mc_vram_size;
 655                 } else {
 656                         if (mc->mc_vram_size > size_af) {
 657                                 dev_warn(rdev->dev, "limiting VRAM\n");
 658                                 mc->real_vram_size = size_af;
 659                                 mc->mc_vram_size = size_af;
 660                         }
 661                         mc->vram_start = mc->gtt_end;
 662                 }
 663                 mc->vram_end = mc->vram_start + mc->mc_vram_size - 1;
 664                 dev_info(rdev->dev, "VRAM: %lluM 0x%08llX - 0x%08llX (%lluM used)\n",
 665                                 mc->mc_vram_size >> 20, mc->vram_start,
 666                                 mc->vram_end, mc->real_vram_size >> 20);
 667         } else {
 668                 u64 base = 0;
 669                 if (rdev->flags & RADEON_IS_IGP)
 670                         base = (RREG32(MC_VM_FB_LOCATION) & 0xFFFF) << 24;
 671                 radeon_vram_location(rdev, &rdev->mc, base);
 672                 radeon_gtt_location(rdev, mc);
 673         }
 674 }
 675
 676 int r600_mc_init(struct radeon_device *rdev)
 677 {
 678         fixed20_12 a;
 679         u32 tmp;
 680         int chansize, numchan;
 681
 682         /* Get VRAM informations */
 683         rdev->mc.vram_is_ddr = true;
 684         tmp = RREG32(RAMCFG);
 685         if (tmp & CHANSIZE_OVERRIDE) {
 686                 chansize = 16;
 687         } else if (tmp & CHANSIZE_MASK) {
 688                 chansize = 64;
 689         } else {
 690                 chansize = 32;
 691         }
 692         tmp = RREG32(CHMAP);
 693         switch ((tmp & NOOFCHAN_MASK) >> NOOFCHAN_SHIFT) {
 694         case 0:
 695         default:
 696                 numchan = 1;
 697                 break;
 698         case 1:
 699                 numchan = 2;
 700                 break;
 701         case 2:
 702                 numchan = 4;
 703                 break;
 704         case 3:
 705                 numchan = 8;
 706                 break;
 707         }
 708         rdev->mc.vram_width = numchan * chansize;
 709         /* Could aper size report 0 ? */
 710         rdev->mc.aper_base = drm_get_resource_start(rdev->ddev, 0);
 711         rdev->mc.aper_size = drm_get_resource_len(rdev->ddev, 0);
 712         /* Setup GPU memory space */
 713         rdev->mc.mc_vram_size = RREG32(CONFIG_MEMSIZE);
 714         rdev->mc.real_vram_size = RREG32(CONFIG_MEMSIZE);
 715         rdev->mc.visible_vram_size = rdev->mc.aper_size;
 716         /* FIXME remove this once we support unmappable VRAM */
 717         if (rdev->mc.mc_vram_size > rdev->mc.aper_size) {
 718                 rdev->mc.mc_vram_size = rdev->mc.aper_size;
 719                 rdev->mc.real_vram_size = rdev->mc.aper_size;
 720         }
 721         r600_vram_gtt_location(rdev, &rdev->mc);
 722         /* FIXME: we should enforce default clock in case GPU is not in
 723          * default setup
 724          */
 725         a.full = rfixed_const(100);
 726         rdev->pm.sclk.full = rfixed_const(rdev->clock.default_sclk);
 727         rdev->pm.sclk.full = rfixed_div(rdev->pm.sclk, a);
 728         if (rdev->flags & RADEON_IS_IGP)
 729                 rdev->mc.igp_sideport_enabled = radeon_atombios_sideport_present(rdev);
 730         return 0;
 731 }
 732
 733 /* We doesn't check that the GPU really needs a reset we simply do the
 734  * reset, it's up to the caller to determine if the GPU needs one. We
 735  * might add an helper function to check that.
 736  */
 737 int r600_gpu_soft_reset(struct radeon_device *rdev)
 738 {
 739         struct rv515_mc_save save;
 740         u32 grbm_busy_mask = S_008010_VC_BUSY(1) | S_008010_VGT_BUSY_NO_DMA(1) |
 741                                 S_008010_VGT_BUSY(1) | S_008010_TA03_BUSY(1) |
 742                                 S_008010_TC_BUSY(1) | S_008010_SX_BUSY(1) |
 743                                 S_008010_SH_BUSY(1) | S_008010_SPI03_BUSY(1) |
 744                                 S_008010_SMX_BUSY(1) | S_008010_SC_BUSY(1) |
 745                                 S_008010_PA_BUSY(1) | S_008010_DB03_BUSY(1) |
 746                                 S_008010_CR_BUSY(1) | S_008010_CB03_BUSY(1) |
 747                                 S_008010_GUI_ACTIVE(1);
 748         u32 grbm2_busy_mask = S_008014_SPI0_BUSY(1) | S_008014_SPI1_BUSY(1) |
 749                         S_008014_SPI2_BUSY(1) | S_008014_SPI3_BUSY(1) |
 750                         S_008014_TA0_BUSY(1) | S_008014_TA1_BUSY(1) |
 751                         S_008014_TA2_BUSY(1) | S_008014_TA3_BUSY(1) |
 752                         S_008014_DB0_BUSY(1) | S_008014_DB1_BUSY(1) |
 753                         S_008014_DB2_BUSY(1) | S_008014_DB3_BUSY(1) |
 754                         S_008014_CB0_BUSY(1) | S_008014_CB1_BUSY(1) |
 755                         S_008014_CB2_BUSY(1) | S_008014_CB3_BUSY(1);
 756         u32 srbm_reset = 0;
 757         u32 tmp;
 758
 759         dev_info(rdev->dev, "GPU softreset \n");
 760         dev_info(rdev->dev, "  R_008010_GRBM_STATUS=0x%08X\n",
 761                 RREG32(R_008010_GRBM_STATUS));
 762         dev_info(rdev->dev, "  R_008014_GRBM_STATUS2=0x%08X\n",
 763                 RREG32(R_008014_GRBM_STATUS2));
 764         dev_info(rdev->dev, "  R_000E50_SRBM_STATUS=0x%08X\n",
 765                 RREG32(R_000E50_SRBM_STATUS));
 766         rv515_mc_stop(rdev, &save);
 767         if (r600_mc_wait_for_idle(rdev)) {
 768                 dev_warn(rdev->dev, "Wait for MC idle timedout !\n");
 769         }
 770         /* Disable CP parsing/prefetching */
 771         WREG32(R_0086D8_CP_ME_CNTL, S_0086D8_CP_ME_HALT(0xff));
 772         /* Check if any of the rendering block is busy and reset it */
 773         if ((RREG32(R_008010_GRBM_STATUS) & grbm_busy_mask) ||
 774             (RREG32(R_008014_GRBM_STATUS2) & grbm2_busy_mask)) {
 775                 tmp = S_008020_SOFT_RESET_CR(1) |
 776                         S_008020_SOFT_RESET_DB(1) |
 777                         S_008020_SOFT_RESET_CB(1) |
 778                         S_008020_SOFT_RESET_PA(1) |
 779                         S_008020_SOFT_RESET_SC(1) |
 780                         S_008020_SOFT_RESET_SMX(1) |
 781                         S_008020_SOFT_RESET_SPI(1) |
 782                         S_008020_SOFT_RESET_SX(1) |
 783                         S_008020_SOFT_RESET_SH(1) |
 784                         S_008020_SOFT_RESET_TC(1) |
 785                         S_008020_SOFT_RESET_TA(1) |
 786                         S_008020_SOFT_RESET_VC(1) |
 787                         S_008020_SOFT_RESET_VGT(1);
 788                 dev_info(rdev->dev, "  R_008020_GRBM_SOFT_RESET=0x%08X\n", tmp);
 789                 WREG32(R_008020_GRBM_SOFT_RESET, tmp);
 790                 (void)RREG32(R_008020_GRBM_SOFT_RESET);
 791                 udelay(50);
 792                 WREG32(R_008020_GRBM_SOFT_RESET, 0);
 793                 (void)RREG32(R_008020_GRBM_SOFT_RESET);
 794         }
 795         /* Reset CP (we always reset CP) */
 796         tmp = S_008020_SOFT_RESET_CP(1);
 797         dev_info(rdev->dev, "R_008020_GRBM_SOFT_RESET=0x%08X\n", tmp);
 798         WREG32(R_008020_GRBM_SOFT_RESET, tmp);
 799         (void)RREG32(R_008020_GRBM_SOFT_RESET);
 800         udelay(50);
 801         WREG32(R_008020_GRBM_SOFT_RESET, 0);
 802         (void)RREG32(R_008020_GRBM_SOFT_RESET);
 803         /* Reset others GPU block if necessary */
 804         if (G_000E50_RLC_BUSY(RREG32(R_000E50_SRBM_STATUS)))
 805                 srbm_reset |= S_000E60_SOFT_RESET_RLC(1);
 806         if (G_000E50_GRBM_RQ_PENDING(RREG32(R_000E50_SRBM_STATUS)))
 807                 srbm_reset |= S_000E60_SOFT_RESET_GRBM(1);
 808         if (G_000E50_HI_RQ_PENDING(RREG32(R_000E50_SRBM_STATUS)))
 809                 srbm_reset |= S_000E60_SOFT_RESET_IH(1);
 810         if (G_000E50_VMC_BUSY(RREG32(R_000E50_SRBM_STATUS)))
 811                 srbm_reset |= S_000E60_SOFT_RESET_VMC(1);
 812         if (G_000E50_MCB_BUSY(RREG32(R_000E50_SRBM_STATUS)))
 813                 srbm_reset |= S_000E60_SOFT_RESET_MC(1);
 814         if (G_000E50_MCDZ_BUSY(RREG32(R_000E50_SRBM_STATUS)))
 815                 srbm_reset |= S_000E60_SOFT_RESET_MC(1);
 816         if (G_000E50_MCDY_BUSY(RREG32(R_000E50_SRBM_STATUS)))
 817                 srbm_reset |= S_000E60_SOFT_RESET_MC(1);
 818         if (G_000E50_MCDX_BUSY(RREG32(R_000E50_SRBM_STATUS)))
 819                 srbm_reset |= S_000E60_SOFT_RESET_MC(1);
 820         if (G_000E50_MCDW_BUSY(RREG32(R_000E50_SRBM_STATUS)))
 821                 srbm_reset |= S_000E60_SOFT_RESET_MC(1);
 822         if (G_000E50_RLC_BUSY(RREG32(R_000E50_SRBM_STATUS)))
 823                 srbm_reset |= S_000E60_SOFT_RESET_RLC(1);
 824         if (G_000E50_SEM_BUSY(RREG32(R_000E50_SRBM_STATUS)))
 825                 srbm_reset |= S_000E60_SOFT_RESET_SEM(1);
 826         if (G_000E50_BIF_BUSY(RREG32(R_000E50_SRBM_STATUS)))
 827                 srbm_reset |= S_000E60_SOFT_RESET_BIF(1);
 828         dev_info(rdev->dev, "  R_000E60_SRBM_SOFT_RESET=0x%08X\n", srbm_reset);
 829         WREG32(R_000E60_SRBM_SOFT_RESET, srbm_reset);
 830         (void)RREG32(R_000E60_SRBM_SOFT_RESET);
 831         udelay(50);
 832         WREG32(R_000E60_SRBM_SOFT_RESET, 0);
 833         (void)RREG32(R_000E60_SRBM_SOFT_RESET);
 834         WREG32(R_000E60_SRBM_SOFT_RESET, srbm_reset);
 835         (void)RREG32(R_000E60_SRBM_SOFT_RESET);
 836         udelay(50);
 837         WREG32(R_000E60_SRBM_SOFT_RESET, 0);
 838         (void)RREG32(R_000E60_SRBM_SOFT_RESET);
 839         /* Wait a little for things to settle down */
 840         udelay(50);
 841         dev_info(rdev->dev, "  R_008010_GRBM_STATUS=0x%08X\n",
 842                 RREG32(R_008010_GRBM_STATUS));
 843         dev_info(rdev->dev, "  R_008014_GRBM_STATUS2=0x%08X\n",
 844                 RREG32(R_008014_GRBM_STATUS2));
 845         dev_info(rdev->dev, "  R_000E50_SRBM_STATUS=0x%08X\n",
 846                 RREG32(R_000E50_SRBM_STATUS));
 847         /* After reset we need to reinit the asic as GPU often endup in an
 848          * incoherent state.
 849          */
 850         atom_asic_init(rdev->mode_info.atom_context);
 851         rv515_mc_resume(rdev, &save);
 852         return 0;
 853 }
 854
 855 int r600_gpu_reset(struct radeon_device *rdev)
 856 {
 857         return r600_gpu_soft_reset(rdev);
 858 }
 859
 860 static u32 r600_get_tile_pipe_to_backend_map(u32 num_tile_pipes,
 861                                              u32 num_backends,
 862                                              u32 backend_disable_mask)
 863 {
 864         u32 backend_map = 0;
 865         u32 enabled_backends_mask;
 866         u32 enabled_backends_count;
 867         u32 cur_pipe;
 868         u32 swizzle_pipe[R6XX_MAX_PIPES];
 869         u32 cur_backend;
 870         u32 i;
 871
 872         if (num_tile_pipes > R6XX_MAX_PIPES)
 873                 num_tile_pipes = R6XX_MAX_PIPES;
 874         if (num_tile_pipes < 1)
 875                 num_tile_pipes = 1;
 876         if (num_backends > R6XX_MAX_BACKENDS)
 877                 num_backends = R6XX_MAX_BACKENDS;
 878         if (num_backends < 1)
 879                 num_backends = 1;
 880
 881         enabled_backends_mask = 0;
 882         enabled_backends_count = 0;
 883         for (i = 0; i < R6XX_MAX_BACKENDS; ++i) {
 884                 if (((backend_disable_mask >> i) & 1) == 0) {
 885                         enabled_backends_mask |= (1 << i);
 886                         ++enabled_backends_count;
 887                 }
 888                 if (enabled_backends_count == num_backends)
 889                         break;
 890         }
 891
 892         if (enabled_backends_count == 0) {
 893                 enabled_backends_mask = 1;
 894                 enabled_backends_count = 1;
 895         }
 896
 897         if (enabled_backends_count != num_backends)
 898                 num_backends = enabled_backends_count;
 899
 900         memset((uint8_t *)&swizzle_pipe[0], 0, sizeof(u32) * R6XX_MAX_PIPES);
 901         switch (num_tile_pipes) {
 902         case 1:
 903                 swizzle_pipe[0] = 0;
 904                 break;
 905         case 2:
 906                 swizzle_pipe[0] = 0;
 907                 swizzle_pipe[1] = 1;
 908                 break;
 909         case 3:
 910                 swizzle_pipe[0] = 0;
 911                 swizzle_pipe[1] = 1;
 912                 swizzle_pipe[2] = 2;
 913                 break;
 914         case 4:
 915                 swizzle_pipe[0] = 0;
 916                 swizzle_pipe[1] = 1;
 917                 swizzle_pipe[2] = 2;
 918                 swizzle_pipe[3] = 3;
 919                 break;
 920         case 5:
 921                 swizzle_pipe[0] = 0;
 922                 swizzle_pipe[1] = 1;
 923                 swizzle_pipe[2] = 2;
 924                 swizzle_pipe[3] = 3;
 925                 swizzle_pipe[4] = 4;
 926                 break;
 927         case 6:
 928                 swizzle_pipe[0] = 0;
 929                 swizzle_pipe[1] = 2;
 930                 swizzle_pipe[2] = 4;
 931                 swizzle_pipe[3] = 5;
 932                 swizzle_pipe[4] = 1;
 933                 swizzle_pipe[5] = 3;
 934                 break;
 935         case 7:
 936                 swizzle_pipe[0] = 0;
 937                 swizzle_pipe[1] = 2;
 938                 swizzle_pipe[2] = 4;
 939                 swizzle_pipe[3] = 6;
 940                 swizzle_pipe[4] = 1;
 941                 swizzle_pipe[5] = 3;
 942                 swizzle_pipe[6] = 5;
 943                 break;
 944         case 8:
 945                 swizzle_pipe[0] = 0;
 946                 swizzle_pipe[1] = 2;
 947                 swizzle_pipe[2] = 4;
 948                 swizzle_pipe[3] = 6;
 949                 swizzle_pipe[4] = 1;
 950                 swizzle_pipe[5] = 3;
 951                 swizzle_pipe[6] = 5;
 952                 swizzle_pipe[7] = 7;
 953                 break;
 954         }
 955
 956         cur_backend = 0;
 957         for (cur_pipe = 0; cur_pipe < num_tile_pipes; ++cur_pipe) {
 958                 while (((1 << cur_backend) & enabled_backends_mask) == 0)
 959                         cur_backend = (cur_backend + 1) % R6XX_MAX_BACKENDS;
 960
 961                 backend_map |= (u32)(((cur_backend & 3) << (swizzle_pipe[cur_pipe] * 2)));
 962
 963                 cur_backend = (cur_backend + 1) % R6XX_MAX_BACKENDS;
 964         }
 965
 966         return backend_map;
 967 }
 968
 969 int r600_count_pipe_bits(uint32_t val)
 970 {
 971         int i, ret = 0;
 972
 973         for (i = 0; i < 32; i++) {
 974                 ret += val & 1;
 975                 val >>= 1;
 976         }
 977         return ret;
 978 }
 979
 980 void r600_gpu_init(struct radeon_device *rdev)
 981 {
 982         u32 tiling_config;
 983         u32 ramcfg;
 984         u32 backend_map;
 985         u32 cc_rb_backend_disable;
 986         u32 cc_gc_shader_pipe_config;
 987         u32 tmp;
 988         int i, j;
 989         u32 sq_config;
 990         u32 sq_gpr_resource_mgmt_1 = 0;
 991         u32 sq_gpr_resource_mgmt_2 = 0;
 992         u32 sq_thread_resource_mgmt = 0;
 993         u32 sq_stack_resource_mgmt_1 = 0;
 994         u32 sq_stack_resource_mgmt_2 = 0;
 995
 996         /* FIXME: implement */
 997         switch (rdev->family) {
 998         case CHIP_R600:
 999                 rdev->config.r600.max_pipes = 4;
1000                 rdev->config.r600.max_tile_pipes = 8;
1001                 rdev->config.r600.max_simds = 4;
1002                 rdev->config.r600.max_backends = 4;
1003                 rdev->config.r600.max_gprs = 256;
1004                 rdev->config.r600.max_threads = 192;
1005                 rdev->config.r600.max_stack_entries = 256;
1006                 rdev->config.r600.max_hw_contexts = 8;
1007                 rdev->config.r600.max_gs_threads = 16;
1008                 rdev->config.r600.sx_max_export_size = 128;
1009                 rdev->config.r600.sx_max_export_pos_size = 16;
1010                 rdev->config.r600.sx_max_export_smx_size = 128;
1011                 rdev->config.r600.sq_num_cf_insts = 2;
1012                 break;
1013         case CHIP_RV630:
1014         case CHIP_RV635:
1015                 rdev->config.r600.max_pipes = 2;
1016                 rdev->config.r600.max_tile_pipes = 2;
1017                 rdev->config.r600.max_simds = 3;
1018                 rdev->config.r600.max_backends = 1;
1019                 rdev->config.r600.max_gprs = 128;
1020                 rdev->config.r600.max_threads = 192;
1021                 rdev->config.r600.max_stack_entries = 128;
1022                 rdev->config.r600.max_hw_contexts = 8;
1023                 rdev->config.r600.max_gs_threads = 4;
1024                 rdev->config.r600.sx_max_export_size = 128;
1025                 rdev->config.r600.sx_max_export_pos_size = 16;
1026                 rdev->config.r600.sx_max_export_smx_size = 128;
1027                 rdev->config.r600.sq_num_cf_insts = 2;
1028                 break;
1029         case CHIP_RV610:
1030         case CHIP_RV620:
1031         case CHIP_RS780:
1032         case CHIP_RS880:
1033                 rdev->config.r600.max_pipes = 1;
1034                 rdev->config.r600.max_tile_pipes = 1;
1035                 rdev->config.r600.max_simds = 2;
1036                 rdev->config.r600.max_backends = 1;
1037                 rdev->config.r600.max_gprs = 128;
1038                 rdev->config.r600.max_threads = 192;
1039                 rdev->config.r600.max_stack_entries = 128;
1040                 rdev->config.r600.max_hw_contexts = 4;
1041                 rdev->config.r600.max_gs_threads = 4;
1042                 rdev->config.r600.sx_max_export_size = 128;
1043                 rdev->config.r600.sx_max_export_pos_size = 16;
1044                 rdev->config.r600.sx_max_export_smx_size = 128;
1045                 rdev->config.r600.sq_num_cf_insts = 1;
1046                 break;
1047         case CHIP_RV670:
1048                 rdev->config.r600.max_pipes = 4;
1049                 rdev->config.r600.max_tile_pipes = 4;
1050                 rdev->config.r600.max_simds = 4;
1051                 rdev->config.r600.max_backends = 4;
1052                 rdev->config.r600.max_gprs = 192;
1053                 rdev->config.r600.max_threads = 192;
1054                 rdev->config.r600.max_stack_entries = 256;
1055                 rdev->config.r600.max_hw_contexts = 8;
1056                 rdev->config.r600.max_gs_threads = 16;
1057                 rdev->config.r600.sx_max_export_size = 128;
1058                 rdev->config.r600.sx_max_export_pos_size = 16;
1059                 rdev->config.r600.sx_max_export_smx_size = 128;
1060                 rdev->config.r600.sq_num_cf_insts = 2;
1061                 break;
1062         default:
1063                 break;
1064         }
1065
1066         /* Initialize HDP */
1067         for (i = 0, j = 0; i < 32; i++, j += 0x18) {
1068                 WREG32((0x2c14 + j), 0x00000000);
1069                 WREG32((0x2c18 + j), 0x00000000);
1070                 WREG32((0x2c1c + j), 0x00000000);
1071                 WREG32((0x2c20 + j), 0x00000000);
1072                 WREG32((0x2c24 + j), 0x00000000);
1073         }
1074
1075         WREG32(GRBM_CNTL, GRBM_READ_TIMEOUT(0xff));
1076
1077         /* Setup tiling */
1078         tiling_config = 0;
1079         ramcfg = RREG32(RAMCFG);
1080         switch (rdev->config.r600.max_tile_pipes) {
1081         case 1:
1082                 tiling_config |= PIPE_TILING(0);
1083                 break;
1084         case 2:
1085                 tiling_config |= PIPE_TILING(1);
1086                 break;
1087         case 4:
1088                 tiling_config |= PIPE_TILING(2);
1089                 break;
1090         case 8:
1091                 tiling_config |= PIPE_TILING(3);
1092                 break;
1093         default:
1094                 break;
1095         }
1096         rdev->config.r600.tiling_npipes = rdev->config.r600.max_tile_pipes;
1097         rdev->config.r600.tiling_nbanks = 4 << ((ramcfg & NOOFBANK_MASK) >> NOOFBANK_SHIFT);
1098         tiling_config |= BANK_TILING((ramcfg & NOOFBANK_MASK) >> NOOFBANK_SHIFT);
1099         tiling_config |= GROUP_SIZE(0);
1100         rdev->config.r600.tiling_group_size = 256;
1101         tmp = (ramcfg & NOOFROWS_MASK) >> NOOFROWS_SHIFT;
1102         if (tmp > 3) {
1103                 tiling_config |= ROW_TILING(3);
1104                 tiling_config |= SAMPLE_SPLIT(3);
1105         } else {
1106                 tiling_config |= ROW_TILING(tmp);
1107                 tiling_config |= SAMPLE_SPLIT(tmp);
1108         }
1109         tiling_config |= BANK_SWAPS(1);
1110
1111         cc_rb_backend_disable = RREG32(CC_RB_BACKEND_DISABLE) & 0x00ff0000;
1112         cc_rb_backend_disable |=
1113                 BACKEND_DISABLE((R6XX_MAX_BACKENDS_MASK << rdev->config.r600.max_backends) & R6XX_MAX_BACKENDS_MASK);
1114
1115         cc_gc_shader_pipe_config = RREG32(CC_GC_SHADER_PIPE_CONFIG) & 0xffffff00;
1116         cc_gc_shader_pipe_config |=
1117                 INACTIVE_QD_PIPES((R6XX_MAX_PIPES_MASK << rdev->config.r600.max_pipes) & R6XX_MAX_PIPES_MASK);
1118         cc_gc_shader_pipe_config |=
1119                 INACTIVE_SIMDS((R6XX_MAX_SIMDS_MASK << rdev->config.r600.max_simds) & R6XX_MAX_SIMDS_MASK);
1120
1121         backend_map = r600_get_tile_pipe_to_backend_map(rdev->config.r600.max_tile_pipes,
1122                                                         (R6XX_MAX_BACKENDS -
1123                                                          r600_count_pipe_bits((cc_rb_backend_disable &
1124                                                                                R6XX_MAX_BACKENDS_MASK) >> 16)),
1125                                                         (cc_rb_backend_disable >> 16));
1126
1127         tiling_config |= BACKEND_MAP(backend_map);
1128         WREG32(GB_TILING_CONFIG, tiling_config);
1129         WREG32(DCP_TILING_CONFIG, tiling_config & 0xffff);
1130         WREG32(HDP_TILING_CONFIG, tiling_config & 0xffff);
1131
1132         /* Setup pipes */
1133         WREG32(CC_RB_BACKEND_DISABLE, cc_rb_backend_disable);
1134         WREG32(CC_GC_SHADER_PIPE_CONFIG, cc_gc_shader_pipe_config);
1135
1136         tmp = R6XX_MAX_PIPES - r600_count_pipe_bits((cc_gc_shader_pipe_config & INACTIVE_QD_PIPES_MASK) >> 8);
1137         WREG32(VGT_OUT_DEALLOC_CNTL, (tmp * 4) & DEALLOC_DIST_MASK);
1138         WREG32(VGT_VERTEX_REUSE_BLOCK_CNTL, ((tmp * 4) - 2) & VTX_REUSE_DEPTH_MASK);
1139
1140         /* Setup some CP states */
1141         WREG32(CP_QUEUE_THRESHOLDS, (ROQ_IB1_START(0x16) | ROQ_IB2_START(0x2b)));
1142         WREG32(CP_MEQ_THRESHOLDS, (MEQ_END(0x40) | ROQ_END(0x40)));
1143
1144         WREG32(TA_CNTL_AUX, (DISABLE_CUBE_ANISO | SYNC_GRADIENT |
1145                              SYNC_WALKER | SYNC_ALIGNER));
1146         /* Setup various GPU states */
1147         if (rdev->family == CHIP_RV670)
1148                 WREG32(ARB_GDEC_RD_CNTL, 0x00000021);
1149
1150         tmp = RREG32(SX_DEBUG_1);
1151         tmp |= SMX_EVENT_RELEASE;
1152         if ((rdev->family > CHIP_R600))
1153                 tmp |= ENABLE_NEW_SMX_ADDRESS;
1154         WREG32(SX_DEBUG_1, tmp);
1155
1156         if (((rdev->family) == CHIP_R600) ||
1157             ((rdev->family) == CHIP_RV630) ||
1158             ((rdev->family) == CHIP_RV610) ||
1159             ((rdev->family) == CHIP_RV620) ||
1160             ((rdev->family) == CHIP_RS780) ||
1161             ((rdev->family) == CHIP_RS880)) {
1162                 WREG32(DB_DEBUG, PREZ_MUST_WAIT_FOR_POSTZ_DONE);
1163         } else {
1164                 WREG32(DB_DEBUG, 0);
1165         }
1166         WREG32(DB_WATERMARKS, (DEPTH_FREE(4) | DEPTH_CACHELINE_FREE(16) |
1167                                DEPTH_FLUSH(16) | DEPTH_PENDING_FREE(4)));
1168
1169         WREG32(PA_SC_MULTI_CHIP_CNTL, 0);
1170         WREG32(VGT_NUM_INSTANCES, 0);
1171
1172         WREG32(SPI_CONFIG_CNTL, GPR_WRITE_PRIORITY(0));
1173         WREG32(SPI_CONFIG_CNTL_1, VTX_DONE_DELAY(0));
1174
1175         tmp = RREG32(SQ_MS_FIFO_SIZES);
1176         if (((rdev->family) == CHIP_RV610) ||
1177             ((rdev->family) == CHIP_RV620) ||
1178             ((rdev->family) == CHIP_RS780) ||
1179             ((rdev->family) == CHIP_RS880)) {
1180                 tmp = (CACHE_FIFO_SIZE(0xa) |
1181                        FETCH_FIFO_HIWATER(0xa) |
1182                        DONE_FIFO_HIWATER(0xe0) |
1183                        ALU_UPDATE_FIFO_HIWATER(0x8));
1184         } else if (((rdev->family) == CHIP_R600) ||
1185                    ((rdev->family) == CHIP_RV630)) {
1186                 tmp &= ~DONE_FIFO_HIWATER(0xff);
1187                 tmp |= DONE_FIFO_HIWATER(0x4);
1188         }
1189         WREG32(SQ_MS_FIFO_SIZES, tmp);
1190
1191         /* SQ_CONFIG, SQ_GPR_RESOURCE_MGMT, SQ_THREAD_RESOURCE_MGMT, SQ_STACK_RESOURCE_MGMT
1192          * should be adjusted as needed by the 2D/3D drivers.  This just sets default values
1193          */
1194         sq_config = RREG32(SQ_CONFIG);
1195         sq_config &= ~(PS_PRIO(3) |
1196                        VS_PRIO(3) |
1197                        GS_PRIO(3) |
1198                        ES_PRIO(3));
1199         sq_config |= (DX9_CONSTS |
1200                       VC_ENABLE |
1201                       PS_PRIO(0) |
1202                       VS_PRIO(1) |
1203                       GS_PRIO(2) |
1204                       ES_PRIO(3));
1205
1206         if ((rdev->family) == CHIP_R600) {
1207                 sq_gpr_resource_mgmt_1 = (NUM_PS_GPRS(124) |
1208                                           NUM_VS_GPRS(124) |
1209                                           NUM_CLAUSE_TEMP_GPRS(4));
1210                 sq_gpr_resource_mgmt_2 = (NUM_GS_GPRS(0) |
1211                                           NUM_ES_GPRS(0));
1212                 sq_thread_resource_mgmt = (NUM_PS_THREADS(136) |
1213                                            NUM_VS_THREADS(48) |
1214                                            NUM_GS_THREADS(4) |
1215                                            NUM_ES_THREADS(4));
1216                 sq_stack_resource_mgmt_1 = (NUM_PS_STACK_ENTRIES(128) |
1217                                             NUM_VS_STACK_ENTRIES(128));
1218                 sq_stack_resource_mgmt_2 = (NUM_GS_STACK_ENTRIES(0) |
1219                                             NUM_ES_STACK_ENTRIES(0));
1220         } else if (((rdev->family) == CHIP_RV610) ||
1221                    ((rdev->family) == CHIP_RV620) ||
1222                    ((rdev->family) == CHIP_RS780) ||
1223                    ((rdev->family) == CHIP_RS880)) {
1224                 /* no vertex cache */
1225                 sq_config &= ~VC_ENABLE;
1226
1227                 sq_gpr_resource_mgmt_1 = (NUM_PS_GPRS(44) |
1228                                           NUM_VS_GPRS(44) |
1229                                           NUM_CLAUSE_TEMP_GPRS(2));
1230                 sq_gpr_resource_mgmt_2 = (NUM_GS_GPRS(17) |
1231                                           NUM_ES_GPRS(17));
1232                 sq_thread_resource_mgmt = (NUM_PS_THREADS(79) |
1233                                            NUM_VS_THREADS(78) |
1234                                            NUM_GS_THREADS(4) |
1235                                            NUM_ES_THREADS(31));
1236                 sq_stack_resource_mgmt_1 = (NUM_PS_STACK_ENTRIES(40) |
1237                                             NUM_VS_STACK_ENTRIES(40));
1238                 sq_stack_resource_mgmt_2 = (NUM_GS_STACK_ENTRIES(32) |
1239                                             NUM_ES_STACK_ENTRIES(16));
1240         } else if (((rdev->family) == CHIP_RV630) ||
1241                    ((rdev->family) == CHIP_RV635)) {
1242                 sq_gpr_resource_mgmt_1 = (NUM_PS_GPRS(44) |
1243                                           NUM_VS_GPRS(44) |
1244                                           NUM_CLAUSE_TEMP_GPRS(2));
1245                 sq_gpr_resource_mgmt_2 = (NUM_GS_GPRS(18) |
1246                                           NUM_ES_GPRS(18));
1247                 sq_thread_resource_mgmt = (NUM_PS_THREADS(79) |
1248                                            NUM_VS_THREADS(78) |
1249                                            NUM_GS_THREADS(4) |
1250                                            NUM_ES_THREADS(31));
1251                 sq_stack_resource_mgmt_1 = (NUM_PS_STACK_ENTRIES(40) |
1252                                             NUM_VS_STACK_ENTRIES(40));
1253                 sq_stack_resource_mgmt_2 = (NUM_GS_STACK_ENTRIES(32) |
1254                                             NUM_ES_STACK_ENTRIES(16));
1255         } else if ((rdev->family) == CHIP_RV670) {
1256                 sq_gpr_resource_mgmt_1 = (NUM_PS_GPRS(44) |
1257                                           NUM_VS_GPRS(44) |
1258                                           NUM_CLAUSE_TEMP_GPRS(2));
1259                 sq_gpr_resource_mgmt_2 = (NUM_GS_GPRS(17) |
1260                                           NUM_ES_GPRS(17));
1261                 sq_thread_resource_mgmt = (NUM_PS_THREADS(79) |
1262                                            NUM_VS_THREADS(78) |
1263                                            NUM_GS_THREADS(4) |
1264                                            NUM_ES_THREADS(31));
1265                 sq_stack_resource_mgmt_1 = (NUM_PS_STACK_ENTRIES(64) |
1266                                             NUM_VS_STACK_ENTRIES(64));
1267                 sq_stack_resource_mgmt_2 = (NUM_GS_STACK_ENTRIES(64) |
1268                                             NUM_ES_STACK_ENTRIES(64));
1269         }
1270
1271         WREG32(SQ_CONFIG, sq_config);
1272         WREG32(SQ_GPR_RESOURCE_MGMT_1,  sq_gpr_resource_mgmt_1);
1273         WREG32(SQ_GPR_RESOURCE_MGMT_2,  sq_gpr_resource_mgmt_2);
1274         WREG32(SQ_THREAD_RESOURCE_MGMT, sq_thread_resource_mgmt);
1275         WREG32(SQ_STACK_RESOURCE_MGMT_1, sq_stack_resource_mgmt_1);
1276         WREG32(SQ_STACK_RESOURCE_MGMT_2, sq_stack_resource_mgmt_2);
1277
1278         if (((rdev->family) == CHIP_RV610) ||
1279             ((rdev->family) == CHIP_RV620) ||
1280             ((rdev->family) == CHIP_RS780) ||
1281             ((rdev->family) == CHIP_RS880)) {
1282                 WREG32(VGT_CACHE_INVALIDATION, CACHE_INVALIDATION(TC_ONLY));
1283         } else {
1284                 WREG32(VGT_CACHE_INVALIDATION, CACHE_INVALIDATION(VC_AND_TC));
1285         }
1286
1287         /* More default values. 2D/3D driver should adjust as needed */
1288         WREG32(PA_SC_AA_SAMPLE_LOCS_2S, (S0_X(0xc) | S0_Y(0x4) |
1289                                          S1_X(0x4) | S1_Y(0xc)));
1290         WREG32(PA_SC_AA_SAMPLE_LOCS_4S, (S0_X(0xe) | S0_Y(0xe) |
1291                                          S1_X(0x2) | S1_Y(0x2) |
1292                                          S2_X(0xa) | S2_Y(0x6) |
1293                                          S3_X(0x6) | S3_Y(0xa)));
1294         WREG32(PA_SC_AA_SAMPLE_LOCS_8S_WD0, (S0_X(0xe) | S0_Y(0xb) |
1295                                              S1_X(0x4) | S1_Y(0xc) |
1296                                              S2_X(0x1) | S2_Y(0x6) |
1297                                              S3_X(0xa) | S3_Y(0xe)));
1298         WREG32(PA_SC_AA_SAMPLE_LOCS_8S_WD1, (S4_X(0x6) | S4_Y(0x1) |
1299                                              S5_X(0x0) | S5_Y(0x0) |
1300                                              S6_X(0xb) | S6_Y(0x4) |
1301                                              S7_X(0x7) | S7_Y(0x8)));
1302
1303         WREG32(VGT_STRMOUT_EN, 0);
1304         tmp = rdev->config.r600.max_pipes * 16;
1305         switch (rdev->family) {
1306         case CHIP_RV610:
1307         case CHIP_RV620:
1308         case CHIP_RS780:
1309         case CHIP_RS880:
1310                 tmp += 32;
1311                 break;
1312         case CHIP_RV670:
1313                 tmp += 128;
1314                 break;
1315         default:
1316                 break;
1317         }
1318         if (tmp > 256) {
1319                 tmp = 256;
1320         }
1321         WREG32(VGT_ES_PER_GS, 128);
1322         WREG32(VGT_GS_PER_ES, tmp);
1323         WREG32(VGT_GS_PER_VS, 2);
1324         WREG32(VGT_GS_VERTEX_REUSE, 16);
1325
1326         /* more default values. 2D/3D driver should adjust as needed */
1327         WREG32(PA_SC_LINE_STIPPLE_STATE, 0);
1328         WREG32(VGT_STRMOUT_EN, 0);
1329         WREG32(SX_MISC, 0);
1330         WREG32(PA_SC_MODE_CNTL, 0);
1331         WREG32(PA_SC_AA_CONFIG, 0);
1332         WREG32(PA_SC_LINE_STIPPLE, 0);
1333         WREG32(SPI_INPUT_Z, 0);
1334         WREG32(SPI_PS_IN_CONTROL_0, NUM_INTERP(2));
1335         WREG32(CB_COLOR7_FRAG, 0);
1336
1337         /* Clear render buffer base addresses */
1338         WREG32(CB_COLOR0_BASE, 0);
1339         WREG32(CB_COLOR1_BASE, 0);
1340         WREG32(CB_COLOR2_BASE, 0);
1341         WREG32(CB_COLOR3_BASE, 0);
1342         WREG32(CB_COLOR4_BASE, 0);
1343         WREG32(CB_COLOR5_BASE, 0);
1344         WREG32(CB_COLOR6_BASE, 0);
1345         WREG32(CB_COLOR7_BASE, 0);
1346         WREG32(CB_COLOR7_FRAG, 0);
1347
1348         switch (rdev->family) {
1349         case CHIP_RV610:
1350         case CHIP_RV620:
1351         case CHIP_RS780:
1352         case CHIP_RS880:
1353                 tmp = TC_L2_SIZE(8);
1354                 break;
1355         case CHIP_RV630:
1356         case CHIP_RV635:
1357                 tmp = TC_L2_SIZE(4);
1358                 break;
1359         case CHIP_R600:
1360                 tmp = TC_L2_SIZE(0) | L2_DISABLE_LATE_HIT;
1361                 break;
1362         default:
1363                 tmp = TC_L2_SIZE(0);
1364                 break;
1365         }
1366         WREG32(TC_CNTL, tmp);
1367
1368         tmp = RREG32(HDP_HOST_PATH_CNTL);
1369         WREG32(HDP_HOST_PATH_CNTL, tmp);
1370
1371         tmp = RREG32(ARB_POP);
1372         tmp |= ENABLE_TC128;
1373         WREG32(ARB_POP, tmp);
1374
1375         WREG32(PA_SC_MULTI_CHIP_CNTL, 0);
1376         WREG32(PA_CL_ENHANCE, (CLIP_VTX_REORDER_ENA |
1377                                NUM_CLIP_SEQ(3)));
1378         WREG32(PA_SC_ENHANCE, FORCE_EOV_MAX_CLK_CNT(4095));
1379 }
1380
1381
1382 /*
1383  * Indirect registers accessor
1384  */
1385 u32 r600_pciep_rreg(struct radeon_device *rdev, u32 reg)
1386 {
1387         u32 r;
1388
1389         WREG32(PCIE_PORT_INDEX, ((reg) & 0xff));
1390         (void)RREG32(PCIE_PORT_INDEX);
1391         r = RREG32(PCIE_PORT_DATA);
1392         return r;
1393 }
1394
1395 void r600_pciep_wreg(struct radeon_device *rdev, u32 reg, u32 v)
1396 {
1397         WREG32(PCIE_PORT_INDEX, ((reg) & 0xff));
1398         (void)RREG32(PCIE_PORT_INDEX);
1399         WREG32(PCIE_PORT_DATA, (v));
1400         (void)RREG32(PCIE_PORT_DATA);
1401 }
1402
1403 /*
1404  * CP & Ring
1405  */
1406 void r600_cp_stop(struct radeon_device *rdev)
1407 {
1408         WREG32(R_0086D8_CP_ME_CNTL, S_0086D8_CP_ME_HALT(1));
1409 }
1410
1411 int r600_init_microcode(struct radeon_device *rdev)
1412 {
1413         struct platform_device *pdev;
1414         const char *chip_name;
1415         const char *rlc_chip_name;
1416         size_t pfp_req_size, me_req_size, rlc_req_size;
1417         char fw_name[30];
1418         int err;
1419
1420         DRM_DEBUG("\n");
1421
1422         pdev = platform_device_register_simple("radeon_cp", 0, NULL, 0);
1423         err = IS_ERR(pdev);
1424         if (err) {
1425                 printk(KERN_ERR "radeon_cp: Failed to register firmware\n");
1426                 return -EINVAL;
1427         }
1428
1429         switch (rdev->family) {
1430         case CHIP_R600:
1431                 chip_name = "R600";
1432                 rlc_chip_name = "R600";
1433                 break;
1434         case CHIP_RV610:
1435                 chip_name = "RV610";
1436                 rlc_chip_name = "R600";
1437                 break;
1438         case CHIP_RV630:
1439                 chip_name = "RV630";
1440                 rlc_chip_name = "R600";
1441                 break;
1442         case CHIP_RV620:
1443                 chip_name = "RV620";
1444                 rlc_chip_name = "R600";
1445                 break;
1446         case CHIP_RV635:
1447                 chip_name = "RV635";
1448                 rlc_chip_name = "R600";
1449                 break;
1450         case CHIP_RV670:
1451                 chip_name = "RV670";
1452                 rlc_chip_name = "R600";
1453                 break;
1454         case CHIP_RS780:
1455         case CHIP_RS880:
1456                 chip_name = "RS780";
1457                 rlc_chip_name = "R600";
1458                 break;
1459         case CHIP_RV770:
1460                 chip_name = "RV770";
1461                 rlc_chip_name = "R700";
1462                 break;
1463         case CHIP_RV730:
1464         case CHIP_RV740:
1465                 chip_name = "RV730";
1466                 rlc_chip_name = "R700";
1467                 break;
1468         case CHIP_RV710:
1469                 chip_name = "RV710";
1470                 rlc_chip_name = "R700";
1471                 break;
1472         default: BUG();
1473         }
1474
1475         if (rdev->family >= CHIP_RV770) {
1476                 pfp_req_size = R700_PFP_UCODE_SIZE * 4;
1477                 me_req_size = R700_PM4_UCODE_SIZE * 4;
1478                 rlc_req_size = R700_RLC_UCODE_SIZE * 4;
1479         } else {
1480                 pfp_req_size = PFP_UCODE_SIZE * 4;
1481                 me_req_size = PM4_UCODE_SIZE * 12;
1482                 rlc_req_size = RLC_UCODE_SIZE * 4;
1483         }
1484
1485         DRM_INFO("Loading %s Microcode\n", chip_name);
1486
1487         snprintf(fw_name, sizeof(fw_name), "radeon/%s_pfp.bin", chip_name);
1488         err = request_firmware(&rdev->pfp_fw, fw_name, &pdev->dev);
1489         if (err)
1490                 goto out;
1491         if (rdev->pfp_fw->size != pfp_req_size) {
1492                 printk(KERN_ERR
1493                        "r600_cp: Bogus length %zu in firmware \"%s\"\n",
1494                        rdev->pfp_fw->size, fw_name);
1495                 err = -EINVAL;
1496                 goto out;
1497         }
1498
1499         snprintf(fw_name, sizeof(fw_name), "radeon/%s_me.bin", chip_name);
1500         err = request_firmware(&rdev->me_fw, fw_name, &pdev->dev);
1501         if (err)
1502                 goto out;
1503         if (rdev->me_fw->size != me_req_size) {
1504                 printk(KERN_ERR
1505                        "r600_cp: Bogus length %zu in firmware \"%s\"\n",
1506                        rdev->me_fw->size, fw_name);
1507                 err = -EINVAL;
1508         }
1509
1510         snprintf(fw_name, sizeof(fw_name), "radeon/%s_rlc.bin", rlc_chip_name);
1511         err = request_firmware(&rdev->rlc_fw, fw_name, &pdev->dev);
1512         if (err)
1513                 goto out;
1514         if (rdev->rlc_fw->size != rlc_req_size) {
1515                 printk(KERN_ERR
1516                        "r600_rlc: Bogus length %zu in firmware \"%s\"\n",
1517                        rdev->rlc_fw->size, fw_name);
1518                 err = -EINVAL;
1519         }
1520
1521 out:
1522         platform_device_unregister(pdev);
1523
1524         if (err) {
1525                 if (err != -EINVAL)
1526                         printk(KERN_ERR
1527                                "r600_cp: Failed to load firmware \"%s\"\n",
1528                                fw_name);
1529                 release_firmware(rdev->pfp_fw);
1530                 rdev->pfp_fw = NULL;
1531                 release_firmware(rdev->me_fw);
1532                 rdev->me_fw = NULL;
1533                 release_firmware(rdev->rlc_fw);
1534                 rdev->rlc_fw = NULL;
1535         }
1536         return err;
1537 }
1538
1539 static int r600_cp_load_microcode(struct radeon_device *rdev)
1540 {
1541         const __be32 *fw_data;
1542         int i;
1543
1544         if (!rdev->me_fw || !rdev->pfp_fw)
1545                 return -EINVAL;
1546
1547         r600_cp_stop(rdev);
1548
1549         WREG32(CP_RB_CNTL, RB_NO_UPDATE | RB_BLKSZ(15) | RB_BUFSZ(3));
1550
1551         /* Reset cp */
1552         WREG32(GRBM_SOFT_RESET, SOFT_RESET_CP);
1553         RREG32(GRBM_SOFT_RESET);
1554         mdelay(15);
1555         WREG32(GRBM_SOFT_RESET, 0);
1556
1557         WREG32(CP_ME_RAM_WADDR, 0);
1558
1559         fw_data = (const __be32 *)rdev->me_fw->data;
1560         WREG32(CP_ME_RAM_WADDR, 0);
1561         for (i = 0; i < PM4_UCODE_SIZE * 3; i++)
1562                 WREG32(CP_ME_RAM_DATA,
1563                        be32_to_cpup(fw_data++));
1564
1565         fw_data = (const __be32 *)rdev->pfp_fw->data;
1566         WREG32(CP_PFP_UCODE_ADDR, 0);
1567         for (i = 0; i < PFP_UCODE_SIZE; i++)
1568                 WREG32(CP_PFP_UCODE_DATA,
1569                        be32_to_cpup(fw_data++));
1570
1571         WREG32(CP_PFP_UCODE_ADDR, 0);
1572         WREG32(CP_ME_RAM_WADDR, 0);
1573         WREG32(CP_ME_RAM_RADDR, 0);
1574         return 0;
1575 }
1576
1577 int r600_cp_start(struct radeon_device *rdev)
1578 {
1579         int r;
1580         uint32_t cp_me;
1581
1582         r = radeon_ring_lock(rdev, 7);
1583         if (r) {
1584                 DRM_ERROR("radeon: cp failed to lock ring (%d).\n", r);
1585                 return r;
1586         }
1587         radeon_ring_write(rdev, PACKET3(PACKET3_ME_INITIALIZE, 5));
1588         radeon_ring_write(rdev, 0x1);
1589         if (rdev->family < CHIP_RV770) {
1590                 radeon_ring_write(rdev, 0x3);
1591                 radeon_ring_write(rdev, rdev->config.r600.max_hw_contexts - 1);
1592         } else {
1593                 radeon_ring_write(rdev, 0x0);
1594                 radeon_ring_write(rdev, rdev->config.rv770.max_hw_contexts - 1);
1595         }
1596         radeon_ring_write(rdev, PACKET3_ME_INITIALIZE_DEVICE_ID(1));
1597         radeon_ring_write(rdev, 0);
1598         radeon_ring_write(rdev, 0);
1599         radeon_ring_unlock_commit(rdev);
1600
1601         cp_me = 0xff;
1602         WREG32(R_0086D8_CP_ME_CNTL, cp_me);
1603         return 0;
1604 }
1605
1606 int r600_cp_resume(struct radeon_device *rdev)
1607 {
1608         u32 tmp;
1609         u32 rb_bufsz;
1610         int r;
1611
1612         /* Reset cp */
1613         WREG32(GRBM_SOFT_RESET, SOFT_RESET_CP);
1614         RREG32(GRBM_SOFT_RESET);
1615         mdelay(15);
1616         WREG32(GRBM_SOFT_RESET, 0);
1617
1618         /* Set ring buffer size */
1619         rb_bufsz = drm_order(rdev->cp.ring_size / 8);
1620         tmp = RB_NO_UPDATE | (drm_order(RADEON_GPU_PAGE_SIZE/8) << 8) | rb_bufsz;
1621 #ifdef __BIG_ENDIAN
1622         tmp |= BUF_SWAP_32BIT;
1623 #endif
1624         WREG32(CP_RB_CNTL, tmp);
1625         WREG32(CP_SEM_WAIT_TIMER, 0x4);
1626
1627         /* Set the write pointer delay */
1628         WREG32(CP_RB_WPTR_DELAY, 0);
1629
1630         /* Initialize the ring buffer's read and write pointers */
1631         WREG32(CP_RB_CNTL, tmp | RB_RPTR_WR_ENA);
1632         WREG32(CP_RB_RPTR_WR, 0);
1633         WREG32(CP_RB_WPTR, 0);
1634         WREG32(CP_RB_RPTR_ADDR, rdev->cp.gpu_addr & 0xFFFFFFFF);
1635         WREG32(CP_RB_RPTR_ADDR_HI, upper_32_bits(rdev->cp.gpu_addr));
1636         mdelay(1);
1637         WREG32(CP_RB_CNTL, tmp);
1638
1639         WREG32(CP_RB_BASE, rdev->cp.gpu_addr >> 8);
1640         WREG32(CP_DEBUG, (1 << 27) | (1 << 28));
1641
1642         rdev->cp.rptr = RREG32(CP_RB_RPTR);
1643         rdev->cp.wptr = RREG32(CP_RB_WPTR);
1644
1645         r600_cp_start(rdev);
1646         rdev->cp.ready = true;
1647         r = radeon_ring_test(rdev);
1648         if (r) {
1649                 rdev->cp.ready = false;
1650                 return r;
1651         }
1652         return 0;
1653 }
1654
1655 void r600_cp_commit(struct radeon_device *rdev)
1656 {
1657         WREG32(CP_RB_WPTR, rdev->cp.wptr);
1658         (void)RREG32(CP_RB_WPTR);
1659 }
1660
1661 void r600_ring_init(struct radeon_device *rdev, unsigned ring_size)
1662 {
1663         u32 rb_bufsz;
1664
1665         /* Align ring size */
1666         rb_bufsz = drm_order(ring_size / 8);
1667         ring_size = (1 << (rb_bufsz + 1)) * 4;
1668         rdev->cp.ring_size = ring_size;
1669         rdev->cp.align_mask = 16 - 1;
1670 }
1671
1672 void r600_cp_fini(struct radeon_device *rdev)
1673 {
1674         r600_cp_stop(rdev);
1675         radeon_ring_fini(rdev);
1676 }
1677
1678
1679 /*
1680  * GPU scratch registers helpers function.
1681  */
1682 void r600_scratch_init(struct radeon_device *rdev)
1683 {
1684         int i;
1685
1686         rdev->scratch.num_reg = 7;
1687         for (i = 0; i < rdev->scratch.num_reg; i++) {
1688                 rdev->scratch.free[i] = true;
1689                 rdev->scratch.reg[i] = SCRATCH_REG0 + (i * 4);
1690         }
1691 }
1692
1693 int r600_ring_test(struct radeon_device *rdev)
1694 {
1695         uint32_t scratch;
1696         uint32_t tmp = 0;
1697         unsigned i;
1698         int r;
1699
1700         r = radeon_scratch_get(rdev, &scratch);
1701         if (r) {
1702                 DRM_ERROR("radeon: cp failed to get scratch reg (%d).\n", r);
1703                 return r;
1704         }
1705         WREG32(scratch, 0xCAFEDEAD);
1706         r = radeon_ring_lock(rdev, 3);
1707         if (r) {
1708                 DRM_ERROR("radeon: cp failed to lock ring (%d).\n", r);
1709                 radeon_scratch_free(rdev, scratch);
1710                 return r;
1711         }
1712         radeon_ring_write(rdev, PACKET3(PACKET3_SET_CONFIG_REG, 1));
1713         radeon_ring_write(rdev, ((scratch - PACKET3_SET_CONFIG_REG_OFFSET) >> 2));
1714         radeon_ring_write(rdev, 0xDEADBEEF);
1715         radeon_ring_unlock_commit(rdev);
1716         for (i = 0; i < rdev->usec_timeout; i++) {
1717                 tmp = RREG32(scratch);
1718                 if (tmp == 0xDEADBEEF)
1719                         break;
1720                 DRM_UDELAY(1);
1721         }
1722         if (i < rdev->usec_timeout) {
1723                 DRM_INFO("ring test succeeded in %d usecs\n", i);
1724         } else {
1725                 DRM_ERROR("radeon: ring test failed (scratch(0x%04X)=0x%08X)\n",
1726                           scratch, tmp);
1727                 r = -EINVAL;
1728         }
1729         radeon_scratch_free(rdev, scratch);
1730         return r;
1731 }
1732
1733 void r600_wb_disable(struct radeon_device *rdev)
1734 {
1735         int r;
1736
1737         WREG32(SCRATCH_UMSK, 0);
1738         if (rdev->wb.wb_obj) {
1739                 r = radeon_bo_reserve(rdev->wb.wb_obj, false);
1740                 if (unlikely(r != 0))
1741                         return;
1742                 radeon_bo_kunmap(rdev->wb.wb_obj);
1743                 radeon_bo_unpin(rdev->wb.wb_obj);
1744                 radeon_bo_unreserve(rdev->wb.wb_obj);
1745         }
1746 }
1747
1748 void r600_wb_fini(struct radeon_device *rdev)
1749 {
1750         r600_wb_disable(rdev);
1751         if (rdev->wb.wb_obj) {
1752                 radeon_bo_unref(&rdev->wb.wb_obj);
1753                 rdev->wb.wb = NULL;
1754                 rdev->wb.wb_obj = NULL;
1755         }
1756 }
1757
1758 int r600_wb_enable(struct radeon_device *rdev)
1759 {
1760         int r;
1761
1762         if (rdev->wb.wb_obj == NULL) {
1763                 r = radeon_bo_create(rdev, NULL, RADEON_GPU_PAGE_SIZE, true,
1764                                 RADEON_GEM_DOMAIN_GTT, &rdev->wb.wb_obj);
1765                 if (r) {
1766                         dev_warn(rdev->dev, "(%d) create WB bo failed\n", r);
1767                         return r;
1768                 }
1769                 r = radeon_bo_reserve(rdev->wb.wb_obj, false);
1770                 if (unlikely(r != 0)) {
1771                         r600_wb_fini(rdev);
1772                         return r;
1773                 }
1774                 r = radeon_bo_pin(rdev->wb.wb_obj, RADEON_GEM_DOMAIN_GTT,
1775                                 &rdev->wb.gpu_addr);
1776                 if (r) {
1777                         radeon_bo_unreserve(rdev->wb.wb_obj);
1778                         dev_warn(rdev->dev, "(%d) pin WB bo failed\n", r);
1779                         r600_wb_fini(rdev);
1780                         return r;
1781                 }
1782                 r = radeon_bo_kmap(rdev->wb.wb_obj, (void **)&rdev->wb.wb);
1783                 radeon_bo_unreserve(rdev->wb.wb_obj);
1784                 if (r) {
1785                         dev_warn(rdev->dev, "(%d) map WB bo failed\n", r);
1786                         r600_wb_fini(rdev);
1787                         return r;
1788                 }
1789         }
1790         WREG32(SCRATCH_ADDR, (rdev->wb.gpu_addr >> 8) & 0xFFFFFFFF);
1791         WREG32(CP_RB_RPTR_ADDR, (rdev->wb.gpu_addr + 1024) & 0xFFFFFFFC);
1792         WREG32(CP_RB_RPTR_ADDR_HI, upper_32_bits(rdev->wb.gpu_addr + 1024) & 0xFF);
1793         WREG32(SCRATCH_UMSK, 0xff);
1794         return 0;
1795 }
1796
1797 void r600_fence_ring_emit(struct radeon_device *rdev,
1798                           struct radeon_fence *fence)
1799 {
1800         /* Also consider EVENT_WRITE_EOP.  it handles the interrupts + timestamps + events */
1801
1802         radeon_ring_write(rdev, PACKET3(PACKET3_EVENT_WRITE, 0));
1803         radeon_ring_write(rdev, CACHE_FLUSH_AND_INV_EVENT);
1804         /* wait for 3D idle clean */
1805         radeon_ring_write(rdev, PACKET3(PACKET3_SET_CONFIG_REG, 1));
1806         radeon_ring_write(rdev, (WAIT_UNTIL - PACKET3_SET_CONFIG_REG_OFFSET) >> 2);
1807         radeon_ring_write(rdev, WAIT_3D_IDLE_bit | WAIT_3D_IDLECLEAN_bit);
1808         /* Emit fence sequence & fire IRQ */
1809         radeon_ring_write(rdev, PACKET3(PACKET3_SET_CONFIG_REG, 1));
1810         radeon_ring_write(rdev, ((rdev->fence_drv.scratch_reg - PACKET3_SET_CONFIG_REG_OFFSET) >> 2));
1811         radeon_ring_write(rdev, fence->seq);
1812         /* CP_INTERRUPT packet 3 no longer exists, use packet 0 */
1813         radeon_ring_write(rdev, PACKET0(CP_INT_STATUS, 0));
1814         radeon_ring_write(rdev, RB_INT_STAT);
1815 }
1816
1817 int r600_copy_blit(struct radeon_device *rdev,
1818                    uint64_t src_offset, uint64_t dst_offset,
1819                    unsigned num_pages, struct radeon_fence *fence)
1820 {
1821         int r;
1822
1823         mutex_lock(&rdev->r600_blit.mutex);
1824         rdev->r600_blit.vb_ib = NULL;
1825         r = r600_blit_prepare_copy(rdev, num_pages * RADEON_GPU_PAGE_SIZE);
1826         if (r) {
1827                 if (rdev->r600_blit.vb_ib)
1828                         radeon_ib_free(rdev, &rdev->r600_blit.vb_ib);
1829                 mutex_unlock(&rdev->r600_blit.mutex);
1830                 return r;
1831         }
1832         r600_kms_blit_copy(rdev, src_offset, dst_offset, num_pages * RADEON_GPU_PAGE_SIZE);
1833         r600_blit_done_copy(rdev, fence);
1834         mutex_unlock(&rdev->r600_blit.mutex);
1835         return 0;
1836 }
1837
1838 int r600_set_surface_reg(struct radeon_device *rdev, int reg,
1839                          uint32_t tiling_flags, uint32_t pitch,
1840                          uint32_t offset, uint32_t obj_size)
1841 {
1842         /* FIXME: implement */
1843         return 0;
1844 }
1845
1846 void r600_clear_surface_reg(struct radeon_device *rdev, int reg)
1847 {
1848         /* FIXME: implement */
1849 }
1850
1851
1852 bool r600_card_posted(struct radeon_device *rdev)
1853 {
1854         uint32_t reg;
1855
1856         /* first check CRTCs */
1857         reg = RREG32(D1CRTC_CONTROL) |
1858                 RREG32(D2CRTC_CONTROL);
1859         if (reg & CRTC_EN)
1860                 return true;
1861
1862         /* then check MEM_SIZE, in case the crtcs are off */
1863         if (RREG32(CONFIG_MEMSIZE))
1864                 return true;
1865
1866         return false;
1867 }
1868
1869 int r600_startup(struct radeon_device *rdev)
1870 {
1871         int r;
1872
1873         if (!rdev->me_fw || !rdev->pfp_fw || !rdev->rlc_fw) {
1874                 r = r600_init_microcode(rdev);
1875                 if (r) {
1876                         DRM_ERROR("Failed to load firmware!\n");
1877                         return r;
1878                 }
1879         }
1880
1881         r600_mc_program(rdev);
1882         if (rdev->flags & RADEON_IS_AGP) {
1883                 r600_agp_enable(rdev);
1884         } else {
1885                 r = r600_pcie_gart_enable(rdev);
1886                 if (r)
1887                         return r;
1888         }
1889         r600_gpu_init(rdev);
1890         r = r600_blit_init(rdev);
1891         if (r) {
1892                 r600_blit_fini(rdev);
1893                 rdev->asic->copy = NULL;
1894                 dev_warn(rdev->dev, "failed blitter (%d) falling back to memcpy\n", r);
1895         }
1896         /* pin copy shader into vram */
1897         if (rdev->r600_blit.shader_obj) {
1898                 r = radeon_bo_reserve(rdev->r600_blit.shader_obj, false);
1899                 if (unlikely(r != 0))
1900                         return r;
1901                 r = radeon_bo_pin(rdev->r600_blit.shader_obj, RADEON_GEM_DOMAIN_VRAM,
1902                                 &rdev->r600_blit.shader_gpu_addr);
1903                 radeon_bo_unreserve(rdev->r600_blit.shader_obj);
1904                 if (r) {
1905                         dev_err(rdev->dev, "(%d) pin blit object failed\n", r);
1906                         return r;
1907                 }
1908         }
1909         /* Enable IRQ */
1910         r = r600_irq_init(rdev);
1911         if (r) {
1912                 DRM_ERROR("radeon: IH init failed (%d).\n", r);
1913                 radeon_irq_kms_fini(rdev);
1914                 return r;
1915         }
1916         r600_irq_set(rdev);
1917
1918         r = radeon_ring_init(rdev, rdev->cp.ring_size);
1919         if (r)
1920                 return r;
1921         r = r600_cp_load_microcode(rdev);
1922         if (r)
1923                 return r;
1924         r = r600_cp_resume(rdev);
1925         if (r)
1926                 return r;
1927         /* write back buffer are not vital so don't worry about failure */
1928         r600_wb_enable(rdev);
1929         return 0;
1930 }
1931
1932 void r600_vga_set_state(struct radeon_device *rdev, bool state)
1933 {
1934         uint32_t temp;
1935
1936         temp = RREG32(CONFIG_CNTL);
1937         if (state == false) {
1938                 temp &= ~(1<<0);
1939                 temp |= (1<<1);
1940         } else {
1941                 temp &= ~(1<<1);
1942         }
1943         WREG32(CONFIG_CNTL, temp);
1944 }
1945
1946 int r600_resume(struct radeon_device *rdev)
1947 {
1948         int r;
1949
1950         /* Do not reset GPU before posting, on r600 hw unlike on r500 hw,
1951          * posting will perform necessary task to bring back GPU into good
1952          * shape.
1953          */
1954         /* post card */
1955         atom_asic_init(rdev->mode_info.atom_context);
1956         /* Initialize clocks */
1957         r = radeon_clocks_init(rdev);
1958         if (r) {
1959                 return r;
1960         }
1961
1962         r = r600_startup(rdev);
1963         if (r) {
1964                 DRM_ERROR("r600 startup failed on resume\n");
1965                 return r;
1966         }
1967
1968         r = r600_ib_test(rdev);
1969         if (r) {
1970                 DRM_ERROR("radeon: failled testing IB (%d).\n", r);
1971                 return r;
1972         }
1973
1974         r = r600_audio_init(rdev);
1975         if (r) {
1976                 DRM_ERROR("radeon: audio resume failed\n");
1977                 return r;
1978         }
1979
1980         return r;
1981 }
1982
1983 int r600_suspend(struct radeon_device *rdev)
1984 {
1985         int r;
1986
1987         r600_audio_fini(rdev);
1988         /* FIXME: we should wait for ring to be empty */
1989         r600_cp_stop(rdev);
1990         rdev->cp.ready = false;
1991         r600_irq_suspend(rdev);
1992         r600_wb_disable(rdev);
1993         r600_pcie_gart_disable(rdev);
1994         /* unpin shaders bo */
1995         if (rdev->r600_blit.shader_obj) {
1996                 r = radeon_bo_reserve(rdev->r600_blit.shader_obj, false);
1997                 if (!r) {
1998                         radeon_bo_unpin(rdev->r600_blit.shader_obj);
1999                         radeon_bo_unreserve(rdev->r600_blit.shader_obj);
2000                 }
2001         }
2002         return 0;
2003 }
2004
2005 /* Plan is to move initialization in that function and use
2006  * helper function so that radeon_device_init pretty much
2007  * do nothing more than calling asic specific function. This
2008  * should also allow to remove a bunch of callback function
2009  * like vram_info.
2010  */
2011 int r600_init(struct radeon_device *rdev)
2012 {
2013         int r;
2014
2015         r = radeon_dummy_page_init(rdev);
2016         if (r)
2017                 return r;
2018         if (r600_debugfs_mc_info_init(rdev)) {
2019                 DRM_ERROR("Failed to register debugfs file for mc !\n");
2020         }
2021         /* This don't do much */
2022         r = radeon_gem_init(rdev);
2023         if (r)
2024                 return r;
2025         /* Read BIOS */
2026         if (!radeon_get_bios(rdev)) {
2027                 if (ASIC_IS_AVIVO(rdev))
2028                         return -EINVAL;
2029         }
2030         /* Must be an ATOMBIOS */
2031         if (!rdev->is_atom_bios) {
2032                 dev_err(rdev->dev, "Expecting atombios for R600 GPU\n");
2033                 return -EINVAL;
2034         }
2035         r = radeon_atombios_init(rdev);
2036         if (r)
2037                 return r;
2038         /* Post card if necessary */
2039         if (!r600_card_posted(rdev)) {
2040                 if (!rdev->bios) {
2041                         dev_err(rdev->dev, "Card not posted and no BIOS - ignoring\n");
2042                         return -EINVAL;
2043                 }
2044                 DRM_INFO("GPU not posted. posting now...\n");
2045                 atom_asic_init(rdev->mode_info.atom_context);
2046         }
2047         /* Initialize scratch registers */
2048         r600_scratch_init(rdev);
2049         /* Initialize surface registers */
2050         radeon_surface_init(rdev);
2051         /* Initialize clocks */
2052         radeon_get_clock_info(rdev->ddev);
2053         r = radeon_clocks_init(rdev);
2054         if (r)
2055                 return r;
2056         /* Initialize power management */
2057         radeon_pm_init(rdev);
2058         /* Fence driver */
2059         r = radeon_fence_driver_init(rdev);
2060         if (r)
2061                 return r;
2062         if (rdev->flags & RADEON_IS_AGP) {
2063                 r = radeon_agp_init(rdev);
2064                 if (r)
2065                         radeon_agp_disable(rdev);
2066         }
2067         r = r600_mc_init(rdev);
2068         if (r)
2069                 return r;
2070         /* Memory manager */
2071         r = radeon_bo_init(rdev);
2072         if (r)
2073                 return r;
2074
2075         r = radeon_irq_kms_init(rdev);
2076         if (r)
2077                 return r;
2078
2079         rdev->cp.ring_obj = NULL;
2080         r600_ring_init(rdev, 1024 * 1024);
2081
2082         rdev->ih.ring_obj = NULL;
2083         r600_ih_ring_init(rdev, 64 * 1024);
2084
2085         r = r600_pcie_gart_init(rdev);
2086         if (r)
2087                 return r;
2088
2089         rdev->accel_working = true;
2090         r = r600_startup(rdev);
2091         if (r) {
2092                 dev_err(rdev->dev, "disabling GPU acceleration\n");
2093                 r600_cp_fini(rdev);
2094                 r600_wb_fini(rdev);
2095                 r600_irq_fini(rdev);
2096                 radeon_irq_kms_fini(rdev);
2097                 r600_pcie_gart_fini(rdev);
2098                 rdev->accel_working = false;
2099         }
2100         if (rdev->accel_working) {
2101                 r = radeon_ib_pool_init(rdev);
2102                 if (r) {
2103                         dev_err(rdev->dev, "IB initialization failed (%d).\n", r);
2104                         rdev->accel_working = false;
2105                 } else {
2106                         r = r600_ib_test(rdev);
2107                         if (r) {
2108                                 dev_err(rdev->dev, "IB test failed (%d).\n", r);
2109                                 rdev->accel_working = false;
2110                         }
2111                 }
2112         }
2113
2114         r = r600_audio_init(rdev);
2115         if (r)
2116                 return r; /* TODO error handling */
2117         return 0;
2118 }
2119
2120 void r600_fini(struct radeon_device *rdev)
2121 {
2122         r600_audio_fini(rdev);
2123         r600_blit_fini(rdev);
2124         r600_cp_fini(rdev);
2125         r600_wb_fini(rdev);
2126         r600_irq_fini(rdev);
2127         radeon_irq_kms_fini(rdev);
2128         r600_pcie_gart_fini(rdev);
2129         radeon_agp_fini(rdev);
2130         radeon_gem_fini(rdev);
2131         radeon_fence_driver_fini(rdev);
2132         radeon_clocks_fini(rdev);
2133         radeon_bo_fini(rdev);
2134         radeon_atombios_fini(rdev);
2135         kfree(rdev->bios);
2136         rdev->bios = NULL;
2137         radeon_dummy_page_fini(rdev);
2138 }
2139
2140
2141 /*
2142  * CS stuff
2143  */
2144 void r600_ring_ib_execute(struct radeon_device *rdev, struct radeon_ib *ib)
2145 {
2146         /* FIXME: implement */
2147         radeon_ring_write(rdev, PACKET3(PACKET3_INDIRECT_BUFFER, 2));
2148         radeon_ring_write(rdev, ib->gpu_addr & 0xFFFFFFFC);
2149         radeon_ring_write(rdev, upper_32_bits(ib->gpu_addr) & 0xFF);
2150         radeon_ring_write(rdev, ib->length_dw);
2151 }
2152
2153 int r600_ib_test(struct radeon_device *rdev)
2154 {
2155         struct radeon_ib *ib;
2156         uint32_t scratch;
2157         uint32_t tmp = 0;
2158         unsigned i;
2159         int r;
2160
2161         r = radeon_scratch_get(rdev, &scratch);
2162         if (r) {
2163                 DRM_ERROR("radeon: failed to get scratch reg (%d).\n", r);
2164                 return r;
2165         }
2166         WREG32(scratch, 0xCAFEDEAD);
2167         r = radeon_ib_get(rdev, &ib);
2168         if (r) {
2169                 DRM_ERROR("radeon: failed to get ib (%d).\n", r);
2170                 return r;
2171         }
2172         ib->ptr[0] = PACKET3(PACKET3_SET_CONFIG_REG, 1);
2173         ib->ptr[1] = ((scratch - PACKET3_SET_CONFIG_REG_OFFSET) >> 2);
2174         ib->ptr[2] = 0xDEADBEEF;
2175         ib->ptr[3] = PACKET2(0);
2176         ib->ptr[4] = PACKET2(0);
2177         ib->ptr[5] = PACKET2(0);
2178         ib->ptr[6] = PACKET2(0);
2179         ib->ptr[7] = PACKET2(0);
2180         ib->ptr[8] = PACKET2(0);
2181         ib->ptr[9] = PACKET2(0);
2182         ib->ptr[10] = PACKET2(0);
2183         ib->ptr[11] = PACKET2(0);
2184         ib->ptr[12] = PACKET2(0);
2185         ib->ptr[13] = PACKET2(0);
2186         ib->ptr[14] = PACKET2(0);
2187         ib->ptr[15] = PACKET2(0);
2188         ib->length_dw = 16;
2189         r = radeon_ib_schedule(rdev, ib);
2190         if (r) {
2191                 radeon_scratch_free(rdev, scratch);
2192                 radeon_ib_free(rdev, &ib);
2193                 DRM_ERROR("radeon: failed to schedule ib (%d).\n", r);
2194                 return r;
2195         }
2196         r = radeon_fence_wait(ib->fence, false);
2197         if (r) {
2198                 DRM_ERROR("radeon: fence wait failed (%d).\n", r);
2199                 return r;
2200         }
2201         for (i = 0; i < rdev->usec_timeout; i++) {
2202                 tmp = RREG32(scratch);
2203                 if (tmp == 0xDEADBEEF)
2204                         break;
2205                 DRM_UDELAY(1);
2206         }
2207         if (i < rdev->usec_timeout) {
2208                 DRM_INFO("ib test succeeded in %u usecs\n", i);
2209         } else {
2210                 DRM_ERROR("radeon: ib test failed (sracth(0x%04X)=0x%08X)\n",
2211                           scratch, tmp);
2212                 r = -EINVAL;
2213         }
2214         radeon_scratch_free(rdev, scratch);
2215         radeon_ib_free(rdev, &ib);
2216         return r;
2217 }
2218
2219 /*
2220  * Interrupts
2221  *
2222  * Interrupts use a ring buffer on r6xx/r7xx hardware.  It works pretty
2223  * the same as the CP ring buffer, but in reverse.  Rather than the CPU
2224  * writing to the ring and the GPU consuming, the GPU writes to the ring
2225  * and host consumes.  As the host irq handler processes interrupts, it
2226  * increments the rptr.  When the rptr catches up with the wptr, all the
2227  * current interrupts have been processed.
2228  */
2229
2230 void r600_ih_ring_init(struct radeon_device *rdev, unsigned ring_size)
2231 {
2232         u32 rb_bufsz;
2233
2234         /* Align ring size */
2235         rb_bufsz = drm_order(ring_size / 4);
2236         ring_size = (1 << rb_bufsz) * 4;
2237         rdev->ih.ring_size = ring_size;
2238         rdev->ih.ptr_mask = rdev->ih.ring_size - 1;
2239         rdev->ih.rptr = 0;
2240 }
2241
2242 static int r600_ih_ring_alloc(struct radeon_device *rdev)
2243 {
2244         int r;
2245
2246         /* Allocate ring buffer */
2247         if (rdev->ih.ring_obj == NULL) {
2248                 r = radeon_bo_create(rdev, NULL, rdev->ih.ring_size,
2249                                      true,
2250                                      RADEON_GEM_DOMAIN_GTT,
2251                                      &rdev->ih.ring_obj);
2252                 if (r) {
2253                         DRM_ERROR("radeon: failed to create ih ring buffer (%d).\n", r);
2254                         return r;
2255                 }
2256                 r = radeon_bo_reserve(rdev->ih.ring_obj, false);
2257                 if (unlikely(r != 0))
2258                         return r;
2259                 r = radeon_bo_pin(rdev->ih.ring_obj,
2260                                   RADEON_GEM_DOMAIN_GTT,
2261                                   &rdev->ih.gpu_addr);
2262                 if (r) {
2263                         radeon_bo_unreserve(rdev->ih.ring_obj);
2264                         DRM_ERROR("radeon: failed to pin ih ring buffer (%d).\n", r);
2265                         return r;
2266                 }
2267                 r = radeon_bo_kmap(rdev->ih.ring_obj,
2268                                    (void **)&rdev->ih.ring);
2269                 radeon_bo_unreserve(rdev->ih.ring_obj);
2270                 if (r) {
2271                         DRM_ERROR("radeon: failed to map ih ring buffer (%d).\n", r);
2272                         return r;
2273                 }
2274         }
2275         return 0;
2276 }
2277
2278 static void r600_ih_ring_fini(struct radeon_device *rdev)
2279 {
2280         int r;
2281         if (rdev->ih.ring_obj) {
2282                 r = radeon_bo_reserve(rdev->ih.ring_obj, false);
2283                 if (likely(r == 0)) {
2284                         radeon_bo_kunmap(rdev->ih.ring_obj);
2285                         radeon_bo_unpin(rdev->ih.ring_obj);
2286                         radeon_bo_unreserve(rdev->ih.ring_obj);
2287                 }
2288                 radeon_bo_unref(&rdev->ih.ring_obj);
2289                 rdev->ih.ring = NULL;
2290                 rdev->ih.ring_obj = NULL;
2291         }
2292 }
2293
2294 static void r600_rlc_stop(struct radeon_device *rdev)
2295 {
2296
2297         if (rdev->family >= CHIP_RV770) {
2298                 /* r7xx asics need to soft reset RLC before halting */
2299                 WREG32(SRBM_SOFT_RESET, SOFT_RESET_RLC);
2300                 RREG32(SRBM_SOFT_RESET);
2301                 udelay(15000);
2302                 WREG32(SRBM_SOFT_RESET, 0);
2303                 RREG32(SRBM_SOFT_RESET);
2304         }
2305
2306         WREG32(RLC_CNTL, 0);
2307 }
2308
2309 static void r600_rlc_start(struct radeon_device *rdev)
2310 {
2311         WREG32(RLC_CNTL, RLC_ENABLE);
2312 }
2313
2314 static int r600_rlc_init(struct radeon_device *rdev)
2315 {
2316         u32 i;
2317         const __be32 *fw_data;
2318
2319         if (!rdev->rlc_fw)
2320                 return -EINVAL;
2321
2322         r600_rlc_stop(rdev);
2323
2324         WREG32(RLC_HB_BASE, 0);
2325         WREG32(RLC_HB_CNTL, 0);
2326         WREG32(RLC_HB_RPTR, 0);
2327         WREG32(RLC_HB_WPTR, 0);
2328         WREG32(RLC_HB_WPTR_LSB_ADDR, 0);
2329         WREG32(RLC_HB_WPTR_MSB_ADDR, 0);
2330         WREG32(RLC_MC_CNTL, 0);
2331         WREG32(RLC_UCODE_CNTL, 0);
2332
2333         fw_data = (const __be32 *)rdev->rlc_fw->data;
2334         if (rdev->family >= CHIP_RV770) {
2335                 for (i = 0; i < R700_RLC_UCODE_SIZE; i++) {
2336                         WREG32(RLC_UCODE_ADDR, i);
2337                         WREG32(RLC_UCODE_DATA, be32_to_cpup(fw_data++));
2338                 }
2339         } else {
2340                 for (i = 0; i < RLC_UCODE_SIZE; i++) {
2341                         WREG32(RLC_UCODE_ADDR, i);
2342                         WREG32(RLC_UCODE_DATA, be32_to_cpup(fw_data++));
2343                 }
2344         }
2345         WREG32(RLC_UCODE_ADDR, 0);
2346
2347         r600_rlc_start(rdev);
2348
2349         return 0;
2350 }
2351
2352 static void r600_enable_interrupts(struct radeon_device *rdev)
2353 {
2354         u32 ih_cntl = RREG32(IH_CNTL);
2355         u32 ih_rb_cntl = RREG32(IH_RB_CNTL);
2356
2357         ih_cntl |= ENABLE_INTR;
2358         ih_rb_cntl |= IH_RB_ENABLE;
2359         WREG32(IH_CNTL, ih_cntl);
2360         WREG32(IH_RB_CNTL, ih_rb_cntl);
2361         rdev->ih.enabled = true;
2362 }
2363
2364 static void r600_disable_interrupts(struct radeon_device *rdev)
2365 {
2366         u32 ih_rb_cntl = RREG32(IH_RB_CNTL);
2367         u32 ih_cntl = RREG32(IH_CNTL);
2368
2369         ih_rb_cntl &= ~IH_RB_ENABLE;
2370         ih_cntl &= ~ENABLE_INTR;
2371         WREG32(IH_RB_CNTL, ih_rb_cntl);
2372         WREG32(IH_CNTL, ih_cntl);
2373         /* set rptr, wptr to 0 */
2374         WREG32(IH_RB_RPTR, 0);
2375         WREG32(IH_RB_WPTR, 0);
2376         rdev->ih.enabled = false;
2377         rdev->ih.wptr = 0;
2378         rdev->ih.rptr = 0;
2379 }
2380
2381 static void r600_disable_interrupt_state(struct radeon_device *rdev)
2382 {
2383         u32 tmp;
2384
2385         WREG32(CP_INT_CNTL, 0);
2386         WREG32(GRBM_INT_CNTL, 0);
2387         WREG32(DxMODE_INT_MASK, 0);
2388         if (ASIC_IS_DCE3(rdev)) {
2389                 WREG32(DCE3_DACA_AUTODETECT_INT_CONTROL, 0);
2390                 WREG32(DCE3_DACB_AUTODETECT_INT_CONTROL, 0);
2391                 tmp = RREG32(DC_HPD1_INT_CONTROL) & DC_HPDx_INT_POLARITY;
2392                 WREG32(DC_HPD1_INT_CONTROL, tmp);
2393                 tmp = RREG32(DC_HPD2_INT_CONTROL) & DC_HPDx_INT_POLARITY;
2394                 WREG32(DC_HPD2_INT_CONTROL, tmp);
2395                 tmp = RREG32(DC_HPD3_INT_CONTROL) & DC_HPDx_INT_POLARITY;
2396                 WREG32(DC_HPD3_INT_CONTROL, tmp);
2397                 tmp = RREG32(DC_HPD4_INT_CONTROL) & DC_HPDx_INT_POLARITY;
2398                 WREG32(DC_HPD4_INT_CONTROL, tmp);
2399                 if (ASIC_IS_DCE32(rdev)) {
2400                         tmp = RREG32(DC_HPD5_INT_CONTROL) & DC_HPDx_INT_POLARITY;
2401                         WREG32(DC_HPD5_INT_CONTROL, 0);
2402                         tmp = RREG32(DC_HPD6_INT_CONTROL) & DC_HPDx_INT_POLARITY;
2403                         WREG32(DC_HPD6_INT_CONTROL, 0);
2404                 }
2405         } else {
2406                 WREG32(DACA_AUTODETECT_INT_CONTROL, 0);
2407                 WREG32(DACB_AUTODETECT_INT_CONTROL, 0);
2408                 tmp = RREG32(DC_HOT_PLUG_DETECT1_INT_CONTROL) & DC_HOT_PLUG_DETECTx_INT_POLARITY;
2409                 WREG32(DC_HOT_PLUG_DETECT1_INT_CONTROL, 0);
2410                 tmp = RREG32(DC_HOT_PLUG_DETECT2_INT_CONTROL) & DC_HOT_PLUG_DETECTx_INT_POLARITY;
2411                 WREG32(DC_HOT_PLUG_DETECT2_INT_CONTROL, 0);
2412                 tmp = RREG32(DC_HOT_PLUG_DETECT3_INT_CONTROL) & DC_HOT_PLUG_DETECTx_INT_POLARITY;
2413                 WREG32(DC_HOT_PLUG_DETECT3_INT_CONTROL, 0);
2414         }
2415 }
2416
2417 int r600_irq_init(struct radeon_device *rdev)
2418 {
2419         int ret = 0;
2420         int rb_bufsz;
2421         u32 interrupt_cntl, ih_cntl, ih_rb_cntl;
2422
2423         /* allocate ring */
2424         ret = r600_ih_ring_alloc(rdev);
2425         if (ret)
2426                 return ret;
2427
2428         /* disable irqs */
2429         r600_disable_interrupts(rdev);
2430
2431         /* init rlc */
2432         ret = r600_rlc_init(rdev);
2433         if (ret) {
2434                 r600_ih_ring_fini(rdev);
2435                 return ret;
2436         }
2437
2438         /* setup interrupt control */
2439         /* set dummy read address to ring address */
2440         WREG32(INTERRUPT_CNTL2, rdev->ih.gpu_addr >> 8);
2441         interrupt_cntl = RREG32(INTERRUPT_CNTL);
2442         /* IH_DUMMY_RD_OVERRIDE=0 - dummy read disabled with msi, enabled without msi
2443          * IH_DUMMY_RD_OVERRIDE=1 - dummy read controlled by IH_DUMMY_RD_EN
2444          */
2445         interrupt_cntl &= ~IH_DUMMY_RD_OVERRIDE;
2446         /* IH_REQ_NONSNOOP_EN=1 if ring is in non-cacheable memory, e.g., vram */
2447         interrupt_cntl &= ~IH_REQ_NONSNOOP_EN;
2448         WREG32(INTERRUPT_CNTL, interrupt_cntl);
2449
2450         WREG32(IH_RB_BASE, rdev->ih.gpu_addr >> 8);
2451         rb_bufsz = drm_order(rdev->ih.ring_size / 4);
2452
2453         ih_rb_cntl = (IH_WPTR_OVERFLOW_ENABLE |
2454                       IH_WPTR_OVERFLOW_CLEAR |
2455                       (rb_bufsz << 1));
2456         /* WPTR writeback, not yet */
2457         /*ih_rb_cntl |= IH_WPTR_WRITEBACK_ENABLE;*/
2458         WREG32(IH_RB_WPTR_ADDR_LO, 0);
2459         WREG32(IH_RB_WPTR_ADDR_HI, 0);
2460
2461         WREG32(IH_RB_CNTL, ih_rb_cntl);
2462
2463         /* set rptr, wptr to 0 */
2464         WREG32(IH_RB_RPTR, 0);
2465         WREG32(IH_RB_WPTR, 0);
2466
2467         /* Default settings for IH_CNTL (disabled at first) */
2468         ih_cntl = MC_WRREQ_CREDIT(0x10) | MC_WR_CLEAN_CNT(0x10);
2469         /* RPTR_REARM only works if msi's are enabled */
2470         if (rdev->msi_enabled)
2471                 ih_cntl |= RPTR_REARM;
2472
2473 #ifdef __BIG_ENDIAN
2474         ih_cntl |= IH_MC_SWAP(IH_MC_SWAP_32BIT);
2475 #endif
2476         WREG32(IH_CNTL, ih_cntl);
2477
2478         /* force the active interrupt state to all disabled */
2479         r600_disable_interrupt_state(rdev);
2480
2481         /* enable irqs */
2482         r600_enable_interrupts(rdev);
2483
2484         return ret;
2485 }
2486
2487 void r600_irq_suspend(struct radeon_device *rdev)
2488 {
2489         r600_disable_interrupts(rdev);
2490         r600_rlc_stop(rdev);
2491 }
2492
2493 void r600_irq_fini(struct radeon_device *rdev)
2494 {
2495         r600_irq_suspend(rdev);
2496         r600_ih_ring_fini(rdev);
2497 }
2498
2499 int r600_irq_set(struct radeon_device *rdev)
2500 {
2501         u32 cp_int_cntl = CNTX_BUSY_INT_ENABLE | CNTX_EMPTY_INT_ENABLE;
2502         u32 mode_int = 0;
2503         u32 hpd1, hpd2, hpd3, hpd4 = 0, hpd5 = 0, hpd6 = 0;
2504
2505         if (!rdev->irq.installed) {
2506                 WARN(1, "Can't enable IRQ/MSI because no handler is installed.\n");
2507                 return -EINVAL;
2508         }
2509         /* don't enable anything if the ih is disabled */
2510         if (!rdev->ih.enabled) {
2511                 r600_disable_interrupts(rdev);
2512                 /* force the active interrupt state to all disabled */
2513                 r600_disable_interrupt_state(rdev);
2514                 return 0;
2515         }
2516
2517         if (ASIC_IS_DCE3(rdev)) {
2518                 hpd1 = RREG32(DC_HPD1_INT_CONTROL) & ~DC_HPDx_INT_EN;
2519                 hpd2 = RREG32(DC_HPD2_INT_CONTROL) & ~DC_HPDx_INT_EN;
2520                 hpd3 = RREG32(DC_HPD3_INT_CONTROL) & ~DC_HPDx_INT_EN;
2521                 hpd4 = RREG32(DC_HPD4_INT_CONTROL) & ~DC_HPDx_INT_EN;
2522                 if (ASIC_IS_DCE32(rdev)) {
2523                         hpd5 = RREG32(DC_HPD5_INT_CONTROL) & ~DC_HPDx_INT_EN;
2524                         hpd6 = RREG32(DC_HPD6_INT_CONTROL) & ~DC_HPDx_INT_EN;
2525                 }
2526         } else {
2527                 hpd1 = RREG32(DC_HOT_PLUG_DETECT1_INT_CONTROL) & ~DC_HPDx_INT_EN;
2528                 hpd2 = RREG32(DC_HOT_PLUG_DETECT2_INT_CONTROL) & ~DC_HPDx_INT_EN;
2529                 hpd3 = RREG32(DC_HOT_PLUG_DETECT3_INT_CONTROL) & ~DC_HPDx_INT_EN;
2530         }
2531
2532         if (rdev->irq.sw_int) {
2533                 DRM_DEBUG("r600_irq_set: sw int\n");
2534                 cp_int_cntl |= RB_INT_ENABLE;
2535         }
2536         if (rdev->irq.crtc_vblank_int[0]) {
2537                 DRM_DEBUG("r600_irq_set: vblank 0\n");
2538                 mode_int |= D1MODE_VBLANK_INT_MASK;
2539         }
2540         if (rdev->irq.crtc_vblank_int[1]) {
2541                 DRM_DEBUG("r600_irq_set: vblank 1\n");
2542                 mode_int |= D2MODE_VBLANK_INT_MASK;
2543         }
2544         if (rdev->irq.hpd[0]) {
2545                 DRM_DEBUG("r600_irq_set: hpd 1\n");
2546                 hpd1 |= DC_HPDx_INT_EN;
2547         }
2548         if (rdev->irq.hpd[1]) {
2549                 DRM_DEBUG("r600_irq_set: hpd 2\n");
2550                 hpd2 |= DC_HPDx_INT_EN;
2551         }
2552         if (rdev->irq.hpd[2]) {
2553                 DRM_DEBUG("r600_irq_set: hpd 3\n");
2554                 hpd3 |= DC_HPDx_INT_EN;
2555         }
2556         if (rdev->irq.hpd[3]) {
2557                 DRM_DEBUG("r600_irq_set: hpd 4\n");
2558                 hpd4 |= DC_HPDx_INT_EN;
2559         }
2560         if (rdev->irq.hpd[4]) {
2561                 DRM_DEBUG("r600_irq_set: hpd 5\n");
2562                 hpd5 |= DC_HPDx_INT_EN;
2563         }
2564         if (rdev->irq.hpd[5]) {
2565                 DRM_DEBUG("r600_irq_set: hpd 6\n");
2566                 hpd6 |= DC_HPDx_INT_EN;
2567         }
2568
2569         WREG32(CP_INT_CNTL, cp_int_cntl);
2570         WREG32(DxMODE_INT_MASK, mode_int);
2571         if (ASIC_IS_DCE3(rdev)) {
2572                 WREG32(DC_HPD1_INT_CONTROL, hpd1);
2573                 WREG32(DC_HPD2_INT_CONTROL, hpd2);
2574                 WREG32(DC_HPD3_INT_CONTROL, hpd3);
2575                 WREG32(DC_HPD4_INT_CONTROL, hpd4);
2576                 if (ASIC_IS_DCE32(rdev)) {
2577                         WREG32(DC_HPD5_INT_CONTROL, hpd5);
2578                         WREG32(DC_HPD6_INT_CONTROL, hpd6);
2579                 }
2580         } else {
2581                 WREG32(DC_HOT_PLUG_DETECT1_INT_CONTROL, hpd1);
2582                 WREG32(DC_HOT_PLUG_DETECT2_INT_CONTROL, hpd2);
2583                 WREG32(DC_HOT_PLUG_DETECT3_INT_CONTROL, hpd3);
2584         }
2585
2586         return 0;
2587 }
2588
2589 static inline void r600_irq_ack(struct radeon_device *rdev,
2590                                 u32 *disp_int,
2591                                 u32 *disp_int_cont,
2592                                 u32 *disp_int_cont2)
2593 {
2594         u32 tmp;
2595
2596         if (ASIC_IS_DCE3(rdev)) {
2597                 *disp_int = RREG32(DCE3_DISP_INTERRUPT_STATUS);
2598                 *disp_int_cont = RREG32(DCE3_DISP_INTERRUPT_STATUS_CONTINUE);
2599                 *disp_int_cont2 = RREG32(DCE3_DISP_INTERRUPT_STATUS_CONTINUE2);
2600         } else {
2601                 *disp_int = RREG32(DISP_INTERRUPT_STATUS);
2602                 *disp_int_cont = RREG32(DISP_INTERRUPT_STATUS_CONTINUE);
2603                 *disp_int_cont2 = 0;
2604         }
2605
2606         if (*disp_int & LB_D1_VBLANK_INTERRUPT)
2607                 WREG32(D1MODE_VBLANK_STATUS, DxMODE_VBLANK_ACK);
2608         if (*disp_int & LB_D1_VLINE_INTERRUPT)
2609                 WREG32(D1MODE_VLINE_STATUS, DxMODE_VLINE_ACK);
2610         if (*disp_int & LB_D2_VBLANK_INTERRUPT)
2611                 WREG32(D2MODE_VBLANK_STATUS, DxMODE_VBLANK_ACK);
2612         if (*disp_int & LB_D2_VLINE_INTERRUPT)
2613                 WREG32(D2MODE_VLINE_STATUS, DxMODE_VLINE_ACK);
2614         if (*disp_int & DC_HPD1_INTERRUPT) {
2615                 if (ASIC_IS_DCE3(rdev)) {
2616                         tmp = RREG32(DC_HPD1_INT_CONTROL);
2617                         tmp |= DC_HPDx_INT_ACK;
2618                         WREG32(DC_HPD1_INT_CONTROL, tmp);
2619                 } else {
2620                         tmp = RREG32(DC_HOT_PLUG_DETECT1_INT_CONTROL);
2621                         tmp |= DC_HPDx_INT_ACK;
2622                         WREG32(DC_HOT_PLUG_DETECT1_INT_CONTROL, tmp);
2623                 }
2624         }
2625         if (*disp_int & DC_HPD2_INTERRUPT) {
2626                 if (ASIC_IS_DCE3(rdev)) {
2627                         tmp = RREG32(DC_HPD2_INT_CONTROL);
2628                         tmp |= DC_HPDx_INT_ACK;
2629                         WREG32(DC_HPD2_INT_CONTROL, tmp);
2630                 } else {
2631                         tmp = RREG32(DC_HOT_PLUG_DETECT2_INT_CONTROL);
2632                         tmp |= DC_HPDx_INT_ACK;
2633                         WREG32(DC_HOT_PLUG_DETECT2_INT_CONTROL, tmp);
2634                 }
2635         }
2636         if (*disp_int_cont & DC_HPD3_INTERRUPT) {
2637                 if (ASIC_IS_DCE3(rdev)) {
2638                         tmp = RREG32(DC_HPD3_INT_CONTROL);
2639                         tmp |= DC_HPDx_INT_ACK;
2640                         WREG32(DC_HPD3_INT_CONTROL, tmp);
2641                 } else {
2642                         tmp = RREG32(DC_HOT_PLUG_DETECT3_INT_CONTROL);
2643                         tmp |= DC_HPDx_INT_ACK;
2644                         WREG32(DC_HOT_PLUG_DETECT3_INT_CONTROL, tmp);
2645                 }
2646         }
2647         if (*disp_int_cont & DC_HPD4_INTERRUPT) {
2648                 tmp = RREG32(DC_HPD4_INT_CONTROL);
2649                 tmp |= DC_HPDx_INT_ACK;
2650                 WREG32(DC_HPD4_INT_CONTROL, tmp);
2651         }
2652         if (ASIC_IS_DCE32(rdev)) {
2653                 if (*disp_int_cont2 & DC_HPD5_INTERRUPT) {
2654                         tmp = RREG32(DC_HPD5_INT_CONTROL);
2655                         tmp |= DC_HPDx_INT_ACK;
2656                         WREG32(DC_HPD5_INT_CONTROL, tmp);
2657                 }
2658                 if (*disp_int_cont2 & DC_HPD6_INTERRUPT) {
2659                         tmp = RREG32(DC_HPD5_INT_CONTROL);
2660                         tmp |= DC_HPDx_INT_ACK;
2661                         WREG32(DC_HPD6_INT_CONTROL, tmp);
2662                 }
2663         }
2664 }
2665
2666 void r600_irq_disable(struct radeon_device *rdev)
2667 {
2668         u32 disp_int, disp_int_cont, disp_int_cont2;
2669
2670         r600_disable_interrupts(rdev);
2671         /* Wait and acknowledge irq */
2672         mdelay(1);
2673         r600_irq_ack(rdev, &disp_int, &disp_int_cont, &disp_int_cont2);
2674         r600_disable_interrupt_state(rdev);
2675 }
2676
2677 static inline u32 r600_get_ih_wptr(struct radeon_device *rdev)
2678 {
2679         u32 wptr, tmp;
2680
2681         /* XXX use writeback */
2682         wptr = RREG32(IH_RB_WPTR);
2683
2684         if (wptr & RB_OVERFLOW) {
2685                 /* When a ring buffer overflow happen start parsing interrupt
2686                  * from the last not overwritten vector (wptr + 16). Hopefully
2687                  * this should allow us to catchup.
2688                  */
2689                 dev_warn(rdev->dev, "IH ring buffer overflow (0x%08X, %d, %d)\n",
2690                         wptr, rdev->ih.rptr, (wptr + 16) + rdev->ih.ptr_mask);
2691                 rdev->ih.rptr = (wptr + 16) & rdev->ih.ptr_mask;
2692                 tmp = RREG32(IH_RB_CNTL);
2693                 tmp |= IH_WPTR_OVERFLOW_CLEAR;
2694                 WREG32(IH_RB_CNTL, tmp);
2695         }
2696         return (wptr & rdev->ih.ptr_mask);
2697 }
2698
2699 /*        r600 IV Ring
2700  * Each IV ring entry is 128 bits:
2701  * [7:0]    - interrupt source id
2702  * [31:8]   - reserved
2703  * [59:32]  - interrupt source data
2704  * [127:60]  - reserved
2705  *
2706  * The basic interrupt vector entries
2707  * are decoded as follows:
2708  * src_id  src_data  description
2709  *      1         0  D1 Vblank
2710  *      1         1  D1 Vline
2711  *      5         0  D2 Vblank
2712  *      5         1  D2 Vline
2713  *     19         0  FP Hot plug detection A
2714  *     19         1  FP Hot plug detection B
2715  *     19         2  DAC A auto-detection
2716  *     19         3  DAC B auto-detection
2717  *    176         -  CP_INT RB
2718  *    177         -  CP_INT IB1
2719  *    178         -  CP_INT IB2
2720  *    181         -  EOP Interrupt
2721  *    233         -  GUI Idle
2722  *
2723  * Note, these are based on r600 and may need to be
2724  * adjusted or added to on newer asics
2725  */
2726
2727 int r600_irq_process(struct radeon_device *rdev)
2728 {
2729         u32 wptr = r600_get_ih_wptr(rdev);
2730         u32 rptr = rdev->ih.rptr;
2731         u32 src_id, src_data;
2732         u32 ring_index, disp_int, disp_int_cont, disp_int_cont2;
2733         unsigned long flags;
2734         bool queue_hotplug = false;
2735
2736         DRM_DEBUG("r600_irq_process start: rptr %d, wptr %d\n", rptr, wptr);
2737         if (!rdev->ih.enabled)
2738                 return IRQ_NONE;
2739
2740         spin_lock_irqsave(&rdev->ih.lock, flags);
2741
2742         if (rptr == wptr) {
2743                 spin_unlock_irqrestore(&rdev->ih.lock, flags);
2744                 return IRQ_NONE;
2745         }
2746         if (rdev->shutdown) {
2747                 spin_unlock_irqrestore(&rdev->ih.lock, flags);
2748                 return IRQ_NONE;
2749         }
2750
2751 restart_ih:
2752         /* display interrupts */
2753         r600_irq_ack(rdev, &disp_int, &disp_int_cont, &disp_int_cont2);
2754
2755         rdev->ih.wptr = wptr;
2756         while (rptr != wptr) {
2757                 /* wptr/rptr are in bytes! */
2758                 ring_index = rptr / 4;
2759                 src_id =  rdev->ih.ring[ring_index] & 0xff;
2760                 src_data = rdev->ih.ring[ring_index + 1] & 0xfffffff;
2761
2762                 switch (src_id) {
2763                 case 1: /* D1 vblank/vline */
2764                         switch (src_data) {
2765                         case 0: /* D1 vblank */
2766                                 if (disp_int & LB_D1_VBLANK_INTERRUPT) {
2767                                         drm_handle_vblank(rdev->ddev, 0);
2768                                         rdev->pm.vblank_sync = true;
2769                                         wake_up(&rdev->irq.vblank_queue);
2770                                         disp_int &= ~LB_D1_VBLANK_INTERRUPT;
2771                                         DRM_DEBUG("IH: D1 vblank\n");
2772                                 }
2773                                 break;
2774                         case 1: /* D1 vline */
2775                                 if (disp_int & LB_D1_VLINE_INTERRUPT) {
2776                                         disp_int &= ~LB_D1_VLINE_INTERRUPT;
2777                                         DRM_DEBUG("IH: D1 vline\n");
2778                                 }
2779                                 break;
2780                         default:
2781                                 DRM_DEBUG("Unhandled interrupt: %d %d\n", src_id, src_data);
2782                                 break;
2783                         }
2784                         break;
2785                 case 5: /* D2 vblank/vline */
2786                         switch (src_data) {
2787                         case 0: /* D2 vblank */
2788                                 if (disp_int & LB_D2_VBLANK_INTERRUPT) {
2789                                         drm_handle_vblank(rdev->ddev, 1);
2790                                         rdev->pm.vblank_sync = true;
2791                                         wake_up(&rdev->irq.vblank_queue);
2792                                         disp_int &= ~LB_D2_VBLANK_INTERRUPT;
2793                                         DRM_DEBUG("IH: D2 vblank\n");
2794                                 }
2795                                 break;
2796                         case 1: /* D1 vline */
2797                                 if (disp_int & LB_D2_VLINE_INTERRUPT) {
2798                                         disp_int &= ~LB_D2_VLINE_INTERRUPT;
2799                                         DRM_DEBUG("IH: D2 vline\n");
2800                                 }
2801                                 break;
2802                         default:
2803                                 DRM_DEBUG("Unhandled interrupt: %d %d\n", src_id, src_data);
2804                                 break;
2805                         }
2806                         break;
2807                 case 19: /* HPD/DAC hotplug */
2808                         switch (src_data) {
2809                         case 0:
2810                                 if (disp_int & DC_HPD1_INTERRUPT) {
2811                                         disp_int &= ~DC_HPD1_INTERRUPT;
2812                                         queue_hotplug = true;
2813                                         DRM_DEBUG("IH: HPD1\n");
2814                                 }
2815                                 break;
2816                         case 1:
2817                                 if (disp_int & DC_HPD2_INTERRUPT) {
2818                                         disp_int &= ~DC_HPD2_INTERRUPT;
2819                                         queue_hotplug = true;
2820                                         DRM_DEBUG("IH: HPD2\n");
2821                                 }
2822                                 break;
2823                         case 4:
2824                                 if (disp_int_cont & DC_HPD3_INTERRUPT) {
2825                                         disp_int_cont &= ~DC_HPD3_INTERRUPT;
2826                                         queue_hotplug = true;
2827                                         DRM_DEBUG("IH: HPD3\n");
2828                                 }
2829                                 break;
2830                         case 5:
2831                                 if (disp_int_cont & DC_HPD4_INTERRUPT) {
2832                                         disp_int_cont &= ~DC_HPD4_INTERRUPT;
2833                                         queue_hotplug = true;
2834                                         DRM_DEBUG("IH: HPD4\n");
2835                                 }
2836                                 break;
2837                         case 10:
2838                                 if (disp_int_cont2 & DC_HPD5_INTERRUPT) {
2839                                         disp_int_cont &= ~DC_HPD5_INTERRUPT;
2840                                         queue_hotplug = true;
2841                                         DRM_DEBUG("IH: HPD5\n");
2842                                 }
2843                                 break;
2844                         case 12:
2845                                 if (disp_int_cont2 & DC_HPD6_INTERRUPT) {
2846                                         disp_int_cont &= ~DC_HPD6_INTERRUPT;
2847                                         queue_hotplug = true;
2848                                         DRM_DEBUG("IH: HPD6\n");
2849                                 }
2850                                 break;
2851                         default:
2852                                 DRM_DEBUG("Unhandled interrupt: %d %d\n", src_id, src_data);
2853                                 break;
2854                         }
2855                         break;
2856                 case 176: /* CP_INT in ring buffer */
2857                 case 177: /* CP_INT in IB1 */
2858                 case 178: /* CP_INT in IB2 */
2859                         DRM_DEBUG("IH: CP int: 0x%08x\n", src_data);
2860                         radeon_fence_process(rdev);
2861                         break;
2862                 case 181: /* CP EOP event */
2863                         DRM_DEBUG("IH: CP EOP\n");
2864                         break;
2865                 default:
2866                         DRM_DEBUG("Unhandled interrupt: %d %d\n", src_id, src_data);
2867                         break;
2868                 }
2869
2870                 /* wptr/rptr are in bytes! */
2871                 rptr += 16;
2872                 rptr &= rdev->ih.ptr_mask;
2873         }
2874         /* make sure wptr hasn't changed while processing */
2875         wptr = r600_get_ih_wptr(rdev);
2876         if (wptr != rdev->ih.wptr)
2877                 goto restart_ih;
2878         if (queue_hotplug)
2879                 queue_work(rdev->wq, &rdev->hotplug_work);
2880         rdev->ih.rptr = rptr;
2881         WREG32(IH_RB_RPTR, rdev->ih.rptr);
2882         spin_unlock_irqrestore(&rdev->ih.lock, flags);
2883         return IRQ_HANDLED;
2884 }
2885
2886 /*
2887  * Debugfs info
2888  */
2889 #if defined(CONFIG_DEBUG_FS)
2890
2891 static int r600_debugfs_cp_ring_info(struct seq_file *m, void *data)
2892 {
2893         struct drm_info_node *node = (struct drm_info_node *) m->private;
2894         struct drm_device *dev = node->minor->dev;
2895         struct radeon_device *rdev = dev->dev_private;
2896         unsigned count, i, j;
2897
2898         radeon_ring_free_size(rdev);
2899         count = (rdev->cp.ring_size / 4) - rdev->cp.ring_free_dw;
2900         seq_printf(m, "CP_STAT 0x%08x\n", RREG32(CP_STAT));
2901         seq_printf(m, "CP_RB_WPTR 0x%08x\n", RREG32(CP_RB_WPTR));
2902         seq_printf(m, "CP_RB_RPTR 0x%08x\n", RREG32(CP_RB_RPTR));
2903         seq_printf(m, "driver's copy of the CP_RB_WPTR 0x%08x\n", rdev->cp.wptr);
2904         seq_printf(m, "driver's copy of the CP_RB_RPTR 0x%08x\n", rdev->cp.rptr);
2905         seq_printf(m, "%u free dwords in ring\n", rdev->cp.ring_free_dw);
2906         seq_printf(m, "%u dwords in ring\n", count);
2907         i = rdev->cp.rptr;
2908         for (j = 0; j <= count; j++) {
2909                 seq_printf(m, "r[%04d]=0x%08x\n", i, rdev->cp.ring[i]);
2910                 i = (i + 1) & rdev->cp.ptr_mask;
2911         }
2912         return 0;
2913 }
2914
2915 static int r600_debugfs_mc_info(struct seq_file *m, void *data)
2916 {
2917         struct drm_info_node *node = (struct drm_info_node *) m->private;
2918         struct drm_device *dev = node->minor->dev;
2919         struct radeon_device *rdev = dev->dev_private;
2920
2921         DREG32_SYS(m, rdev, R_000E50_SRBM_STATUS);
2922         DREG32_SYS(m, rdev, VM_L2_STATUS);
2923         return 0;
2924 }
2925
2926 static struct drm_info_list r600_mc_info_list[] = {
2927         {"r600_mc_info", r600_debugfs_mc_info, 0, NULL},
2928         {"r600_ring_info", r600_debugfs_cp_ring_info, 0, NULL},
2929 };
2930 #endif
2931
2932 int r600_debugfs_mc_info_init(struct radeon_device *rdev)
2933 {
2934 #if defined(CONFIG_DEBUG_FS)
2935         return radeon_debugfs_add_files(rdev, r600_mc_info_list, ARRAY_SIZE(r600_mc_info_list));
2936 #else
2937         return 0;
2938 #endif
2939 }
2940
2941 /**
2942  * r600_ioctl_wait_idle - flush host path cache on wait idle ioctl
2943  * rdev: radeon device structure
2944  * bo: buffer object struct which userspace is waiting for idle
2945  *
2946  * Some R6XX/R7XX doesn't seems to take into account HDP flush performed
2947  * through ring buffer, this leads to corruption in rendering, see
2948  * http://bugzilla.kernel.org/show_bug.cgi?id=15186 to avoid this we
2949  * directly perform HDP flush by writing register through MMIO.
2950  */
2951 void r600_ioctl_wait_idle(struct radeon_device *rdev, struct radeon_bo *bo)
2952 {
2953         WREG32(R_005480_HDP_MEM_COHERENCY_FLUSH_CNTL, 0x1);
2954 }