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 int r600_gart_clear_page(struct radeon_device *rdev, int i)
 357 {
 358         void __iomem *ptr = (void *)rdev->gart.table.vram.ptr;
 359         u64 pte;
 360
 361         if (i < 0 || i > rdev->gart.num_gpu_pages)
 362                 return -EINVAL;
 363         pte = 0;
 364         writeq(pte, ((void __iomem *)ptr) + (i * 8));
 365         return 0;
 366 }
 367
 368 void r600_pcie_gart_tlb_flush(struct radeon_device *rdev)
 369 {
 370         unsigned i;
 371         u32 tmp;
 372
 373         WREG32(VM_CONTEXT0_INVALIDATION_LOW_ADDR, rdev->mc.gtt_start >> 12);
 374         WREG32(VM_CONTEXT0_INVALIDATION_HIGH_ADDR, (rdev->mc.gtt_end - 1) >> 12);
 375         WREG32(VM_CONTEXT0_REQUEST_RESPONSE, REQUEST_TYPE(1));
 376         for (i = 0; i < rdev->usec_timeout; i++) {
 377                 /* read MC_STATUS */
 378                 tmp = RREG32(VM_CONTEXT0_REQUEST_RESPONSE);
 379                 tmp = (tmp & RESPONSE_TYPE_MASK) >> RESPONSE_TYPE_SHIFT;
 380                 if (tmp == 2) {
 381                         printk(KERN_WARNING "[drm] r600 flush TLB failed\n");
 382                         return;
 383                 }
 384                 if (tmp) {
 385                         return;
 386                 }
 387                 udelay(1);
 388         }
 389 }
 390
 391 int r600_pcie_gart_init(struct radeon_device *rdev)
 392 {
 393         int r;
 394
 395         if (rdev->gart.table.vram.robj) {
 396                 WARN(1, "R600 PCIE GART already initialized.\n");
 397                 return 0;
 398         }
 399         /* Initialize common gart structure */
 400         r = radeon_gart_init(rdev);
 401         if (r)
 402                 return r;
 403         rdev->gart.table_size = rdev->gart.num_gpu_pages * 8;
 404         return radeon_gart_table_vram_alloc(rdev);
 405 }
 406
 407 int r600_pcie_gart_enable(struct radeon_device *rdev)
 408 {
 409         u32 tmp;
 410         int r, i;
 411
 412         if (rdev->gart.table.vram.robj == NULL) {
 413                 dev_err(rdev->dev, "No VRAM object for PCIE GART.\n");
 414                 return -EINVAL;
 415         }
 416         r = radeon_gart_table_vram_pin(rdev);
 417         if (r)
 418                 return r;
 419
 420         /* Setup L2 cache */
 421         WREG32(VM_L2_CNTL, ENABLE_L2_CACHE | ENABLE_L2_FRAGMENT_PROCESSING |
 422                                 ENABLE_L2_PTE_CACHE_LRU_UPDATE_BY_WRITE |
 423                                 EFFECTIVE_L2_QUEUE_SIZE(7));
 424         WREG32(VM_L2_CNTL2, 0);
 425         WREG32(VM_L2_CNTL3, BANK_SELECT_0(0) | BANK_SELECT_1(1));
 426         /* Setup TLB control */
 427         tmp = ENABLE_L1_TLB | ENABLE_L1_FRAGMENT_PROCESSING |
 428                 SYSTEM_ACCESS_MODE_NOT_IN_SYS |
 429                 EFFECTIVE_L1_TLB_SIZE(5) | EFFECTIVE_L1_QUEUE_SIZE(5) |
 430                 ENABLE_WAIT_L2_QUERY;
 431         WREG32(MC_VM_L1_TLB_MCB_RD_SYS_CNTL, tmp);
 432         WREG32(MC_VM_L1_TLB_MCB_WR_SYS_CNTL, tmp);
 433         WREG32(MC_VM_L1_TLB_MCB_RD_HDP_CNTL, tmp | ENABLE_L1_STRICT_ORDERING);
 434         WREG32(MC_VM_L1_TLB_MCB_WR_HDP_CNTL, tmp);
 435         WREG32(MC_VM_L1_TLB_MCD_RD_A_CNTL, tmp);
 436         WREG32(MC_VM_L1_TLB_MCD_WR_A_CNTL, tmp);
 437         WREG32(MC_VM_L1_TLB_MCD_RD_B_CNTL, tmp);
 438         WREG32(MC_VM_L1_TLB_MCD_WR_B_CNTL, tmp);
 439         WREG32(MC_VM_L1_TLB_MCB_RD_GFX_CNTL, tmp);
 440         WREG32(MC_VM_L1_TLB_MCB_WR_GFX_CNTL, tmp);
 441         WREG32(MC_VM_L1_TLB_MCB_RD_PDMA_CNTL, tmp);
 442         WREG32(MC_VM_L1_TLB_MCB_WR_PDMA_CNTL, tmp);
 443         WREG32(MC_VM_L1_TLB_MCB_RD_SEM_CNTL, tmp | ENABLE_SEMAPHORE_MODE);
 444         WREG32(MC_VM_L1_TLB_MCB_WR_SEM_CNTL, tmp | ENABLE_SEMAPHORE_MODE);
 445         WREG32(VM_CONTEXT0_PAGE_TABLE_START_ADDR, rdev->mc.gtt_start >> 12);
 446         WREG32(VM_CONTEXT0_PAGE_TABLE_END_ADDR, rdev->mc.gtt_end >> 12);
 447         WREG32(VM_CONTEXT0_PAGE_TABLE_BASE_ADDR, rdev->gart.table_addr >> 12);
 448         WREG32(VM_CONTEXT0_CNTL, ENABLE_CONTEXT | PAGE_TABLE_DEPTH(0) |
 449                                 RANGE_PROTECTION_FAULT_ENABLE_DEFAULT);
 450         WREG32(VM_CONTEXT0_PROTECTION_FAULT_DEFAULT_ADDR,
 451                         (u32)(rdev->dummy_page.addr >> 12));
 452         for (i = 1; i < 7; i++)
 453                 WREG32(VM_CONTEXT0_CNTL + (i * 4), 0);
 454
 455         r600_pcie_gart_tlb_flush(rdev);
 456         rdev->gart.ready = true;
 457         return 0;
 458 }
 459
 460 void r600_pcie_gart_disable(struct radeon_device *rdev)
 461 {
 462         u32 tmp;
 463         int i, r;
 464
 465         /* Disable all tables */
 466         for (i = 0; i < 7; i++)
 467                 WREG32(VM_CONTEXT0_CNTL + (i * 4), 0);
 468
 469         /* Disable L2 cache */
 470         WREG32(VM_L2_CNTL, ENABLE_L2_FRAGMENT_PROCESSING |
 471                                 EFFECTIVE_L2_QUEUE_SIZE(7));
 472         WREG32(VM_L2_CNTL3, BANK_SELECT_0(0) | BANK_SELECT_1(1));
 473         /* Setup L1 TLB control */
 474         tmp = EFFECTIVE_L1_TLB_SIZE(5) | EFFECTIVE_L1_QUEUE_SIZE(5) |
 475                 ENABLE_WAIT_L2_QUERY;
 476         WREG32(MC_VM_L1_TLB_MCD_RD_A_CNTL, tmp);
 477         WREG32(MC_VM_L1_TLB_MCD_WR_A_CNTL, tmp);
 478         WREG32(MC_VM_L1_TLB_MCD_RD_B_CNTL, tmp);
 479         WREG32(MC_VM_L1_TLB_MCD_WR_B_CNTL, tmp);
 480         WREG32(MC_VM_L1_TLB_MCB_RD_GFX_CNTL, tmp);
 481         WREG32(MC_VM_L1_TLB_MCB_WR_GFX_CNTL, tmp);
 482         WREG32(MC_VM_L1_TLB_MCB_RD_PDMA_CNTL, tmp);
 483         WREG32(MC_VM_L1_TLB_MCB_WR_PDMA_CNTL, tmp);
 484         WREG32(MC_VM_L1_TLB_MCB_RD_SEM_CNTL, tmp);
 485         WREG32(MC_VM_L1_TLB_MCB_WR_SEM_CNTL, tmp);
 486         WREG32(MC_VM_L1_TLB_MCB_RD_SYS_CNTL, tmp);
 487         WREG32(MC_VM_L1_TLB_MCB_WR_SYS_CNTL, tmp);
 488         WREG32(MC_VM_L1_TLB_MCB_RD_HDP_CNTL, tmp);
 489         WREG32(MC_VM_L1_TLB_MCB_WR_HDP_CNTL, tmp);
 490         if (rdev->gart.table.vram.robj) {
 491                 r = radeon_bo_reserve(rdev->gart.table.vram.robj, false);
 492                 if (likely(r == 0)) {
 493                         radeon_bo_kunmap(rdev->gart.table.vram.robj);
 494                         radeon_bo_unpin(rdev->gart.table.vram.robj);
 495                         radeon_bo_unreserve(rdev->gart.table.vram.robj);
 496                 }
 497         }
 498 }
 499
 500 void r600_pcie_gart_fini(struct radeon_device *rdev)
 501 {
 502         r600_pcie_gart_disable(rdev);
 503         radeon_gart_table_vram_free(rdev);
 504         radeon_gart_fini(rdev);
 505 }
 506
 507 void r600_agp_enable(struct radeon_device *rdev)
 508 {
 509         u32 tmp;
 510         int i;
 511
 512         /* Setup L2 cache */
 513         WREG32(VM_L2_CNTL, ENABLE_L2_CACHE | ENABLE_L2_FRAGMENT_PROCESSING |
 514                                 ENABLE_L2_PTE_CACHE_LRU_UPDATE_BY_WRITE |
 515                                 EFFECTIVE_L2_QUEUE_SIZE(7));
 516         WREG32(VM_L2_CNTL2, 0);
 517         WREG32(VM_L2_CNTL3, BANK_SELECT_0(0) | BANK_SELECT_1(1));
 518         /* Setup TLB control */
 519         tmp = ENABLE_L1_TLB | ENABLE_L1_FRAGMENT_PROCESSING |
 520                 SYSTEM_ACCESS_MODE_NOT_IN_SYS |
 521                 EFFECTIVE_L1_TLB_SIZE(5) | EFFECTIVE_L1_QUEUE_SIZE(5) |
 522                 ENABLE_WAIT_L2_QUERY;
 523         WREG32(MC_VM_L1_TLB_MCB_RD_SYS_CNTL, tmp);
 524         WREG32(MC_VM_L1_TLB_MCB_WR_SYS_CNTL, tmp);
 525         WREG32(MC_VM_L1_TLB_MCB_RD_HDP_CNTL, tmp | ENABLE_L1_STRICT_ORDERING);
 526         WREG32(MC_VM_L1_TLB_MCB_WR_HDP_CNTL, tmp);
 527         WREG32(MC_VM_L1_TLB_MCD_RD_A_CNTL, tmp);
 528         WREG32(MC_VM_L1_TLB_MCD_WR_A_CNTL, tmp);
 529         WREG32(MC_VM_L1_TLB_MCD_RD_B_CNTL, tmp);
 530         WREG32(MC_VM_L1_TLB_MCD_WR_B_CNTL, tmp);
 531         WREG32(MC_VM_L1_TLB_MCB_RD_GFX_CNTL, tmp);
 532         WREG32(MC_VM_L1_TLB_MCB_WR_GFX_CNTL, tmp);
 533         WREG32(MC_VM_L1_TLB_MCB_RD_PDMA_CNTL, tmp);
 534         WREG32(MC_VM_L1_TLB_MCB_WR_PDMA_CNTL, tmp);
 535         WREG32(MC_VM_L1_TLB_MCB_RD_SEM_CNTL, tmp | ENABLE_SEMAPHORE_MODE);
 536         WREG32(MC_VM_L1_TLB_MCB_WR_SEM_CNTL, tmp | ENABLE_SEMAPHORE_MODE);
 537         for (i = 0; i < 7; i++)
 538                 WREG32(VM_CONTEXT0_CNTL + (i * 4), 0);
 539 }
 540
 541 int r600_mc_wait_for_idle(struct radeon_device *rdev)
 542 {
 543         unsigned i;
 544         u32 tmp;
 545
 546         for (i = 0; i < rdev->usec_timeout; i++) {
 547                 /* read MC_STATUS */
 548                 tmp = RREG32(R_000E50_SRBM_STATUS) & 0x3F00;
 549                 if (!tmp)
 550                         return 0;
 551                 udelay(1);
 552         }
 553         return -1;
 554 }
 555
 556 static void r600_mc_program(struct radeon_device *rdev)
 557 {
 558         struct rv515_mc_save save;
 559         u32 tmp;
 560         int i, j;
 561
 562         /* Initialize HDP */
 563         for (i = 0, j = 0; i < 32; i++, j += 0x18) {
 564                 WREG32((0x2c14 + j), 0x00000000);
 565                 WREG32((0x2c18 + j), 0x00000000);
 566                 WREG32((0x2c1c + j), 0x00000000);
 567                 WREG32((0x2c20 + j), 0x00000000);
 568                 WREG32((0x2c24 + j), 0x00000000);
 569         }
 570         WREG32(HDP_REG_COHERENCY_FLUSH_CNTL, 0);
 571
 572         rv515_mc_stop(rdev, &save);
 573         if (r600_mc_wait_for_idle(rdev)) {
 574                 dev_warn(rdev->dev, "Wait for MC idle timedout !\n");
 575         }
 576         /* Lockout access through VGA aperture (doesn't exist before R600) */
 577         WREG32(VGA_HDP_CONTROL, VGA_MEMORY_DISABLE);
 578         /* Update configuration */
 579         if (rdev->flags & RADEON_IS_AGP) {
 580                 if (rdev->mc.vram_start < rdev->mc.gtt_start) {
 581                         /* VRAM before AGP */
 582                         WREG32(MC_VM_SYSTEM_APERTURE_LOW_ADDR,
 583                                 rdev->mc.vram_start >> 12);
 584                         WREG32(MC_VM_SYSTEM_APERTURE_HIGH_ADDR,
 585                                 rdev->mc.gtt_end >> 12);
 586                 } else {
 587                         /* VRAM after AGP */
 588                         WREG32(MC_VM_SYSTEM_APERTURE_LOW_ADDR,
 589                                 rdev->mc.gtt_start >> 12);
 590                         WREG32(MC_VM_SYSTEM_APERTURE_HIGH_ADDR,
 591                                 rdev->mc.vram_end >> 12);
 592                 }
 593         } else {
 594                 WREG32(MC_VM_SYSTEM_APERTURE_LOW_ADDR, rdev->mc.vram_start >> 12);
 595                 WREG32(MC_VM_SYSTEM_APERTURE_HIGH_ADDR, rdev->mc.vram_end >> 12);
 596         }
 597         WREG32(MC_VM_SYSTEM_APERTURE_DEFAULT_ADDR, 0);
 598         tmp = ((rdev->mc.vram_end >> 24) & 0xFFFF) << 16;
 599         tmp |= ((rdev->mc.vram_start >> 24) & 0xFFFF);
 600         WREG32(MC_VM_FB_LOCATION, tmp);
 601         WREG32(HDP_NONSURFACE_BASE, (rdev->mc.vram_start >> 8));
 602         WREG32(HDP_NONSURFACE_INFO, (2 << 7));
 603         WREG32(HDP_NONSURFACE_SIZE, rdev->mc.mc_vram_size | 0x3FF);
 604         if (rdev->flags & RADEON_IS_AGP) {
 605                 WREG32(MC_VM_AGP_TOP, rdev->mc.gtt_end >> 22);
 606                 WREG32(MC_VM_AGP_BOT, rdev->mc.gtt_start >> 22);
 607                 WREG32(MC_VM_AGP_BASE, rdev->mc.agp_base >> 22);
 608         } else {
 609                 WREG32(MC_VM_AGP_BASE, 0);
 610                 WREG32(MC_VM_AGP_TOP, 0x0FFFFFFF);
 611                 WREG32(MC_VM_AGP_BOT, 0x0FFFFFFF);
 612         }
 613         if (r600_mc_wait_for_idle(rdev)) {
 614                 dev_warn(rdev->dev, "Wait for MC idle timedout !\n");
 615         }
 616         rv515_mc_resume(rdev, &save);
 617         /* we need to own VRAM, so turn off the VGA renderer here
 618          * to stop it overwriting our objects */
 619         rv515_vga_render_disable(rdev);
 620 }
 621
 622 int r600_mc_init(struct radeon_device *rdev)
 623 {
 624         fixed20_12 a;
 625         u32 tmp;
 626         int chansize, numchan;
 627
 628         /* Get VRAM informations */
 629         rdev->mc.vram_is_ddr = true;
 630         tmp = RREG32(RAMCFG);
 631         if (tmp & CHANSIZE_OVERRIDE) {
 632                 chansize = 16;
 633         } else if (tmp & CHANSIZE_MASK) {
 634                 chansize = 64;
 635         } else {
 636                 chansize = 32;
 637         }
 638         tmp = RREG32(CHMAP);
 639         switch ((tmp & NOOFCHAN_MASK) >> NOOFCHAN_SHIFT) {
 640         case 0:
 641         default:
 642                 numchan = 1;
 643                 break;
 644         case 1:
 645                 numchan = 2;
 646                 break;
 647         case 2:
 648                 numchan = 4;
 649                 break;
 650         case 3:
 651                 numchan = 8;
 652                 break;
 653         }
 654         rdev->mc.vram_width = numchan * chansize;
 655         /* Could aper size report 0 ? */
 656         rdev->mc.aper_base = drm_get_resource_start(rdev->ddev, 0);
 657         rdev->mc.aper_size = drm_get_resource_len(rdev->ddev, 0);
 658         /* Setup GPU memory space */
 659         rdev->mc.mc_vram_size = RREG32(CONFIG_MEMSIZE);
 660         rdev->mc.real_vram_size = RREG32(CONFIG_MEMSIZE);
 661
 662         if (rdev->mc.mc_vram_size > rdev->mc.aper_size)
 663                 rdev->mc.mc_vram_size = rdev->mc.aper_size;
 664
 665         if (rdev->mc.real_vram_size > rdev->mc.aper_size)
 666                 rdev->mc.real_vram_size = rdev->mc.aper_size;
 667
 668         if (rdev->flags & RADEON_IS_AGP) {
 669                 /* gtt_size is setup by radeon_agp_init */
 670                 rdev->mc.gtt_location = rdev->mc.agp_base;
 671                 tmp = 0xFFFFFFFFUL - rdev->mc.agp_base - rdev->mc.gtt_size;
 672                 /* Try to put vram before or after AGP because we
 673                  * we want SYSTEM_APERTURE to cover both VRAM and
 674                  * AGP so that GPU can catch out of VRAM/AGP access
 675                  */
 676                 if (rdev->mc.gtt_location > rdev->mc.mc_vram_size) {
 677                         /* Enough place before */
 678                         rdev->mc.vram_location = rdev->mc.gtt_location -
 679                                                         rdev->mc.mc_vram_size;
 680                 } else if (tmp > rdev->mc.mc_vram_size) {
 681                         /* Enough place after */
 682                         rdev->mc.vram_location = rdev->mc.gtt_location +
 683                                                         rdev->mc.gtt_size;
 684                 } else {
 685                         /* Try to setup VRAM then AGP might not
 686                          * not work on some card
 687                          */
 688                         rdev->mc.vram_location = 0x00000000UL;
 689                         rdev->mc.gtt_location = rdev->mc.mc_vram_size;
 690                 }
 691         } else {
 692                 rdev->mc.gtt_size = radeon_gart_size * 1024 * 1024;
 693                 rdev->mc.vram_location = (RREG32(MC_VM_FB_LOCATION) &
 694                                                         0xFFFF) << 24;
 695                 tmp = rdev->mc.vram_location + rdev->mc.mc_vram_size;
 696                 if ((0xFFFFFFFFUL - tmp) >= rdev->mc.gtt_size) {
 697                         /* Enough place after vram */
 698                         rdev->mc.gtt_location = tmp;
 699                 } else if (rdev->mc.vram_location >= rdev->mc.gtt_size) {
 700                         /* Enough place before vram */
 701                         rdev->mc.gtt_location = 0;
 702                 } else {
 703                         /* Not enough place after or before shrink
 704                          * gart size
 705                          */
 706                         if (rdev->mc.vram_location > (0xFFFFFFFFUL - tmp)) {
 707                                 rdev->mc.gtt_location = 0;
 708                                 rdev->mc.gtt_size = rdev->mc.vram_location;
 709                         } else {
 710                                 rdev->mc.gtt_location = tmp;
 711                                 rdev->mc.gtt_size = 0xFFFFFFFFUL - tmp;
 712                         }
 713                 }
 714                 rdev->mc.gtt_location = rdev->mc.mc_vram_size;
 715         }
 716         rdev->mc.vram_start = rdev->mc.vram_location;
 717         rdev->mc.vram_end = rdev->mc.vram_location + rdev->mc.mc_vram_size - 1;
 718         rdev->mc.gtt_start = rdev->mc.gtt_location;
 719         rdev->mc.gtt_end = rdev->mc.gtt_location + rdev->mc.gtt_size - 1;
 720         /* FIXME: we should enforce default clock in case GPU is not in
 721          * default setup
 722          */
 723         a.full = rfixed_const(100);
 724         rdev->pm.sclk.full = rfixed_const(rdev->clock.default_sclk);
 725         rdev->pm.sclk.full = rfixed_div(rdev->pm.sclk, a);
 726
 727         if (rdev->flags & RADEON_IS_IGP)
 728                 rdev->mc.igp_sideport_enabled = radeon_atombios_sideport_present(rdev);
 729
 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 tmp;
 985         int i, j;
 986         u32 sq_config;
 987         u32 sq_gpr_resource_mgmt_1 = 0;
 988         u32 sq_gpr_resource_mgmt_2 = 0;
 989         u32 sq_thread_resource_mgmt = 0;
 990         u32 sq_stack_resource_mgmt_1 = 0;
 991         u32 sq_stack_resource_mgmt_2 = 0;
 992
 993         /* FIXME: implement */
 994         switch (rdev->family) {
 995         case CHIP_R600:
 996                 rdev->config.r600.max_pipes = 4;
 997                 rdev->config.r600.max_tile_pipes = 8;
 998                 rdev->config.r600.max_simds = 4;
 999                 rdev->config.r600.max_backends = 4;
1000                 rdev->config.r600.max_gprs = 256;
1001                 rdev->config.r600.max_threads = 192;
1002                 rdev->config.r600.max_stack_entries = 256;
1003                 rdev->config.r600.max_hw_contexts = 8;
1004                 rdev->config.r600.max_gs_threads = 16;
1005                 rdev->config.r600.sx_max_export_size = 128;
1006                 rdev->config.r600.sx_max_export_pos_size = 16;
1007                 rdev->config.r600.sx_max_export_smx_size = 128;
1008                 rdev->config.r600.sq_num_cf_insts = 2;
1009                 break;
1010         case CHIP_RV630:
1011         case CHIP_RV635:
1012                 rdev->config.r600.max_pipes = 2;
1013                 rdev->config.r600.max_tile_pipes = 2;
1014                 rdev->config.r600.max_simds = 3;
1015                 rdev->config.r600.max_backends = 1;
1016                 rdev->config.r600.max_gprs = 128;
1017                 rdev->config.r600.max_threads = 192;
1018                 rdev->config.r600.max_stack_entries = 128;
1019                 rdev->config.r600.max_hw_contexts = 8;
1020                 rdev->config.r600.max_gs_threads = 4;
1021                 rdev->config.r600.sx_max_export_size = 128;
1022                 rdev->config.r600.sx_max_export_pos_size = 16;
1023                 rdev->config.r600.sx_max_export_smx_size = 128;
1024                 rdev->config.r600.sq_num_cf_insts = 2;
1025                 break;
1026         case CHIP_RV610:
1027         case CHIP_RV620:
1028         case CHIP_RS780:
1029         case CHIP_RS880:
1030                 rdev->config.r600.max_pipes = 1;
1031                 rdev->config.r600.max_tile_pipes = 1;
1032                 rdev->config.r600.max_simds = 2;
1033                 rdev->config.r600.max_backends = 1;
1034                 rdev->config.r600.max_gprs = 128;
1035                 rdev->config.r600.max_threads = 192;
1036                 rdev->config.r600.max_stack_entries = 128;
1037                 rdev->config.r600.max_hw_contexts = 4;
1038                 rdev->config.r600.max_gs_threads = 4;
1039                 rdev->config.r600.sx_max_export_size = 128;
1040                 rdev->config.r600.sx_max_export_pos_size = 16;
1041                 rdev->config.r600.sx_max_export_smx_size = 128;
1042                 rdev->config.r600.sq_num_cf_insts = 1;
1043                 break;
1044         case CHIP_RV670:
1045                 rdev->config.r600.max_pipes = 4;
1046                 rdev->config.r600.max_tile_pipes = 4;
1047                 rdev->config.r600.max_simds = 4;
1048                 rdev->config.r600.max_backends = 4;
1049                 rdev->config.r600.max_gprs = 192;
1050                 rdev->config.r600.max_threads = 192;
1051                 rdev->config.r600.max_stack_entries = 256;
1052                 rdev->config.r600.max_hw_contexts = 8;
1053                 rdev->config.r600.max_gs_threads = 16;
1054                 rdev->config.r600.sx_max_export_size = 128;
1055                 rdev->config.r600.sx_max_export_pos_size = 16;
1056                 rdev->config.r600.sx_max_export_smx_size = 128;
1057                 rdev->config.r600.sq_num_cf_insts = 2;
1058                 break;
1059         default:
1060                 break;
1061         }
1062
1063         /* Initialize HDP */
1064         for (i = 0, j = 0; i < 32; i++, j += 0x18) {
1065                 WREG32((0x2c14 + j), 0x00000000);
1066                 WREG32((0x2c18 + j), 0x00000000);
1067                 WREG32((0x2c1c + j), 0x00000000);
1068                 WREG32((0x2c20 + j), 0x00000000);
1069                 WREG32((0x2c24 + j), 0x00000000);
1070         }
1071
1072         WREG32(GRBM_CNTL, GRBM_READ_TIMEOUT(0xff));
1073
1074         /* Setup tiling */
1075         tiling_config = 0;
1076         ramcfg = RREG32(RAMCFG);
1077         switch (rdev->config.r600.max_tile_pipes) {
1078         case 1:
1079                 tiling_config |= PIPE_TILING(0);
1080                 break;
1081         case 2:
1082                 tiling_config |= PIPE_TILING(1);
1083                 break;
1084         case 4:
1085                 tiling_config |= PIPE_TILING(2);
1086                 break;
1087         case 8:
1088                 tiling_config |= PIPE_TILING(3);
1089                 break;
1090         default:
1091                 break;
1092         }
1093         tiling_config |= BANK_TILING((ramcfg & NOOFBANK_MASK) >> NOOFBANK_SHIFT);
1094         tiling_config |= GROUP_SIZE(0);
1095         tmp = (ramcfg & NOOFROWS_MASK) >> NOOFROWS_SHIFT;
1096         if (tmp > 3) {
1097                 tiling_config |= ROW_TILING(3);
1098                 tiling_config |= SAMPLE_SPLIT(3);
1099         } else {
1100                 tiling_config |= ROW_TILING(tmp);
1101                 tiling_config |= SAMPLE_SPLIT(tmp);
1102         }
1103         tiling_config |= BANK_SWAPS(1);
1104         tmp = r600_get_tile_pipe_to_backend_map(rdev->config.r600.max_tile_pipes,
1105                                                 rdev->config.r600.max_backends,
1106                                                 (0xff << rdev->config.r600.max_backends) & 0xff);
1107         tiling_config |= BACKEND_MAP(tmp);
1108         WREG32(GB_TILING_CONFIG, tiling_config);
1109         WREG32(DCP_TILING_CONFIG, tiling_config & 0xffff);
1110         WREG32(HDP_TILING_CONFIG, tiling_config & 0xffff);
1111
1112         tmp = BACKEND_DISABLE((R6XX_MAX_BACKENDS_MASK << rdev->config.r600.max_backends) & R6XX_MAX_BACKENDS_MASK);
1113         WREG32(CC_RB_BACKEND_DISABLE, tmp);
1114
1115         /* Setup pipes */
1116         tmp = INACTIVE_QD_PIPES((R6XX_MAX_PIPES_MASK << rdev->config.r600.max_pipes) & R6XX_MAX_PIPES_MASK);
1117         tmp |= INACTIVE_SIMDS((R6XX_MAX_SIMDS_MASK << rdev->config.r600.max_simds) & R6XX_MAX_SIMDS_MASK);
1118         WREG32(CC_GC_SHADER_PIPE_CONFIG, tmp);
1119         WREG32(GC_USER_SHADER_PIPE_CONFIG, tmp);
1120
1121         tmp = R6XX_MAX_BACKENDS - r600_count_pipe_bits(tmp & INACTIVE_QD_PIPES_MASK);
1122         WREG32(VGT_OUT_DEALLOC_CNTL, (tmp * 4) & DEALLOC_DIST_MASK);
1123         WREG32(VGT_VERTEX_REUSE_BLOCK_CNTL, ((tmp * 4) - 2) & VTX_REUSE_DEPTH_MASK);
1124
1125         /* Setup some CP states */
1126         WREG32(CP_QUEUE_THRESHOLDS, (ROQ_IB1_START(0x16) | ROQ_IB2_START(0x2b)));
1127         WREG32(CP_MEQ_THRESHOLDS, (MEQ_END(0x40) | ROQ_END(0x40)));
1128
1129         WREG32(TA_CNTL_AUX, (DISABLE_CUBE_ANISO | SYNC_GRADIENT |
1130                              SYNC_WALKER | SYNC_ALIGNER));
1131         /* Setup various GPU states */
1132         if (rdev->family == CHIP_RV670)
1133                 WREG32(ARB_GDEC_RD_CNTL, 0x00000021);
1134
1135         tmp = RREG32(SX_DEBUG_1);
1136         tmp |= SMX_EVENT_RELEASE;
1137         if ((rdev->family > CHIP_R600))
1138                 tmp |= ENABLE_NEW_SMX_ADDRESS;
1139         WREG32(SX_DEBUG_1, tmp);
1140
1141         if (((rdev->family) == CHIP_R600) ||
1142             ((rdev->family) == CHIP_RV630) ||
1143             ((rdev->family) == CHIP_RV610) ||
1144             ((rdev->family) == CHIP_RV620) ||
1145             ((rdev->family) == CHIP_RS780) ||
1146             ((rdev->family) == CHIP_RS880)) {
1147                 WREG32(DB_DEBUG, PREZ_MUST_WAIT_FOR_POSTZ_DONE);
1148         } else {
1149                 WREG32(DB_DEBUG, 0);
1150         }
1151         WREG32(DB_WATERMARKS, (DEPTH_FREE(4) | DEPTH_CACHELINE_FREE(16) |
1152                                DEPTH_FLUSH(16) | DEPTH_PENDING_FREE(4)));
1153
1154         WREG32(PA_SC_MULTI_CHIP_CNTL, 0);
1155         WREG32(VGT_NUM_INSTANCES, 0);
1156
1157         WREG32(SPI_CONFIG_CNTL, GPR_WRITE_PRIORITY(0));
1158         WREG32(SPI_CONFIG_CNTL_1, VTX_DONE_DELAY(0));
1159
1160         tmp = RREG32(SQ_MS_FIFO_SIZES);
1161         if (((rdev->family) == CHIP_RV610) ||
1162             ((rdev->family) == CHIP_RV620) ||
1163             ((rdev->family) == CHIP_RS780) ||
1164             ((rdev->family) == CHIP_RS880)) {
1165                 tmp = (CACHE_FIFO_SIZE(0xa) |
1166                        FETCH_FIFO_HIWATER(0xa) |
1167                        DONE_FIFO_HIWATER(0xe0) |
1168                        ALU_UPDATE_FIFO_HIWATER(0x8));
1169         } else if (((rdev->family) == CHIP_R600) ||
1170                    ((rdev->family) == CHIP_RV630)) {
1171                 tmp &= ~DONE_FIFO_HIWATER(0xff);
1172                 tmp |= DONE_FIFO_HIWATER(0x4);
1173         }
1174         WREG32(SQ_MS_FIFO_SIZES, tmp);
1175
1176         /* SQ_CONFIG, SQ_GPR_RESOURCE_MGMT, SQ_THREAD_RESOURCE_MGMT, SQ_STACK_RESOURCE_MGMT
1177          * should be adjusted as needed by the 2D/3D drivers.  This just sets default values
1178          */
1179         sq_config = RREG32(SQ_CONFIG);
1180         sq_config &= ~(PS_PRIO(3) |
1181                        VS_PRIO(3) |
1182                        GS_PRIO(3) |
1183                        ES_PRIO(3));
1184         sq_config |= (DX9_CONSTS |
1185                       VC_ENABLE |
1186                       PS_PRIO(0) |
1187                       VS_PRIO(1) |
1188                       GS_PRIO(2) |
1189                       ES_PRIO(3));
1190
1191         if ((rdev->family) == CHIP_R600) {
1192                 sq_gpr_resource_mgmt_1 = (NUM_PS_GPRS(124) |
1193                                           NUM_VS_GPRS(124) |
1194                                           NUM_CLAUSE_TEMP_GPRS(4));
1195                 sq_gpr_resource_mgmt_2 = (NUM_GS_GPRS(0) |
1196                                           NUM_ES_GPRS(0));
1197                 sq_thread_resource_mgmt = (NUM_PS_THREADS(136) |
1198                                            NUM_VS_THREADS(48) |
1199                                            NUM_GS_THREADS(4) |
1200                                            NUM_ES_THREADS(4));
1201                 sq_stack_resource_mgmt_1 = (NUM_PS_STACK_ENTRIES(128) |
1202                                             NUM_VS_STACK_ENTRIES(128));
1203                 sq_stack_resource_mgmt_2 = (NUM_GS_STACK_ENTRIES(0) |
1204                                             NUM_ES_STACK_ENTRIES(0));
1205         } else if (((rdev->family) == CHIP_RV610) ||
1206                    ((rdev->family) == CHIP_RV620) ||
1207                    ((rdev->family) == CHIP_RS780) ||
1208                    ((rdev->family) == CHIP_RS880)) {
1209                 /* no vertex cache */
1210                 sq_config &= ~VC_ENABLE;
1211
1212                 sq_gpr_resource_mgmt_1 = (NUM_PS_GPRS(44) |
1213                                           NUM_VS_GPRS(44) |
1214                                           NUM_CLAUSE_TEMP_GPRS(2));
1215                 sq_gpr_resource_mgmt_2 = (NUM_GS_GPRS(17) |
1216                                           NUM_ES_GPRS(17));
1217                 sq_thread_resource_mgmt = (NUM_PS_THREADS(79) |
1218                                            NUM_VS_THREADS(78) |
1219                                            NUM_GS_THREADS(4) |
1220                                            NUM_ES_THREADS(31));
1221                 sq_stack_resource_mgmt_1 = (NUM_PS_STACK_ENTRIES(40) |
1222                                             NUM_VS_STACK_ENTRIES(40));
1223                 sq_stack_resource_mgmt_2 = (NUM_GS_STACK_ENTRIES(32) |
1224                                             NUM_ES_STACK_ENTRIES(16));
1225         } else if (((rdev->family) == CHIP_RV630) ||
1226                    ((rdev->family) == CHIP_RV635)) {
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(18) |
1231                                           NUM_ES_GPRS(18));
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_RV670) {
1241                 sq_gpr_resource_mgmt_1 = (NUM_PS_GPRS(44) |
1242                                           NUM_VS_GPRS(44) |
1243                                           NUM_CLAUSE_TEMP_GPRS(2));
1244                 sq_gpr_resource_mgmt_2 = (NUM_GS_GPRS(17) |
1245                                           NUM_ES_GPRS(17));
1246                 sq_thread_resource_mgmt = (NUM_PS_THREADS(79) |
1247                                            NUM_VS_THREADS(78) |
1248                                            NUM_GS_THREADS(4) |
1249                                            NUM_ES_THREADS(31));
1250                 sq_stack_resource_mgmt_1 = (NUM_PS_STACK_ENTRIES(64) |
1251                                             NUM_VS_STACK_ENTRIES(64));
1252                 sq_stack_resource_mgmt_2 = (NUM_GS_STACK_ENTRIES(64) |
1253                                             NUM_ES_STACK_ENTRIES(64));
1254         }
1255
1256         WREG32(SQ_CONFIG, sq_config);
1257         WREG32(SQ_GPR_RESOURCE_MGMT_1,  sq_gpr_resource_mgmt_1);
1258         WREG32(SQ_GPR_RESOURCE_MGMT_2,  sq_gpr_resource_mgmt_2);
1259         WREG32(SQ_THREAD_RESOURCE_MGMT, sq_thread_resource_mgmt);
1260         WREG32(SQ_STACK_RESOURCE_MGMT_1, sq_stack_resource_mgmt_1);
1261         WREG32(SQ_STACK_RESOURCE_MGMT_2, sq_stack_resource_mgmt_2);
1262
1263         if (((rdev->family) == CHIP_RV610) ||
1264             ((rdev->family) == CHIP_RV620) ||
1265             ((rdev->family) == CHIP_RS780) ||
1266             ((rdev->family) == CHIP_RS880)) {
1267                 WREG32(VGT_CACHE_INVALIDATION, CACHE_INVALIDATION(TC_ONLY));
1268         } else {
1269                 WREG32(VGT_CACHE_INVALIDATION, CACHE_INVALIDATION(VC_AND_TC));
1270         }
1271
1272         /* More default values. 2D/3D driver should adjust as needed */
1273         WREG32(PA_SC_AA_SAMPLE_LOCS_2S, (S0_X(0xc) | S0_Y(0x4) |
1274                                          S1_X(0x4) | S1_Y(0xc)));
1275         WREG32(PA_SC_AA_SAMPLE_LOCS_4S, (S0_X(0xe) | S0_Y(0xe) |
1276                                          S1_X(0x2) | S1_Y(0x2) |
1277                                          S2_X(0xa) | S2_Y(0x6) |
1278                                          S3_X(0x6) | S3_Y(0xa)));
1279         WREG32(PA_SC_AA_SAMPLE_LOCS_8S_WD0, (S0_X(0xe) | S0_Y(0xb) |
1280                                              S1_X(0x4) | S1_Y(0xc) |
1281                                              S2_X(0x1) | S2_Y(0x6) |
1282                                              S3_X(0xa) | S3_Y(0xe)));
1283         WREG32(PA_SC_AA_SAMPLE_LOCS_8S_WD1, (S4_X(0x6) | S4_Y(0x1) |
1284                                              S5_X(0x0) | S5_Y(0x0) |
1285                                              S6_X(0xb) | S6_Y(0x4) |
1286                                              S7_X(0x7) | S7_Y(0x8)));
1287
1288         WREG32(VGT_STRMOUT_EN, 0);
1289         tmp = rdev->config.r600.max_pipes * 16;
1290         switch (rdev->family) {
1291         case CHIP_RV610:
1292         case CHIP_RV620:
1293         case CHIP_RS780:
1294         case CHIP_RS880:
1295                 tmp += 32;
1296                 break;
1297         case CHIP_RV670:
1298                 tmp += 128;
1299                 break;
1300         default:
1301                 break;
1302         }
1303         if (tmp > 256) {
1304                 tmp = 256;
1305         }
1306         WREG32(VGT_ES_PER_GS, 128);
1307         WREG32(VGT_GS_PER_ES, tmp);
1308         WREG32(VGT_GS_PER_VS, 2);
1309         WREG32(VGT_GS_VERTEX_REUSE, 16);
1310
1311         /* more default values. 2D/3D driver should adjust as needed */
1312         WREG32(PA_SC_LINE_STIPPLE_STATE, 0);
1313         WREG32(VGT_STRMOUT_EN, 0);
1314         WREG32(SX_MISC, 0);
1315         WREG32(PA_SC_MODE_CNTL, 0);
1316         WREG32(PA_SC_AA_CONFIG, 0);
1317         WREG32(PA_SC_LINE_STIPPLE, 0);
1318         WREG32(SPI_INPUT_Z, 0);
1319         WREG32(SPI_PS_IN_CONTROL_0, NUM_INTERP(2));
1320         WREG32(CB_COLOR7_FRAG, 0);
1321
1322         /* Clear render buffer base addresses */
1323         WREG32(CB_COLOR0_BASE, 0);
1324         WREG32(CB_COLOR1_BASE, 0);
1325         WREG32(CB_COLOR2_BASE, 0);
1326         WREG32(CB_COLOR3_BASE, 0);
1327         WREG32(CB_COLOR4_BASE, 0);
1328         WREG32(CB_COLOR5_BASE, 0);
1329         WREG32(CB_COLOR6_BASE, 0);
1330         WREG32(CB_COLOR7_BASE, 0);
1331         WREG32(CB_COLOR7_FRAG, 0);
1332
1333         switch (rdev->family) {
1334         case CHIP_RV610:
1335         case CHIP_RV620:
1336         case CHIP_RS780:
1337         case CHIP_RS880:
1338                 tmp = TC_L2_SIZE(8);
1339                 break;
1340         case CHIP_RV630:
1341         case CHIP_RV635:
1342                 tmp = TC_L2_SIZE(4);
1343                 break;
1344         case CHIP_R600:
1345                 tmp = TC_L2_SIZE(0) | L2_DISABLE_LATE_HIT;
1346                 break;
1347         default:
1348                 tmp = TC_L2_SIZE(0);
1349                 break;
1350         }
1351         WREG32(TC_CNTL, tmp);
1352
1353         tmp = RREG32(HDP_HOST_PATH_CNTL);
1354         WREG32(HDP_HOST_PATH_CNTL, tmp);
1355
1356         tmp = RREG32(ARB_POP);
1357         tmp |= ENABLE_TC128;
1358         WREG32(ARB_POP, tmp);
1359
1360         WREG32(PA_SC_MULTI_CHIP_CNTL, 0);
1361         WREG32(PA_CL_ENHANCE, (CLIP_VTX_REORDER_ENA |
1362                                NUM_CLIP_SEQ(3)));
1363         WREG32(PA_SC_ENHANCE, FORCE_EOV_MAX_CLK_CNT(4095));
1364 }
1365
1366
1367 /*
1368  * Indirect registers accessor
1369  */
1370 u32 r600_pciep_rreg(struct radeon_device *rdev, u32 reg)
1371 {
1372         u32 r;
1373
1374         WREG32(PCIE_PORT_INDEX, ((reg) & 0xff));
1375         (void)RREG32(PCIE_PORT_INDEX);
1376         r = RREG32(PCIE_PORT_DATA);
1377         return r;
1378 }
1379
1380 void r600_pciep_wreg(struct radeon_device *rdev, u32 reg, u32 v)
1381 {
1382         WREG32(PCIE_PORT_INDEX, ((reg) & 0xff));
1383         (void)RREG32(PCIE_PORT_INDEX);
1384         WREG32(PCIE_PORT_DATA, (v));
1385         (void)RREG32(PCIE_PORT_DATA);
1386 }
1387
1388 /*
1389  * CP & Ring
1390  */
1391 void r600_cp_stop(struct radeon_device *rdev)
1392 {
1393         WREG32(R_0086D8_CP_ME_CNTL, S_0086D8_CP_ME_HALT(1));
1394 }
1395
1396 int r600_init_microcode(struct radeon_device *rdev)
1397 {
1398         struct platform_device *pdev;
1399         const char *chip_name;
1400         const char *rlc_chip_name;
1401         size_t pfp_req_size, me_req_size, rlc_req_size;
1402         char fw_name[30];
1403         int err;
1404
1405         DRM_DEBUG("\n");
1406
1407         pdev = platform_device_register_simple("radeon_cp", 0, NULL, 0);
1408         err = IS_ERR(pdev);
1409         if (err) {
1410                 printk(KERN_ERR "radeon_cp: Failed to register firmware\n");
1411                 return -EINVAL;
1412         }
1413
1414         switch (rdev->family) {
1415         case CHIP_R600:
1416                 chip_name = "R600";
1417                 rlc_chip_name = "R600";
1418                 break;
1419         case CHIP_RV610:
1420                 chip_name = "RV610";
1421                 rlc_chip_name = "R600";
1422                 break;
1423         case CHIP_RV630:
1424                 chip_name = "RV630";
1425                 rlc_chip_name = "R600";
1426                 break;
1427         case CHIP_RV620:
1428                 chip_name = "RV620";
1429                 rlc_chip_name = "R600";
1430                 break;
1431         case CHIP_RV635:
1432                 chip_name = "RV635";
1433                 rlc_chip_name = "R600";
1434                 break;
1435         case CHIP_RV670:
1436                 chip_name = "RV670";
1437                 rlc_chip_name = "R600";
1438                 break;
1439         case CHIP_RS780:
1440         case CHIP_RS880:
1441                 chip_name = "RS780";
1442                 rlc_chip_name = "R600";
1443                 break;
1444         case CHIP_RV770:
1445                 chip_name = "RV770";
1446                 rlc_chip_name = "R700";
1447                 break;
1448         case CHIP_RV730:
1449         case CHIP_RV740:
1450                 chip_name = "RV730";
1451                 rlc_chip_name = "R700";
1452                 break;
1453         case CHIP_RV710:
1454                 chip_name = "RV710";
1455                 rlc_chip_name = "R700";
1456                 break;
1457         default: BUG();
1458         }
1459
1460         if (rdev->family >= CHIP_RV770) {
1461                 pfp_req_size = R700_PFP_UCODE_SIZE * 4;
1462                 me_req_size = R700_PM4_UCODE_SIZE * 4;
1463                 rlc_req_size = R700_RLC_UCODE_SIZE * 4;
1464         } else {
1465                 pfp_req_size = PFP_UCODE_SIZE * 4;
1466                 me_req_size = PM4_UCODE_SIZE * 12;
1467                 rlc_req_size = RLC_UCODE_SIZE * 4;
1468         }
1469
1470         DRM_INFO("Loading %s Microcode\n", chip_name);
1471
1472         snprintf(fw_name, sizeof(fw_name), "radeon/%s_pfp.bin", chip_name);
1473         err = request_firmware(&rdev->pfp_fw, fw_name, &pdev->dev);
1474         if (err)
1475                 goto out;
1476         if (rdev->pfp_fw->size != pfp_req_size) {
1477                 printk(KERN_ERR
1478                        "r600_cp: Bogus length %zu in firmware \"%s\"\n",
1479                        rdev->pfp_fw->size, fw_name);
1480                 err = -EINVAL;
1481                 goto out;
1482         }
1483
1484         snprintf(fw_name, sizeof(fw_name), "radeon/%s_me.bin", chip_name);
1485         err = request_firmware(&rdev->me_fw, fw_name, &pdev->dev);
1486         if (err)
1487                 goto out;
1488         if (rdev->me_fw->size != me_req_size) {
1489                 printk(KERN_ERR
1490                        "r600_cp: Bogus length %zu in firmware \"%s\"\n",
1491                        rdev->me_fw->size, fw_name);
1492                 err = -EINVAL;
1493         }
1494
1495         snprintf(fw_name, sizeof(fw_name), "radeon/%s_rlc.bin", rlc_chip_name);
1496         err = request_firmware(&rdev->rlc_fw, fw_name, &pdev->dev);
1497         if (err)
1498                 goto out;
1499         if (rdev->rlc_fw->size != rlc_req_size) {
1500                 printk(KERN_ERR
1501                        "r600_rlc: Bogus length %zu in firmware \"%s\"\n",
1502                        rdev->rlc_fw->size, fw_name);
1503                 err = -EINVAL;
1504         }
1505
1506 out:
1507         platform_device_unregister(pdev);
1508
1509         if (err) {
1510                 if (err != -EINVAL)
1511                         printk(KERN_ERR
1512                                "r600_cp: Failed to load firmware \"%s\"\n",
1513                                fw_name);
1514                 release_firmware(rdev->pfp_fw);
1515                 rdev->pfp_fw = NULL;
1516                 release_firmware(rdev->me_fw);
1517                 rdev->me_fw = NULL;
1518                 release_firmware(rdev->rlc_fw);
1519                 rdev->rlc_fw = NULL;
1520         }
1521         return err;
1522 }
1523
1524 static int r600_cp_load_microcode(struct radeon_device *rdev)
1525 {
1526         const __be32 *fw_data;
1527         int i;
1528
1529         if (!rdev->me_fw || !rdev->pfp_fw)
1530                 return -EINVAL;
1531
1532         r600_cp_stop(rdev);
1533
1534         WREG32(CP_RB_CNTL, RB_NO_UPDATE | RB_BLKSZ(15) | RB_BUFSZ(3));
1535
1536         /* Reset cp */
1537         WREG32(GRBM_SOFT_RESET, SOFT_RESET_CP);
1538         RREG32(GRBM_SOFT_RESET);
1539         mdelay(15);
1540         WREG32(GRBM_SOFT_RESET, 0);
1541
1542         WREG32(CP_ME_RAM_WADDR, 0);
1543
1544         fw_data = (const __be32 *)rdev->me_fw->data;
1545         WREG32(CP_ME_RAM_WADDR, 0);
1546         for (i = 0; i < PM4_UCODE_SIZE * 3; i++)
1547                 WREG32(CP_ME_RAM_DATA,
1548                        be32_to_cpup(fw_data++));
1549
1550         fw_data = (const __be32 *)rdev->pfp_fw->data;
1551         WREG32(CP_PFP_UCODE_ADDR, 0);
1552         for (i = 0; i < PFP_UCODE_SIZE; i++)
1553                 WREG32(CP_PFP_UCODE_DATA,
1554                        be32_to_cpup(fw_data++));
1555
1556         WREG32(CP_PFP_UCODE_ADDR, 0);
1557         WREG32(CP_ME_RAM_WADDR, 0);
1558         WREG32(CP_ME_RAM_RADDR, 0);
1559         return 0;
1560 }
1561
1562 int r600_cp_start(struct radeon_device *rdev)
1563 {
1564         int r;
1565         uint32_t cp_me;
1566
1567         r = radeon_ring_lock(rdev, 7);
1568         if (r) {
1569                 DRM_ERROR("radeon: cp failed to lock ring (%d).\n", r);
1570                 return r;
1571         }
1572         radeon_ring_write(rdev, PACKET3(PACKET3_ME_INITIALIZE, 5));
1573         radeon_ring_write(rdev, 0x1);
1574         if (rdev->family < CHIP_RV770) {
1575                 radeon_ring_write(rdev, 0x3);
1576                 radeon_ring_write(rdev, rdev->config.r600.max_hw_contexts - 1);
1577         } else {
1578                 radeon_ring_write(rdev, 0x0);
1579                 radeon_ring_write(rdev, rdev->config.rv770.max_hw_contexts - 1);
1580         }
1581         radeon_ring_write(rdev, PACKET3_ME_INITIALIZE_DEVICE_ID(1));
1582         radeon_ring_write(rdev, 0);
1583         radeon_ring_write(rdev, 0);
1584         radeon_ring_unlock_commit(rdev);
1585
1586         cp_me = 0xff;
1587         WREG32(R_0086D8_CP_ME_CNTL, cp_me);
1588         return 0;
1589 }
1590
1591 int r600_cp_resume(struct radeon_device *rdev)
1592 {
1593         u32 tmp;
1594         u32 rb_bufsz;
1595         int r;
1596
1597         /* Reset cp */
1598         WREG32(GRBM_SOFT_RESET, SOFT_RESET_CP);
1599         RREG32(GRBM_SOFT_RESET);
1600         mdelay(15);
1601         WREG32(GRBM_SOFT_RESET, 0);
1602
1603         /* Set ring buffer size */
1604         rb_bufsz = drm_order(rdev->cp.ring_size / 8);
1605         tmp = RB_NO_UPDATE | (drm_order(RADEON_GPU_PAGE_SIZE/8) << 8) | rb_bufsz;
1606 #ifdef __BIG_ENDIAN
1607         tmp |= BUF_SWAP_32BIT;
1608 #endif
1609         WREG32(CP_RB_CNTL, tmp);
1610         WREG32(CP_SEM_WAIT_TIMER, 0x4);
1611
1612         /* Set the write pointer delay */
1613         WREG32(CP_RB_WPTR_DELAY, 0);
1614
1615         /* Initialize the ring buffer's read and write pointers */
1616         WREG32(CP_RB_CNTL, tmp | RB_RPTR_WR_ENA);
1617         WREG32(CP_RB_RPTR_WR, 0);
1618         WREG32(CP_RB_WPTR, 0);
1619         WREG32(CP_RB_RPTR_ADDR, rdev->cp.gpu_addr & 0xFFFFFFFF);
1620         WREG32(CP_RB_RPTR_ADDR_HI, upper_32_bits(rdev->cp.gpu_addr));
1621         mdelay(1);
1622         WREG32(CP_RB_CNTL, tmp);
1623
1624         WREG32(CP_RB_BASE, rdev->cp.gpu_addr >> 8);
1625         WREG32(CP_DEBUG, (1 << 27) | (1 << 28));
1626
1627         rdev->cp.rptr = RREG32(CP_RB_RPTR);
1628         rdev->cp.wptr = RREG32(CP_RB_WPTR);
1629
1630         r600_cp_start(rdev);
1631         rdev->cp.ready = true;
1632         r = radeon_ring_test(rdev);
1633         if (r) {
1634                 rdev->cp.ready = false;
1635                 return r;
1636         }
1637         return 0;
1638 }
1639
1640 void r600_cp_commit(struct radeon_device *rdev)
1641 {
1642         WREG32(CP_RB_WPTR, rdev->cp.wptr);
1643         (void)RREG32(CP_RB_WPTR);
1644 }
1645
1646 void r600_ring_init(struct radeon_device *rdev, unsigned ring_size)
1647 {
1648         u32 rb_bufsz;
1649
1650         /* Align ring size */
1651         rb_bufsz = drm_order(ring_size / 8);
1652         ring_size = (1 << (rb_bufsz + 1)) * 4;
1653         rdev->cp.ring_size = ring_size;
1654         rdev->cp.align_mask = 16 - 1;
1655 }
1656
1657 void r600_cp_fini(struct radeon_device *rdev)
1658 {
1659         r600_cp_stop(rdev);
1660         radeon_ring_fini(rdev);
1661 }
1662
1663
1664 /*
1665  * GPU scratch registers helpers function.
1666  */
1667 void r600_scratch_init(struct radeon_device *rdev)
1668 {
1669         int i;
1670
1671         rdev->scratch.num_reg = 7;
1672         for (i = 0; i < rdev->scratch.num_reg; i++) {
1673                 rdev->scratch.free[i] = true;
1674                 rdev->scratch.reg[i] = SCRATCH_REG0 + (i * 4);
1675         }
1676 }
1677
1678 int r600_ring_test(struct radeon_device *rdev)
1679 {
1680         uint32_t scratch;
1681         uint32_t tmp = 0;
1682         unsigned i;
1683         int r;
1684
1685         r = radeon_scratch_get(rdev, &scratch);
1686         if (r) {
1687                 DRM_ERROR("radeon: cp failed to get scratch reg (%d).\n", r);
1688                 return r;
1689         }
1690         WREG32(scratch, 0xCAFEDEAD);
1691         r = radeon_ring_lock(rdev, 3);
1692         if (r) {
1693                 DRM_ERROR("radeon: cp failed to lock ring (%d).\n", r);
1694                 radeon_scratch_free(rdev, scratch);
1695                 return r;
1696         }
1697         radeon_ring_write(rdev, PACKET3(PACKET3_SET_CONFIG_REG, 1));
1698         radeon_ring_write(rdev, ((scratch - PACKET3_SET_CONFIG_REG_OFFSET) >> 2));
1699         radeon_ring_write(rdev, 0xDEADBEEF);
1700         radeon_ring_unlock_commit(rdev);
1701         for (i = 0; i < rdev->usec_timeout; i++) {
1702                 tmp = RREG32(scratch);
1703                 if (tmp == 0xDEADBEEF)
1704                         break;
1705                 DRM_UDELAY(1);
1706         }
1707         if (i < rdev->usec_timeout) {
1708                 DRM_INFO("ring test succeeded in %d usecs\n", i);
1709         } else {
1710                 DRM_ERROR("radeon: ring test failed (scratch(0x%04X)=0x%08X)\n",
1711                           scratch, tmp);
1712                 r = -EINVAL;
1713         }
1714         radeon_scratch_free(rdev, scratch);
1715         return r;
1716 }
1717
1718 void r600_wb_disable(struct radeon_device *rdev)
1719 {
1720         int r;
1721
1722         WREG32(SCRATCH_UMSK, 0);
1723         if (rdev->wb.wb_obj) {
1724                 r = radeon_bo_reserve(rdev->wb.wb_obj, false);
1725                 if (unlikely(r != 0))
1726                         return;
1727                 radeon_bo_kunmap(rdev->wb.wb_obj);
1728                 radeon_bo_unpin(rdev->wb.wb_obj);
1729                 radeon_bo_unreserve(rdev->wb.wb_obj);
1730         }
1731 }
1732
1733 void r600_wb_fini(struct radeon_device *rdev)
1734 {
1735         r600_wb_disable(rdev);
1736         if (rdev->wb.wb_obj) {
1737                 radeon_bo_unref(&rdev->wb.wb_obj);
1738                 rdev->wb.wb = NULL;
1739                 rdev->wb.wb_obj = NULL;
1740         }
1741 }
1742
1743 int r600_wb_enable(struct radeon_device *rdev)
1744 {
1745         int r;
1746
1747         if (rdev->wb.wb_obj == NULL) {
1748                 r = radeon_bo_create(rdev, NULL, RADEON_GPU_PAGE_SIZE, true,
1749                                 RADEON_GEM_DOMAIN_GTT, &rdev->wb.wb_obj);
1750                 if (r) {
1751                         dev_warn(rdev->dev, "(%d) create WB bo failed\n", r);
1752                         return r;
1753                 }
1754                 r = radeon_bo_reserve(rdev->wb.wb_obj, false);
1755                 if (unlikely(r != 0)) {
1756                         r600_wb_fini(rdev);
1757                         return r;
1758                 }
1759                 r = radeon_bo_pin(rdev->wb.wb_obj, RADEON_GEM_DOMAIN_GTT,
1760                                 &rdev->wb.gpu_addr);
1761                 if (r) {
1762                         radeon_bo_unreserve(rdev->wb.wb_obj);
1763                         dev_warn(rdev->dev, "(%d) pin WB bo failed\n", r);
1764                         r600_wb_fini(rdev);
1765                         return r;
1766                 }
1767                 r = radeon_bo_kmap(rdev->wb.wb_obj, (void **)&rdev->wb.wb);
1768                 radeon_bo_unreserve(rdev->wb.wb_obj);
1769                 if (r) {
1770                         dev_warn(rdev->dev, "(%d) map WB bo failed\n", r);
1771                         r600_wb_fini(rdev);
1772                         return r;
1773                 }
1774         }
1775         WREG32(SCRATCH_ADDR, (rdev->wb.gpu_addr >> 8) & 0xFFFFFFFF);
1776         WREG32(CP_RB_RPTR_ADDR, (rdev->wb.gpu_addr + 1024) & 0xFFFFFFFC);
1777         WREG32(CP_RB_RPTR_ADDR_HI, upper_32_bits(rdev->wb.gpu_addr + 1024) & 0xFF);
1778         WREG32(SCRATCH_UMSK, 0xff);
1779         return 0;
1780 }
1781
1782 void r600_fence_ring_emit(struct radeon_device *rdev,
1783                           struct radeon_fence *fence)
1784 {
1785         /* Also consider EVENT_WRITE_EOP.  it handles the interrupts + timestamps + events */
1786         /* Emit fence sequence & fire IRQ */
1787         radeon_ring_write(rdev, PACKET3(PACKET3_SET_CONFIG_REG, 1));
1788         radeon_ring_write(rdev, ((rdev->fence_drv.scratch_reg - PACKET3_SET_CONFIG_REG_OFFSET) >> 2));
1789         radeon_ring_write(rdev, fence->seq);
1790         radeon_ring_write(rdev, PACKET0(R_005480_HDP_MEM_COHERENCY_FLUSH_CNTL, 0));
1791         radeon_ring_write(rdev, 1);
1792         /* CP_INTERRUPT packet 3 no longer exists, use packet 0 */
1793         radeon_ring_write(rdev, PACKET0(CP_INT_STATUS, 0));
1794         radeon_ring_write(rdev, RB_INT_STAT);
1795 }
1796
1797 int r600_copy_blit(struct radeon_device *rdev,
1798                    uint64_t src_offset, uint64_t dst_offset,
1799                    unsigned num_pages, struct radeon_fence *fence)
1800 {
1801         int r;
1802
1803         mutex_lock(&rdev->r600_blit.mutex);
1804         rdev->r600_blit.vb_ib = NULL;
1805         r = r600_blit_prepare_copy(rdev, num_pages * RADEON_GPU_PAGE_SIZE);
1806         if (r) {
1807                 if (rdev->r600_blit.vb_ib)
1808                         radeon_ib_free(rdev, &rdev->r600_blit.vb_ib);
1809                 mutex_unlock(&rdev->r600_blit.mutex);
1810                 return r;
1811         }
1812         r600_kms_blit_copy(rdev, src_offset, dst_offset, num_pages * RADEON_GPU_PAGE_SIZE);
1813         r600_blit_done_copy(rdev, fence);
1814         mutex_unlock(&rdev->r600_blit.mutex);
1815         return 0;
1816 }
1817
1818 int r600_set_surface_reg(struct radeon_device *rdev, int reg,
1819                          uint32_t tiling_flags, uint32_t pitch,
1820                          uint32_t offset, uint32_t obj_size)
1821 {
1822         /* FIXME: implement */
1823         return 0;
1824 }
1825
1826 void r600_clear_surface_reg(struct radeon_device *rdev, int reg)
1827 {
1828         /* FIXME: implement */
1829 }
1830
1831
1832 bool r600_card_posted(struct radeon_device *rdev)
1833 {
1834         uint32_t reg;
1835
1836         /* first check CRTCs */
1837         reg = RREG32(D1CRTC_CONTROL) |
1838                 RREG32(D2CRTC_CONTROL);
1839         if (reg & CRTC_EN)
1840                 return true;
1841
1842         /* then check MEM_SIZE, in case the crtcs are off */
1843         if (RREG32(CONFIG_MEMSIZE))
1844                 return true;
1845
1846         return false;
1847 }
1848
1849 int r600_startup(struct radeon_device *rdev)
1850 {
1851         int r;
1852
1853         if (!rdev->me_fw || !rdev->pfp_fw || !rdev->rlc_fw) {
1854                 r = r600_init_microcode(rdev);
1855                 if (r) {
1856                         DRM_ERROR("Failed to load firmware!\n");
1857                         return r;
1858                 }
1859         }
1860
1861         r600_mc_program(rdev);
1862         if (rdev->flags & RADEON_IS_AGP) {
1863                 r600_agp_enable(rdev);
1864         } else {
1865                 r = r600_pcie_gart_enable(rdev);
1866                 if (r)
1867                         return r;
1868         }
1869         r600_gpu_init(rdev);
1870         r = r600_blit_init(rdev);
1871         if (r) {
1872                 r600_blit_fini(rdev);
1873                 rdev->asic->copy = NULL;
1874                 dev_warn(rdev->dev, "failed blitter (%d) falling back to memcpy\n", r);
1875         }
1876         /* pin copy shader into vram */
1877         if (rdev->r600_blit.shader_obj) {
1878                 r = radeon_bo_reserve(rdev->r600_blit.shader_obj, false);
1879                 if (unlikely(r != 0))
1880                         return r;
1881                 r = radeon_bo_pin(rdev->r600_blit.shader_obj, RADEON_GEM_DOMAIN_VRAM,
1882                                 &rdev->r600_blit.shader_gpu_addr);
1883                 radeon_bo_unreserve(rdev->r600_blit.shader_obj);
1884                 if (r) {
1885                         dev_err(rdev->dev, "(%d) pin blit object failed\n", r);
1886                         return r;
1887                 }
1888         }
1889         /* Enable IRQ */
1890         r = r600_irq_init(rdev);
1891         if (r) {
1892                 DRM_ERROR("radeon: IH init failed (%d).\n", r);
1893                 radeon_irq_kms_fini(rdev);
1894                 return r;
1895         }
1896         r600_irq_set(rdev);
1897
1898         r = radeon_ring_init(rdev, rdev->cp.ring_size);
1899         if (r)
1900                 return r;
1901         r = r600_cp_load_microcode(rdev);
1902         if (r)
1903                 return r;
1904         r = r600_cp_resume(rdev);
1905         if (r)
1906                 return r;
1907         /* write back buffer are not vital so don't worry about failure */
1908         r600_wb_enable(rdev);
1909         return 0;
1910 }
1911
1912 void r600_vga_set_state(struct radeon_device *rdev, bool state)
1913 {
1914         uint32_t temp;
1915
1916         temp = RREG32(CONFIG_CNTL);
1917         if (state == false) {
1918                 temp &= ~(1<<0);
1919                 temp |= (1<<1);
1920         } else {
1921                 temp &= ~(1<<1);
1922         }
1923         WREG32(CONFIG_CNTL, temp);
1924 }
1925
1926 int r600_resume(struct radeon_device *rdev)
1927 {
1928         int r;
1929
1930         /* Do not reset GPU before posting, on r600 hw unlike on r500 hw,
1931          * posting will perform necessary task to bring back GPU into good
1932          * shape.
1933          */
1934         /* post card */
1935         atom_asic_init(rdev->mode_info.atom_context);
1936         /* Initialize clocks */
1937         r = radeon_clocks_init(rdev);
1938         if (r) {
1939                 return r;
1940         }
1941
1942         r = r600_startup(rdev);
1943         if (r) {
1944                 DRM_ERROR("r600 startup failed on resume\n");
1945                 return r;
1946         }
1947
1948         r = r600_ib_test(rdev);
1949         if (r) {
1950                 DRM_ERROR("radeon: failled testing IB (%d).\n", r);
1951                 return r;
1952         }
1953         return r;
1954 }
1955
1956 int r600_suspend(struct radeon_device *rdev)
1957 {
1958         int r;
1959
1960         /* FIXME: we should wait for ring to be empty */
1961         r600_cp_stop(rdev);
1962         rdev->cp.ready = false;
1963         r600_irq_suspend(rdev);
1964         r600_wb_disable(rdev);
1965         r600_pcie_gart_disable(rdev);
1966         /* unpin shaders bo */
1967         if (rdev->r600_blit.shader_obj) {
1968                 r = radeon_bo_reserve(rdev->r600_blit.shader_obj, false);
1969                 if (!r) {
1970                         radeon_bo_unpin(rdev->r600_blit.shader_obj);
1971                         radeon_bo_unreserve(rdev->r600_blit.shader_obj);
1972                 }
1973         }
1974         return 0;
1975 }
1976
1977 /* Plan is to move initialization in that function and use
1978  * helper function so that radeon_device_init pretty much
1979  * do nothing more than calling asic specific function. This
1980  * should also allow to remove a bunch of callback function
1981  * like vram_info.
1982  */
1983 int r600_init(struct radeon_device *rdev)
1984 {
1985         int r;
1986
1987         r = radeon_dummy_page_init(rdev);
1988         if (r)
1989                 return r;
1990         if (r600_debugfs_mc_info_init(rdev)) {
1991                 DRM_ERROR("Failed to register debugfs file for mc !\n");
1992         }
1993         /* This don't do much */
1994         r = radeon_gem_init(rdev);
1995         if (r)
1996                 return r;
1997         /* Read BIOS */
1998         if (!radeon_get_bios(rdev)) {
1999                 if (ASIC_IS_AVIVO(rdev))
2000                         return -EINVAL;
2001         }
2002         /* Must be an ATOMBIOS */
2003         if (!rdev->is_atom_bios) {
2004                 dev_err(rdev->dev, "Expecting atombios for R600 GPU\n");
2005                 return -EINVAL;
2006         }
2007         r = radeon_atombios_init(rdev);
2008         if (r)
2009                 return r;
2010         /* Post card if necessary */
2011         if (!r600_card_posted(rdev)) {
2012                 if (!rdev->bios) {
2013                         dev_err(rdev->dev, "Card not posted and no BIOS - ignoring\n");
2014                         return -EINVAL;
2015                 }
2016                 DRM_INFO("GPU not posted. posting now...\n");
2017                 atom_asic_init(rdev->mode_info.atom_context);
2018         }
2019         /* Initialize scratch registers */
2020         r600_scratch_init(rdev);
2021         /* Initialize surface registers */
2022         radeon_surface_init(rdev);
2023         /* Initialize clocks */
2024         radeon_get_clock_info(rdev->ddev);
2025         r = radeon_clocks_init(rdev);
2026         if (r)
2027                 return r;
2028         /* Initialize power management */
2029         radeon_pm_init(rdev);
2030         /* Fence driver */
2031         r = radeon_fence_driver_init(rdev);
2032         if (r)
2033                 return r;
2034         if (rdev->flags & RADEON_IS_AGP) {
2035                 r = radeon_agp_init(rdev);
2036                 if (r)
2037                         radeon_agp_disable(rdev);
2038         }
2039         r = r600_mc_init(rdev);
2040         if (r)
2041                 return r;
2042         /* Memory manager */
2043         r = radeon_bo_init(rdev);
2044         if (r)
2045                 return r;
2046
2047         r = radeon_irq_kms_init(rdev);
2048         if (r)
2049                 return r;
2050
2051         rdev->cp.ring_obj = NULL;
2052         r600_ring_init(rdev, 1024 * 1024);
2053
2054         rdev->ih.ring_obj = NULL;
2055         r600_ih_ring_init(rdev, 64 * 1024);
2056
2057         r = r600_pcie_gart_init(rdev);
2058         if (r)
2059                 return r;
2060
2061         rdev->accel_working = true;
2062         r = r600_startup(rdev);
2063         if (r) {
2064                 dev_err(rdev->dev, "disabling GPU acceleration\n");
2065                 r600_cp_fini(rdev);
2066                 r600_wb_fini(rdev);
2067                 r600_irq_fini(rdev);
2068                 radeon_irq_kms_fini(rdev);
2069                 r600_pcie_gart_fini(rdev);
2070                 rdev->accel_working = false;
2071         }
2072         if (rdev->accel_working) {
2073                 r = radeon_ib_pool_init(rdev);
2074                 if (r) {
2075                         dev_err(rdev->dev, "IB initialization failed (%d).\n", r);
2076                         rdev->accel_working = false;
2077                 } else {
2078                         r = r600_ib_test(rdev);
2079                         if (r) {
2080                                 dev_err(rdev->dev, "IB test failed (%d).\n", r);
2081                                 rdev->accel_working = false;
2082                         }
2083                 }
2084         }
2085
2086         r = r600_audio_init(rdev);
2087         if (r)
2088                 return r; /* TODO error handling */
2089         return 0;
2090 }
2091
2092 void r600_fini(struct radeon_device *rdev)
2093 {
2094         r600_audio_fini(rdev);
2095         r600_blit_fini(rdev);
2096         r600_cp_fini(rdev);
2097         r600_wb_fini(rdev);
2098         r600_irq_fini(rdev);
2099         radeon_irq_kms_fini(rdev);
2100         r600_pcie_gart_fini(rdev);
2101         radeon_agp_fini(rdev);
2102         radeon_gem_fini(rdev);
2103         radeon_fence_driver_fini(rdev);
2104         radeon_clocks_fini(rdev);
2105         radeon_bo_fini(rdev);
2106         radeon_atombios_fini(rdev);
2107         kfree(rdev->bios);
2108         rdev->bios = NULL;
2109         radeon_dummy_page_fini(rdev);
2110 }
2111
2112
2113 /*
2114  * CS stuff
2115  */
2116 void r600_ring_ib_execute(struct radeon_device *rdev, struct radeon_ib *ib)
2117 {
2118         /* FIXME: implement */
2119         radeon_ring_write(rdev, PACKET3(PACKET3_INDIRECT_BUFFER, 2));
2120         radeon_ring_write(rdev, ib->gpu_addr & 0xFFFFFFFC);
2121         radeon_ring_write(rdev, upper_32_bits(ib->gpu_addr) & 0xFF);
2122         radeon_ring_write(rdev, ib->length_dw);
2123 }
2124
2125 int r600_ib_test(struct radeon_device *rdev)
2126 {
2127         struct radeon_ib *ib;
2128         uint32_t scratch;
2129         uint32_t tmp = 0;
2130         unsigned i;
2131         int r;
2132
2133         r = radeon_scratch_get(rdev, &scratch);
2134         if (r) {
2135                 DRM_ERROR("radeon: failed to get scratch reg (%d).\n", r);
2136                 return r;
2137         }
2138         WREG32(scratch, 0xCAFEDEAD);
2139         r = radeon_ib_get(rdev, &ib);
2140         if (r) {
2141                 DRM_ERROR("radeon: failed to get ib (%d).\n", r);
2142                 return r;
2143         }
2144         ib->ptr[0] = PACKET3(PACKET3_SET_CONFIG_REG, 1);
2145         ib->ptr[1] = ((scratch - PACKET3_SET_CONFIG_REG_OFFSET) >> 2);
2146         ib->ptr[2] = 0xDEADBEEF;
2147         ib->ptr[3] = PACKET2(0);
2148         ib->ptr[4] = PACKET2(0);
2149         ib->ptr[5] = PACKET2(0);
2150         ib->ptr[6] = PACKET2(0);
2151         ib->ptr[7] = PACKET2(0);
2152         ib->ptr[8] = PACKET2(0);
2153         ib->ptr[9] = PACKET2(0);
2154         ib->ptr[10] = PACKET2(0);
2155         ib->ptr[11] = PACKET2(0);
2156         ib->ptr[12] = PACKET2(0);
2157         ib->ptr[13] = PACKET2(0);
2158         ib->ptr[14] = PACKET2(0);
2159         ib->ptr[15] = PACKET2(0);
2160         ib->length_dw = 16;
2161         r = radeon_ib_schedule(rdev, ib);
2162         if (r) {
2163                 radeon_scratch_free(rdev, scratch);
2164                 radeon_ib_free(rdev, &ib);
2165                 DRM_ERROR("radeon: failed to schedule ib (%d).\n", r);
2166                 return r;
2167         }
2168         r = radeon_fence_wait(ib->fence, false);
2169         if (r) {
2170                 DRM_ERROR("radeon: fence wait failed (%d).\n", r);
2171                 return r;
2172         }
2173         for (i = 0; i < rdev->usec_timeout; i++) {
2174                 tmp = RREG32(scratch);
2175                 if (tmp == 0xDEADBEEF)
2176                         break;
2177                 DRM_UDELAY(1);
2178         }
2179         if (i < rdev->usec_timeout) {
2180                 DRM_INFO("ib test succeeded in %u usecs\n", i);
2181         } else {
2182                 DRM_ERROR("radeon: ib test failed (sracth(0x%04X)=0x%08X)\n",
2183                           scratch, tmp);
2184                 r = -EINVAL;
2185         }
2186         radeon_scratch_free(rdev, scratch);
2187         radeon_ib_free(rdev, &ib);
2188         return r;
2189 }
2190
2191 /*
2192  * Interrupts
2193  *
2194  * Interrupts use a ring buffer on r6xx/r7xx hardware.  It works pretty
2195  * the same as the CP ring buffer, but in reverse.  Rather than the CPU
2196  * writing to the ring and the GPU consuming, the GPU writes to the ring
2197  * and host consumes.  As the host irq handler processes interrupts, it
2198  * increments the rptr.  When the rptr catches up with the wptr, all the
2199  * current interrupts have been processed.
2200  */
2201
2202 void r600_ih_ring_init(struct radeon_device *rdev, unsigned ring_size)
2203 {
2204         u32 rb_bufsz;
2205
2206         /* Align ring size */
2207         rb_bufsz = drm_order(ring_size / 4);
2208         ring_size = (1 << rb_bufsz) * 4;
2209         rdev->ih.ring_size = ring_size;
2210         rdev->ih.ptr_mask = rdev->ih.ring_size - 1;
2211         rdev->ih.rptr = 0;
2212 }
2213
2214 static int r600_ih_ring_alloc(struct radeon_device *rdev)
2215 {
2216         int r;
2217
2218         /* Allocate ring buffer */
2219         if (rdev->ih.ring_obj == NULL) {
2220                 r = radeon_bo_create(rdev, NULL, rdev->ih.ring_size,
2221                                      true,
2222                                      RADEON_GEM_DOMAIN_GTT,
2223                                      &rdev->ih.ring_obj);
2224                 if (r) {
2225                         DRM_ERROR("radeon: failed to create ih ring buffer (%d).\n", r);
2226                         return r;
2227                 }
2228                 r = radeon_bo_reserve(rdev->ih.ring_obj, false);
2229                 if (unlikely(r != 0))
2230                         return r;
2231                 r = radeon_bo_pin(rdev->ih.ring_obj,
2232                                   RADEON_GEM_DOMAIN_GTT,
2233                                   &rdev->ih.gpu_addr);
2234                 if (r) {
2235                         radeon_bo_unreserve(rdev->ih.ring_obj);
2236                         DRM_ERROR("radeon: failed to pin ih ring buffer (%d).\n", r);
2237                         return r;
2238                 }
2239                 r = radeon_bo_kmap(rdev->ih.ring_obj,
2240                                    (void **)&rdev->ih.ring);
2241                 radeon_bo_unreserve(rdev->ih.ring_obj);
2242                 if (r) {
2243                         DRM_ERROR("radeon: failed to map ih ring buffer (%d).\n", r);
2244                         return r;
2245                 }
2246         }
2247         return 0;
2248 }
2249
2250 static void r600_ih_ring_fini(struct radeon_device *rdev)
2251 {
2252         int r;
2253         if (rdev->ih.ring_obj) {
2254                 r = radeon_bo_reserve(rdev->ih.ring_obj, false);
2255                 if (likely(r == 0)) {
2256                         radeon_bo_kunmap(rdev->ih.ring_obj);
2257                         radeon_bo_unpin(rdev->ih.ring_obj);
2258                         radeon_bo_unreserve(rdev->ih.ring_obj);
2259                 }
2260                 radeon_bo_unref(&rdev->ih.ring_obj);
2261                 rdev->ih.ring = NULL;
2262                 rdev->ih.ring_obj = NULL;
2263         }
2264 }
2265
2266 static void r600_rlc_stop(struct radeon_device *rdev)
2267 {
2268
2269         if (rdev->family >= CHIP_RV770) {
2270                 /* r7xx asics need to soft reset RLC before halting */
2271                 WREG32(SRBM_SOFT_RESET, SOFT_RESET_RLC);
2272                 RREG32(SRBM_SOFT_RESET);
2273                 udelay(15000);
2274                 WREG32(SRBM_SOFT_RESET, 0);
2275                 RREG32(SRBM_SOFT_RESET);
2276         }
2277
2278         WREG32(RLC_CNTL, 0);
2279 }
2280
2281 static void r600_rlc_start(struct radeon_device *rdev)
2282 {
2283         WREG32(RLC_CNTL, RLC_ENABLE);
2284 }
2285
2286 static int r600_rlc_init(struct radeon_device *rdev)
2287 {
2288         u32 i;
2289         const __be32 *fw_data;
2290
2291         if (!rdev->rlc_fw)
2292                 return -EINVAL;
2293
2294         r600_rlc_stop(rdev);
2295
2296         WREG32(RLC_HB_BASE, 0);
2297         WREG32(RLC_HB_CNTL, 0);
2298         WREG32(RLC_HB_RPTR, 0);
2299         WREG32(RLC_HB_WPTR, 0);
2300         WREG32(RLC_HB_WPTR_LSB_ADDR, 0);
2301         WREG32(RLC_HB_WPTR_MSB_ADDR, 0);
2302         WREG32(RLC_MC_CNTL, 0);
2303         WREG32(RLC_UCODE_CNTL, 0);
2304
2305         fw_data = (const __be32 *)rdev->rlc_fw->data;
2306         if (rdev->family >= CHIP_RV770) {
2307                 for (i = 0; i < R700_RLC_UCODE_SIZE; i++) {
2308                         WREG32(RLC_UCODE_ADDR, i);
2309                         WREG32(RLC_UCODE_DATA, be32_to_cpup(fw_data++));
2310                 }
2311         } else {
2312                 for (i = 0; i < RLC_UCODE_SIZE; i++) {
2313                         WREG32(RLC_UCODE_ADDR, i);
2314                         WREG32(RLC_UCODE_DATA, be32_to_cpup(fw_data++));
2315                 }
2316         }
2317         WREG32(RLC_UCODE_ADDR, 0);
2318
2319         r600_rlc_start(rdev);
2320
2321         return 0;
2322 }
2323
2324 static void r600_enable_interrupts(struct radeon_device *rdev)
2325 {
2326         u32 ih_cntl = RREG32(IH_CNTL);
2327         u32 ih_rb_cntl = RREG32(IH_RB_CNTL);
2328
2329         ih_cntl |= ENABLE_INTR;
2330         ih_rb_cntl |= IH_RB_ENABLE;
2331         WREG32(IH_CNTL, ih_cntl);
2332         WREG32(IH_RB_CNTL, ih_rb_cntl);
2333         rdev->ih.enabled = true;
2334 }
2335
2336 static void r600_disable_interrupts(struct radeon_device *rdev)
2337 {
2338         u32 ih_rb_cntl = RREG32(IH_RB_CNTL);
2339         u32 ih_cntl = RREG32(IH_CNTL);
2340
2341         ih_rb_cntl &= ~IH_RB_ENABLE;
2342         ih_cntl &= ~ENABLE_INTR;
2343         WREG32(IH_RB_CNTL, ih_rb_cntl);
2344         WREG32(IH_CNTL, ih_cntl);
2345         /* set rptr, wptr to 0 */
2346         WREG32(IH_RB_RPTR, 0);
2347         WREG32(IH_RB_WPTR, 0);
2348         rdev->ih.enabled = false;
2349         rdev->ih.wptr = 0;
2350         rdev->ih.rptr = 0;
2351 }
2352
2353 static void r600_disable_interrupt_state(struct radeon_device *rdev)
2354 {
2355         u32 tmp;
2356
2357         WREG32(CP_INT_CNTL, 0);
2358         WREG32(GRBM_INT_CNTL, 0);
2359         WREG32(DxMODE_INT_MASK, 0);
2360         if (ASIC_IS_DCE3(rdev)) {
2361                 WREG32(DCE3_DACA_AUTODETECT_INT_CONTROL, 0);
2362                 WREG32(DCE3_DACB_AUTODETECT_INT_CONTROL, 0);
2363                 tmp = RREG32(DC_HPD1_INT_CONTROL) & DC_HPDx_INT_POLARITY;
2364                 WREG32(DC_HPD1_INT_CONTROL, tmp);
2365                 tmp = RREG32(DC_HPD2_INT_CONTROL) & DC_HPDx_INT_POLARITY;
2366                 WREG32(DC_HPD2_INT_CONTROL, tmp);
2367                 tmp = RREG32(DC_HPD3_INT_CONTROL) & DC_HPDx_INT_POLARITY;
2368                 WREG32(DC_HPD3_INT_CONTROL, tmp);
2369                 tmp = RREG32(DC_HPD4_INT_CONTROL) & DC_HPDx_INT_POLARITY;
2370                 WREG32(DC_HPD4_INT_CONTROL, tmp);
2371                 if (ASIC_IS_DCE32(rdev)) {
2372                         tmp = RREG32(DC_HPD5_INT_CONTROL) & DC_HPDx_INT_POLARITY;
2373                         WREG32(DC_HPD5_INT_CONTROL, 0);
2374                         tmp = RREG32(DC_HPD6_INT_CONTROL) & DC_HPDx_INT_POLARITY;
2375                         WREG32(DC_HPD6_INT_CONTROL, 0);
2376                 }
2377         } else {
2378                 WREG32(DACA_AUTODETECT_INT_CONTROL, 0);
2379                 WREG32(DACB_AUTODETECT_INT_CONTROL, 0);
2380                 tmp = RREG32(DC_HOT_PLUG_DETECT1_INT_CONTROL) & DC_HOT_PLUG_DETECTx_INT_POLARITY;
2381                 WREG32(DC_HOT_PLUG_DETECT1_INT_CONTROL, 0);
2382                 tmp = RREG32(DC_HOT_PLUG_DETECT2_INT_CONTROL) & DC_HOT_PLUG_DETECTx_INT_POLARITY;
2383                 WREG32(DC_HOT_PLUG_DETECT2_INT_CONTROL, 0);
2384                 tmp = RREG32(DC_HOT_PLUG_DETECT3_INT_CONTROL) & DC_HOT_PLUG_DETECTx_INT_POLARITY;
2385                 WREG32(DC_HOT_PLUG_DETECT3_INT_CONTROL, 0);
2386         }
2387 }
2388
2389 int r600_irq_init(struct radeon_device *rdev)
2390 {
2391         int ret = 0;
2392         int rb_bufsz;
2393         u32 interrupt_cntl, ih_cntl, ih_rb_cntl;
2394
2395         /* allocate ring */
2396         ret = r600_ih_ring_alloc(rdev);
2397         if (ret)
2398                 return ret;
2399
2400         /* disable irqs */
2401         r600_disable_interrupts(rdev);
2402
2403         /* init rlc */
2404         ret = r600_rlc_init(rdev);
2405         if (ret) {
2406                 r600_ih_ring_fini(rdev);
2407                 return ret;
2408         }
2409
2410         /* setup interrupt control */
2411         /* set dummy read address to ring address */
2412         WREG32(INTERRUPT_CNTL2, rdev->ih.gpu_addr >> 8);
2413         interrupt_cntl = RREG32(INTERRUPT_CNTL);
2414         /* IH_DUMMY_RD_OVERRIDE=0 - dummy read disabled with msi, enabled without msi
2415          * IH_DUMMY_RD_OVERRIDE=1 - dummy read controlled by IH_DUMMY_RD_EN
2416          */
2417         interrupt_cntl &= ~IH_DUMMY_RD_OVERRIDE;
2418         /* IH_REQ_NONSNOOP_EN=1 if ring is in non-cacheable memory, e.g., vram */
2419         interrupt_cntl &= ~IH_REQ_NONSNOOP_EN;
2420         WREG32(INTERRUPT_CNTL, interrupt_cntl);
2421
2422         WREG32(IH_RB_BASE, rdev->ih.gpu_addr >> 8);
2423         rb_bufsz = drm_order(rdev->ih.ring_size / 4);
2424
2425         ih_rb_cntl = (IH_WPTR_OVERFLOW_ENABLE |
2426                       IH_WPTR_OVERFLOW_CLEAR |
2427                       (rb_bufsz << 1));
2428         /* WPTR writeback, not yet */
2429         /*ih_rb_cntl |= IH_WPTR_WRITEBACK_ENABLE;*/
2430         WREG32(IH_RB_WPTR_ADDR_LO, 0);
2431         WREG32(IH_RB_WPTR_ADDR_HI, 0);
2432
2433         WREG32(IH_RB_CNTL, ih_rb_cntl);
2434
2435         /* set rptr, wptr to 0 */
2436         WREG32(IH_RB_RPTR, 0);
2437         WREG32(IH_RB_WPTR, 0);
2438
2439         /* Default settings for IH_CNTL (disabled at first) */
2440         ih_cntl = MC_WRREQ_CREDIT(0x10) | MC_WR_CLEAN_CNT(0x10);
2441         /* RPTR_REARM only works if msi's are enabled */
2442         if (rdev->msi_enabled)
2443                 ih_cntl |= RPTR_REARM;
2444
2445 #ifdef __BIG_ENDIAN
2446         ih_cntl |= IH_MC_SWAP(IH_MC_SWAP_32BIT);
2447 #endif
2448         WREG32(IH_CNTL, ih_cntl);
2449
2450         /* force the active interrupt state to all disabled */
2451         r600_disable_interrupt_state(rdev);
2452
2453         /* enable irqs */
2454         r600_enable_interrupts(rdev);
2455
2456         return ret;
2457 }
2458
2459 void r600_irq_suspend(struct radeon_device *rdev)
2460 {
2461         r600_disable_interrupts(rdev);
2462         r600_rlc_stop(rdev);
2463 }
2464
2465 void r600_irq_fini(struct radeon_device *rdev)
2466 {
2467         r600_irq_suspend(rdev);
2468         r600_ih_ring_fini(rdev);
2469 }
2470
2471 int r600_irq_set(struct radeon_device *rdev)
2472 {
2473         u32 cp_int_cntl = CNTX_BUSY_INT_ENABLE | CNTX_EMPTY_INT_ENABLE;
2474         u32 mode_int = 0;
2475         u32 hpd1, hpd2, hpd3, hpd4 = 0, hpd5 = 0, hpd6 = 0;
2476
2477         if (!rdev->irq.installed) {
2478                 WARN(1, "Can't enable IRQ/MSI because no handler is installed.\n");
2479                 return -EINVAL;
2480         }
2481         /* don't enable anything if the ih is disabled */
2482         if (!rdev->ih.enabled) {
2483                 r600_disable_interrupts(rdev);
2484                 /* force the active interrupt state to all disabled */
2485                 r600_disable_interrupt_state(rdev);
2486                 return 0;
2487         }
2488
2489         if (ASIC_IS_DCE3(rdev)) {
2490                 hpd1 = RREG32(DC_HPD1_INT_CONTROL) & ~DC_HPDx_INT_EN;
2491                 hpd2 = RREG32(DC_HPD2_INT_CONTROL) & ~DC_HPDx_INT_EN;
2492                 hpd3 = RREG32(DC_HPD3_INT_CONTROL) & ~DC_HPDx_INT_EN;
2493                 hpd4 = RREG32(DC_HPD4_INT_CONTROL) & ~DC_HPDx_INT_EN;
2494                 if (ASIC_IS_DCE32(rdev)) {
2495                         hpd5 = RREG32(DC_HPD5_INT_CONTROL) & ~DC_HPDx_INT_EN;
2496                         hpd6 = RREG32(DC_HPD6_INT_CONTROL) & ~DC_HPDx_INT_EN;
2497                 }
2498         } else {
2499                 hpd1 = RREG32(DC_HOT_PLUG_DETECT1_INT_CONTROL) & ~DC_HPDx_INT_EN;
2500                 hpd2 = RREG32(DC_HOT_PLUG_DETECT2_INT_CONTROL) & ~DC_HPDx_INT_EN;
2501                 hpd3 = RREG32(DC_HOT_PLUG_DETECT3_INT_CONTROL) & ~DC_HPDx_INT_EN;
2502         }
2503
2504         if (rdev->irq.sw_int) {
2505                 DRM_DEBUG("r600_irq_set: sw int\n");
2506                 cp_int_cntl |= RB_INT_ENABLE;
2507         }
2508         if (rdev->irq.crtc_vblank_int[0]) {
2509                 DRM_DEBUG("r600_irq_set: vblank 0\n");
2510                 mode_int |= D1MODE_VBLANK_INT_MASK;
2511         }
2512         if (rdev->irq.crtc_vblank_int[1]) {
2513                 DRM_DEBUG("r600_irq_set: vblank 1\n");
2514                 mode_int |= D2MODE_VBLANK_INT_MASK;
2515         }
2516         if (rdev->irq.hpd[0]) {
2517                 DRM_DEBUG("r600_irq_set: hpd 1\n");
2518                 hpd1 |= DC_HPDx_INT_EN;
2519         }
2520         if (rdev->irq.hpd[1]) {
2521                 DRM_DEBUG("r600_irq_set: hpd 2\n");
2522                 hpd2 |= DC_HPDx_INT_EN;
2523         }
2524         if (rdev->irq.hpd[2]) {
2525                 DRM_DEBUG("r600_irq_set: hpd 3\n");
2526                 hpd3 |= DC_HPDx_INT_EN;
2527         }
2528         if (rdev->irq.hpd[3]) {
2529                 DRM_DEBUG("r600_irq_set: hpd 4\n");
2530                 hpd4 |= DC_HPDx_INT_EN;
2531         }
2532         if (rdev->irq.hpd[4]) {
2533                 DRM_DEBUG("r600_irq_set: hpd 5\n");
2534                 hpd5 |= DC_HPDx_INT_EN;
2535         }
2536         if (rdev->irq.hpd[5]) {
2537                 DRM_DEBUG("r600_irq_set: hpd 6\n");
2538                 hpd6 |= DC_HPDx_INT_EN;
2539         }
2540
2541         WREG32(CP_INT_CNTL, cp_int_cntl);
2542         WREG32(DxMODE_INT_MASK, mode_int);
2543         if (ASIC_IS_DCE3(rdev)) {
2544                 WREG32(DC_HPD1_INT_CONTROL, hpd1);
2545                 WREG32(DC_HPD2_INT_CONTROL, hpd2);
2546                 WREG32(DC_HPD3_INT_CONTROL, hpd3);
2547                 WREG32(DC_HPD4_INT_CONTROL, hpd4);
2548                 if (ASIC_IS_DCE32(rdev)) {
2549                         WREG32(DC_HPD5_INT_CONTROL, hpd5);
2550                         WREG32(DC_HPD6_INT_CONTROL, hpd6);
2551                 }
2552         } else {
2553                 WREG32(DC_HOT_PLUG_DETECT1_INT_CONTROL, hpd1);
2554                 WREG32(DC_HOT_PLUG_DETECT2_INT_CONTROL, hpd2);
2555                 WREG32(DC_HOT_PLUG_DETECT3_INT_CONTROL, hpd3);
2556         }
2557
2558         return 0;
2559 }
2560
2561 static inline void r600_irq_ack(struct radeon_device *rdev,
2562                                 u32 *disp_int,
2563                                 u32 *disp_int_cont,
2564                                 u32 *disp_int_cont2)
2565 {
2566         u32 tmp;
2567
2568         if (ASIC_IS_DCE3(rdev)) {
2569                 *disp_int = RREG32(DCE3_DISP_INTERRUPT_STATUS);
2570                 *disp_int_cont = RREG32(DCE3_DISP_INTERRUPT_STATUS_CONTINUE);
2571                 *disp_int_cont2 = RREG32(DCE3_DISP_INTERRUPT_STATUS_CONTINUE2);
2572         } else {
2573                 *disp_int = RREG32(DISP_INTERRUPT_STATUS);
2574                 *disp_int_cont = RREG32(DISP_INTERRUPT_STATUS_CONTINUE);
2575                 *disp_int_cont2 = 0;
2576         }
2577
2578         if (*disp_int & LB_D1_VBLANK_INTERRUPT)
2579                 WREG32(D1MODE_VBLANK_STATUS, DxMODE_VBLANK_ACK);
2580         if (*disp_int & LB_D1_VLINE_INTERRUPT)
2581                 WREG32(D1MODE_VLINE_STATUS, DxMODE_VLINE_ACK);
2582         if (*disp_int & LB_D2_VBLANK_INTERRUPT)
2583                 WREG32(D2MODE_VBLANK_STATUS, DxMODE_VBLANK_ACK);
2584         if (*disp_int & LB_D2_VLINE_INTERRUPT)
2585                 WREG32(D2MODE_VLINE_STATUS, DxMODE_VLINE_ACK);
2586         if (*disp_int & DC_HPD1_INTERRUPT) {
2587                 if (ASIC_IS_DCE3(rdev)) {
2588                         tmp = RREG32(DC_HPD1_INT_CONTROL);
2589                         tmp |= DC_HPDx_INT_ACK;
2590                         WREG32(DC_HPD1_INT_CONTROL, tmp);
2591                 } else {
2592                         tmp = RREG32(DC_HOT_PLUG_DETECT1_INT_CONTROL);
2593                         tmp |= DC_HPDx_INT_ACK;
2594                         WREG32(DC_HOT_PLUG_DETECT1_INT_CONTROL, tmp);
2595                 }
2596         }
2597         if (*disp_int & DC_HPD2_INTERRUPT) {
2598                 if (ASIC_IS_DCE3(rdev)) {
2599                         tmp = RREG32(DC_HPD2_INT_CONTROL);
2600                         tmp |= DC_HPDx_INT_ACK;
2601                         WREG32(DC_HPD2_INT_CONTROL, tmp);
2602                 } else {
2603                         tmp = RREG32(DC_HOT_PLUG_DETECT2_INT_CONTROL);
2604                         tmp |= DC_HPDx_INT_ACK;
2605                         WREG32(DC_HOT_PLUG_DETECT2_INT_CONTROL, tmp);
2606                 }
2607         }
2608         if (*disp_int_cont & DC_HPD3_INTERRUPT) {
2609                 if (ASIC_IS_DCE3(rdev)) {
2610                         tmp = RREG32(DC_HPD3_INT_CONTROL);
2611                         tmp |= DC_HPDx_INT_ACK;
2612                         WREG32(DC_HPD3_INT_CONTROL, tmp);
2613                 } else {
2614                         tmp = RREG32(DC_HOT_PLUG_DETECT3_INT_CONTROL);
2615                         tmp |= DC_HPDx_INT_ACK;
2616                         WREG32(DC_HOT_PLUG_DETECT3_INT_CONTROL, tmp);
2617                 }
2618         }
2619         if (*disp_int_cont & DC_HPD4_INTERRUPT) {
2620                 tmp = RREG32(DC_HPD4_INT_CONTROL);
2621                 tmp |= DC_HPDx_INT_ACK;
2622                 WREG32(DC_HPD4_INT_CONTROL, tmp);
2623         }
2624         if (ASIC_IS_DCE32(rdev)) {
2625                 if (*disp_int_cont2 & DC_HPD5_INTERRUPT) {
2626                         tmp = RREG32(DC_HPD5_INT_CONTROL);
2627                         tmp |= DC_HPDx_INT_ACK;
2628                         WREG32(DC_HPD5_INT_CONTROL, tmp);
2629                 }
2630                 if (*disp_int_cont2 & DC_HPD6_INTERRUPT) {
2631                         tmp = RREG32(DC_HPD5_INT_CONTROL);
2632                         tmp |= DC_HPDx_INT_ACK;
2633                         WREG32(DC_HPD6_INT_CONTROL, tmp);
2634                 }
2635         }
2636 }
2637
2638 void r600_irq_disable(struct radeon_device *rdev)
2639 {
2640         u32 disp_int, disp_int_cont, disp_int_cont2;
2641
2642         r600_disable_interrupts(rdev);
2643         /* Wait and acknowledge irq */
2644         mdelay(1);
2645         r600_irq_ack(rdev, &disp_int, &disp_int_cont, &disp_int_cont2);
2646         r600_disable_interrupt_state(rdev);
2647 }
2648
2649 static inline u32 r600_get_ih_wptr(struct radeon_device *rdev)
2650 {
2651         u32 wptr, tmp;
2652
2653         /* XXX use writeback */
2654         wptr = RREG32(IH_RB_WPTR);
2655
2656         if (wptr & RB_OVERFLOW) {
2657                 /* When a ring buffer overflow happen start parsing interrupt
2658                  * from the last not overwritten vector (wptr + 16). Hopefully
2659                  * this should allow us to catchup.
2660                  */
2661                 dev_warn(rdev->dev, "IH ring buffer overflow (0x%08X, %d, %d)\n",
2662                         wptr, rdev->ih.rptr, (wptr + 16) + rdev->ih.ptr_mask);
2663                 rdev->ih.rptr = (wptr + 16) & rdev->ih.ptr_mask;
2664                 tmp = RREG32(IH_RB_CNTL);
2665                 tmp |= IH_WPTR_OVERFLOW_CLEAR;
2666                 WREG32(IH_RB_CNTL, tmp);
2667         }
2668         return (wptr & rdev->ih.ptr_mask);
2669 }
2670
2671 /*        r600 IV Ring
2672  * Each IV ring entry is 128 bits:
2673  * [7:0]    - interrupt source id
2674  * [31:8]   - reserved
2675  * [59:32]  - interrupt source data
2676  * [127:60]  - reserved
2677  *
2678  * The basic interrupt vector entries
2679  * are decoded as follows:
2680  * src_id  src_data  description
2681  *      1         0  D1 Vblank
2682  *      1         1  D1 Vline
2683  *      5         0  D2 Vblank
2684  *      5         1  D2 Vline
2685  *     19         0  FP Hot plug detection A
2686  *     19         1  FP Hot plug detection B
2687  *     19         2  DAC A auto-detection
2688  *     19         3  DAC B auto-detection
2689  *    176         -  CP_INT RB
2690  *    177         -  CP_INT IB1
2691  *    178         -  CP_INT IB2
2692  *    181         -  EOP Interrupt
2693  *    233         -  GUI Idle
2694  *
2695  * Note, these are based on r600 and may need to be
2696  * adjusted or added to on newer asics
2697  */
2698
2699 int r600_irq_process(struct radeon_device *rdev)
2700 {
2701         u32 wptr = r600_get_ih_wptr(rdev);
2702         u32 rptr = rdev->ih.rptr;
2703         u32 src_id, src_data;
2704         u32 ring_index, disp_int, disp_int_cont, disp_int_cont2;
2705         unsigned long flags;
2706         bool queue_hotplug = false;
2707
2708         DRM_DEBUG("r600_irq_process start: rptr %d, wptr %d\n", rptr, wptr);
2709         if (!rdev->ih.enabled)
2710                 return IRQ_NONE;
2711
2712         spin_lock_irqsave(&rdev->ih.lock, flags);
2713
2714         if (rptr == wptr) {
2715                 spin_unlock_irqrestore(&rdev->ih.lock, flags);
2716                 return IRQ_NONE;
2717         }
2718         if (rdev->shutdown) {
2719                 spin_unlock_irqrestore(&rdev->ih.lock, flags);
2720                 return IRQ_NONE;
2721         }
2722
2723 restart_ih:
2724         /* display interrupts */
2725         r600_irq_ack(rdev, &disp_int, &disp_int_cont, &disp_int_cont2);
2726
2727         rdev->ih.wptr = wptr;
2728         while (rptr != wptr) {
2729                 /* wptr/rptr are in bytes! */
2730                 ring_index = rptr / 4;
2731                 src_id =  rdev->ih.ring[ring_index] & 0xff;
2732                 src_data = rdev->ih.ring[ring_index + 1] & 0xfffffff;
2733
2734                 switch (src_id) {
2735                 case 1: /* D1 vblank/vline */
2736                         switch (src_data) {
2737                         case 0: /* D1 vblank */
2738                                 if (disp_int & LB_D1_VBLANK_INTERRUPT) {
2739                                         drm_handle_vblank(rdev->ddev, 0);
2740                                         disp_int &= ~LB_D1_VBLANK_INTERRUPT;
2741                                         DRM_DEBUG("IH: D1 vblank\n");
2742                                 }
2743                                 break;
2744                         case 1: /* D1 vline */
2745                                 if (disp_int & LB_D1_VLINE_INTERRUPT) {
2746                                         disp_int &= ~LB_D1_VLINE_INTERRUPT;
2747                                         DRM_DEBUG("IH: D1 vline\n");
2748                                 }
2749                                 break;
2750                         default:
2751                                 DRM_DEBUG("Unhandled interrupt: %d %d\n", src_id, src_data);
2752                                 break;
2753                         }
2754                         break;
2755                 case 5: /* D2 vblank/vline */
2756                         switch (src_data) {
2757                         case 0: /* D2 vblank */
2758                                 if (disp_int & LB_D2_VBLANK_INTERRUPT) {
2759                                         drm_handle_vblank(rdev->ddev, 1);
2760                                         disp_int &= ~LB_D2_VBLANK_INTERRUPT;
2761                                         DRM_DEBUG("IH: D2 vblank\n");
2762                                 }
2763                                 break;
2764                         case 1: /* D1 vline */
2765                                 if (disp_int & LB_D2_VLINE_INTERRUPT) {
2766                                         disp_int &= ~LB_D2_VLINE_INTERRUPT;
2767                                         DRM_DEBUG("IH: D2 vline\n");
2768                                 }
2769                                 break;
2770                         default:
2771                                 DRM_DEBUG("Unhandled interrupt: %d %d\n", src_id, src_data);
2772                                 break;
2773                         }
2774                         break;
2775                 case 19: /* HPD/DAC hotplug */
2776                         switch (src_data) {
2777                         case 0:
2778                                 if (disp_int & DC_HPD1_INTERRUPT) {
2779                                         disp_int &= ~DC_HPD1_INTERRUPT;
2780                                         queue_hotplug = true;
2781                                         DRM_DEBUG("IH: HPD1\n");
2782                                 }
2783                                 break;
2784                         case 1:
2785                                 if (disp_int & DC_HPD2_INTERRUPT) {
2786                                         disp_int &= ~DC_HPD2_INTERRUPT;
2787                                         queue_hotplug = true;
2788                                         DRM_DEBUG("IH: HPD2\n");
2789                                 }
2790                                 break;
2791                         case 4:
2792                                 if (disp_int_cont & DC_HPD3_INTERRUPT) {
2793                                         disp_int_cont &= ~DC_HPD3_INTERRUPT;
2794                                         queue_hotplug = true;
2795                                         DRM_DEBUG("IH: HPD3\n");
2796                                 }
2797                                 break;
2798                         case 5:
2799                                 if (disp_int_cont & DC_HPD4_INTERRUPT) {
2800                                         disp_int_cont &= ~DC_HPD4_INTERRUPT;
2801                                         queue_hotplug = true;
2802                                         DRM_DEBUG("IH: HPD4\n");
2803                                 }
2804                                 break;
2805                         case 10:
2806                                 if (disp_int_cont2 & DC_HPD5_INTERRUPT) {
2807                                         disp_int_cont &= ~DC_HPD5_INTERRUPT;
2808                                         queue_hotplug = true;
2809                                         DRM_DEBUG("IH: HPD5\n");
2810                                 }
2811                                 break;
2812                         case 12:
2813                                 if (disp_int_cont2 & DC_HPD6_INTERRUPT) {
2814                                         disp_int_cont &= ~DC_HPD6_INTERRUPT;
2815                                         queue_hotplug = true;
2816                                         DRM_DEBUG("IH: HPD6\n");
2817                                 }
2818                                 break;
2819                         default:
2820                                 DRM_DEBUG("Unhandled interrupt: %d %d\n", src_id, src_data);
2821                                 break;
2822                         }
2823                         break;
2824                 case 176: /* CP_INT in ring buffer */
2825                 case 177: /* CP_INT in IB1 */
2826                 case 178: /* CP_INT in IB2 */
2827                         DRM_DEBUG("IH: CP int: 0x%08x\n", src_data);
2828                         radeon_fence_process(rdev);
2829                         break;
2830                 case 181: /* CP EOP event */
2831                         DRM_DEBUG("IH: CP EOP\n");
2832                         break;
2833                 default:
2834                         DRM_DEBUG("Unhandled interrupt: %d %d\n", src_id, src_data);
2835                         break;
2836                 }
2837
2838                 /* wptr/rptr are in bytes! */
2839                 rptr += 16;
2840                 rptr &= rdev->ih.ptr_mask;
2841         }
2842         /* make sure wptr hasn't changed while processing */
2843         wptr = r600_get_ih_wptr(rdev);
2844         if (wptr != rdev->ih.wptr)
2845                 goto restart_ih;
2846         if (queue_hotplug)
2847                 queue_work(rdev->wq, &rdev->hotplug_work);
2848         rdev->ih.rptr = rptr;
2849         WREG32(IH_RB_RPTR, rdev->ih.rptr);
2850         spin_unlock_irqrestore(&rdev->ih.lock, flags);
2851         return IRQ_HANDLED;
2852 }
2853
2854 /*
2855  * Debugfs info
2856  */
2857 #if defined(CONFIG_DEBUG_FS)
2858
2859 static int r600_debugfs_cp_ring_info(struct seq_file *m, void *data)
2860 {
2861         struct drm_info_node *node = (struct drm_info_node *) m->private;
2862         struct drm_device *dev = node->minor->dev;
2863         struct radeon_device *rdev = dev->dev_private;
2864         unsigned count, i, j;
2865
2866         radeon_ring_free_size(rdev);
2867         count = (rdev->cp.ring_size / 4) - rdev->cp.ring_free_dw;
2868         seq_printf(m, "CP_STAT 0x%08x\n", RREG32(CP_STAT));
2869         seq_printf(m, "CP_RB_WPTR 0x%08x\n", RREG32(CP_RB_WPTR));
2870         seq_printf(m, "CP_RB_RPTR 0x%08x\n", RREG32(CP_RB_RPTR));
2871         seq_printf(m, "driver's copy of the CP_RB_WPTR 0x%08x\n", rdev->cp.wptr);
2872         seq_printf(m, "driver's copy of the CP_RB_RPTR 0x%08x\n", rdev->cp.rptr);
2873         seq_printf(m, "%u free dwords in ring\n", rdev->cp.ring_free_dw);
2874         seq_printf(m, "%u dwords in ring\n", count);
2875         i = rdev->cp.rptr;
2876         for (j = 0; j <= count; j++) {
2877                 seq_printf(m, "r[%04d]=0x%08x\n", i, rdev->cp.ring[i]);
2878                 i = (i + 1) & rdev->cp.ptr_mask;
2879         }
2880         return 0;
2881 }
2882
2883 static int r600_debugfs_mc_info(struct seq_file *m, void *data)
2884 {
2885         struct drm_info_node *node = (struct drm_info_node *) m->private;
2886         struct drm_device *dev = node->minor->dev;
2887         struct radeon_device *rdev = dev->dev_private;
2888
2889         DREG32_SYS(m, rdev, R_000E50_SRBM_STATUS);
2890         DREG32_SYS(m, rdev, VM_L2_STATUS);
2891         return 0;
2892 }
2893
2894 static struct drm_info_list r600_mc_info_list[] = {
2895         {"r600_mc_info", r600_debugfs_mc_info, 0, NULL},
2896         {"r600_ring_info", r600_debugfs_cp_ring_info, 0, NULL},
2897 };
2898 #endif
2899
2900 int r600_debugfs_mc_info_init(struct radeon_device *rdev)
2901 {
2902 #if defined(CONFIG_DEBUG_FS)
2903         return radeon_debugfs_add_files(rdev, r600_mc_info_list, ARRAY_SIZE(r600_mc_info_list));
2904 #else
2905         return 0;
2906 #endif
2907 }
2908
2909 /**
2910  * r600_ioctl_wait_idle - flush host path cache on wait idle ioctl
2911  * rdev: radeon device structure
2912  * bo: buffer object struct which userspace is waiting for idle
2913  *
2914  * Some R6XX/R7XX doesn't seems to take into account HDP flush performed
2915  * through ring buffer, this leads to corruption in rendering, see
2916  * http://bugzilla.kernel.org/show_bug.cgi?id=15186 to avoid this we
2917  * directly perform HDP flush by writing register through MMIO.
2918  */
2919 void r600_ioctl_wait_idle(struct radeon_device *rdev, struct radeon_bo *bo)
2920 {
2921         WREG32(R_005480_HDP_MEM_COHERENCY_FLUSH_CNTL, 0x1);
2922 }