SAFE public projects git trees. - safe/jmp/linux-2.6/blob - drivers/gpu/drm/radeon/r600.c

   1 /*
   2  * Copyright 2008 Advanced Micro Devices, Inc.
   3  * Copyright 2008 Red Hat Inc.
   4  * Copyright 2009 Jerome Glisse.
   5  *
   6  * Permission is hereby granted, free of charge, to any person obtaining a
   7  * copy of this software and associated documentation files (the "Software"),
   8  * to deal in the Software without restriction, including without limitation
   9  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
  10  * and/or sell copies of the Software, and to permit persons to whom the
  11  * Software is furnished to do so, subject to the following conditions:
  12  *
  13  * The above copyright notice and this permission notice shall be included in
  14  * all copies or substantial portions of the Software.
  15  *
  16  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  17  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  18  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
  19  * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
  20  * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
  21  * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
  22  * OTHER DEALINGS IN THE SOFTWARE.
  23  *
  24  * Authors: Dave Airlie
  25  *          Alex Deucher
  26  *          Jerome Glisse
  27  */
  28 #include <linux/seq_file.h>
  29 #include <linux/firmware.h>
  30 #include <linux/platform_device.h>
  31 #include "drmP.h"
  32 #include "radeon_drm.h"
  33 #include "radeon.h"
  34 #include "radeon_mode.h"
  35 #include "r600d.h"
  36 #include "atom.h"
  37 #include "avivod.h"
  38
  39 #define PFP_UCODE_SIZE 576
  40 #define PM4_UCODE_SIZE 1792
  41 #define RLC_UCODE_SIZE 768
  42 #define R700_PFP_UCODE_SIZE 848
  43 #define R700_PM4_UCODE_SIZE 1360
  44 #define R700_RLC_UCODE_SIZE 1024
  45
  46 /* Firmware Names */
  47 MODULE_FIRMWARE("radeon/R600_pfp.bin");
  48 MODULE_FIRMWARE("radeon/R600_me.bin");
  49 MODULE_FIRMWARE("radeon/RV610_pfp.bin");
  50 MODULE_FIRMWARE("radeon/RV610_me.bin");
  51 MODULE_FIRMWARE("radeon/RV630_pfp.bin");
  52 MODULE_FIRMWARE("radeon/RV630_me.bin");
  53 MODULE_FIRMWARE("radeon/RV620_pfp.bin");
  54 MODULE_FIRMWARE("radeon/RV620_me.bin");
  55 MODULE_FIRMWARE("radeon/RV635_pfp.bin");
  56 MODULE_FIRMWARE("radeon/RV635_me.bin");
  57 MODULE_FIRMWARE("radeon/RV670_pfp.bin");
  58 MODULE_FIRMWARE("radeon/RV670_me.bin");
  59 MODULE_FIRMWARE("radeon/RS780_pfp.bin");
  60 MODULE_FIRMWARE("radeon/RS780_me.bin");
  61 MODULE_FIRMWARE("radeon/RV770_pfp.bin");
  62 MODULE_FIRMWARE("radeon/RV770_me.bin");
  63 MODULE_FIRMWARE("radeon/RV730_pfp.bin");
  64 MODULE_FIRMWARE("radeon/RV730_me.bin");
  65 MODULE_FIRMWARE("radeon/RV710_pfp.bin");
  66 MODULE_FIRMWARE("radeon/RV710_me.bin");
  67 MODULE_FIRMWARE("radeon/R600_rlc.bin");
  68 MODULE_FIRMWARE("radeon/R700_rlc.bin");
  69
  70 int r600_debugfs_mc_info_init(struct radeon_device *rdev);
  71
  72 /* r600,rv610,rv630,rv620,rv635,rv670 */
  73 int r600_mc_wait_for_idle(struct radeon_device *rdev);
  74 void r600_gpu_init(struct radeon_device *rdev);
  75 void r600_fini(struct radeon_device *rdev);
  76
  77 /*
  78  * R600 PCIE GART
  79  */
  80 int r600_gart_clear_page(struct radeon_device *rdev, int i)
  81 {
  82         void __iomem *ptr = (void *)rdev->gart.table.vram.ptr;
  83         u64 pte;
  84
  85         if (i < 0 || i > rdev->gart.num_gpu_pages)
  86                 return -EINVAL;
  87         pte = 0;
  88         writeq(pte, ((void __iomem *)ptr) + (i * 8));
  89         return 0;
  90 }
  91
  92 void r600_pcie_gart_tlb_flush(struct radeon_device *rdev)
  93 {
  94         unsigned i;
  95         u32 tmp;
  96
  97         WREG32(VM_CONTEXT0_INVALIDATION_LOW_ADDR, rdev->mc.gtt_start >> 12);
  98         WREG32(VM_CONTEXT0_INVALIDATION_HIGH_ADDR, (rdev->mc.gtt_end - 1) >> 12);
  99         WREG32(VM_CONTEXT0_REQUEST_RESPONSE, REQUEST_TYPE(1));
 100         for (i = 0; i < rdev->usec_timeout; i++) {
 101                 /* read MC_STATUS */
 102                 tmp = RREG32(VM_CONTEXT0_REQUEST_RESPONSE);
 103                 tmp = (tmp & RESPONSE_TYPE_MASK) >> RESPONSE_TYPE_SHIFT;
 104                 if (tmp == 2) {
 105                         printk(KERN_WARNING "[drm] r600 flush TLB failed\n");
 106                         return;
 107                 }
 108                 if (tmp) {
 109                         return;
 110                 }
 111                 udelay(1);
 112         }
 113 }
 114
 115 int r600_pcie_gart_init(struct radeon_device *rdev)
 116 {
 117         int r;
 118
 119         if (rdev->gart.table.vram.robj) {
 120                 WARN(1, "R600 PCIE GART already initialized.\n");
 121                 return 0;
 122         }
 123         /* Initialize common gart structure */
 124         r = radeon_gart_init(rdev);
 125         if (r)
 126                 return r;
 127         rdev->gart.table_size = rdev->gart.num_gpu_pages * 8;
 128         return radeon_gart_table_vram_alloc(rdev);
 129 }
 130
 131 int r600_pcie_gart_enable(struct radeon_device *rdev)
 132 {
 133         u32 tmp;
 134         int r, i;
 135
 136         if (rdev->gart.table.vram.robj == NULL) {
 137                 dev_err(rdev->dev, "No VRAM object for PCIE GART.\n");
 138                 return -EINVAL;
 139         }
 140         r = radeon_gart_table_vram_pin(rdev);
 141         if (r)
 142                 return r;
 143
 144         /* Setup L2 cache */
 145         WREG32(VM_L2_CNTL, ENABLE_L2_CACHE | ENABLE_L2_FRAGMENT_PROCESSING |
 146                                 ENABLE_L2_PTE_CACHE_LRU_UPDATE_BY_WRITE |
 147                                 EFFECTIVE_L2_QUEUE_SIZE(7));
 148         WREG32(VM_L2_CNTL2, 0);
 149         WREG32(VM_L2_CNTL3, BANK_SELECT_0(0) | BANK_SELECT_1(1));
 150         /* Setup TLB control */
 151         tmp = ENABLE_L1_TLB | ENABLE_L1_FRAGMENT_PROCESSING |
 152                 SYSTEM_ACCESS_MODE_NOT_IN_SYS |
 153                 EFFECTIVE_L1_TLB_SIZE(5) | EFFECTIVE_L1_QUEUE_SIZE(5) |
 154                 ENABLE_WAIT_L2_QUERY;
 155         WREG32(MC_VM_L1_TLB_MCB_RD_SYS_CNTL, tmp);
 156         WREG32(MC_VM_L1_TLB_MCB_WR_SYS_CNTL, tmp);
 157         WREG32(MC_VM_L1_TLB_MCB_RD_HDP_CNTL, tmp | ENABLE_L1_STRICT_ORDERING);
 158         WREG32(MC_VM_L1_TLB_MCB_WR_HDP_CNTL, tmp);
 159         WREG32(MC_VM_L1_TLB_MCD_RD_A_CNTL, tmp);
 160         WREG32(MC_VM_L1_TLB_MCD_WR_A_CNTL, tmp);
 161         WREG32(MC_VM_L1_TLB_MCD_RD_B_CNTL, tmp);
 162         WREG32(MC_VM_L1_TLB_MCD_WR_B_CNTL, tmp);
 163         WREG32(MC_VM_L1_TLB_MCB_RD_GFX_CNTL, tmp);
 164         WREG32(MC_VM_L1_TLB_MCB_WR_GFX_CNTL, tmp);
 165         WREG32(MC_VM_L1_TLB_MCB_RD_PDMA_CNTL, tmp);
 166         WREG32(MC_VM_L1_TLB_MCB_WR_PDMA_CNTL, tmp);
 167         WREG32(MC_VM_L1_TLB_MCB_RD_SEM_CNTL, tmp | ENABLE_SEMAPHORE_MODE);
 168         WREG32(MC_VM_L1_TLB_MCB_WR_SEM_CNTL, tmp | ENABLE_SEMAPHORE_MODE);
 169         WREG32(VM_CONTEXT0_PAGE_TABLE_START_ADDR, rdev->mc.gtt_start >> 12);
 170         WREG32(VM_CONTEXT0_PAGE_TABLE_END_ADDR, rdev->mc.gtt_end >> 12);
 171         WREG32(VM_CONTEXT0_PAGE_TABLE_BASE_ADDR, rdev->gart.table_addr >> 12);
 172         WREG32(VM_CONTEXT0_CNTL, ENABLE_CONTEXT | PAGE_TABLE_DEPTH(0) |
 173                                 RANGE_PROTECTION_FAULT_ENABLE_DEFAULT);
 174         WREG32(VM_CONTEXT0_PROTECTION_FAULT_DEFAULT_ADDR,
 175                         (u32)(rdev->dummy_page.addr >> 12));
 176         for (i = 1; i < 7; i++)
 177                 WREG32(VM_CONTEXT0_CNTL + (i * 4), 0);
 178
 179         r600_pcie_gart_tlb_flush(rdev);
 180         rdev->gart.ready = true;
 181         return 0;
 182 }
 183
 184 void r600_pcie_gart_disable(struct radeon_device *rdev)
 185 {
 186         u32 tmp;
 187         int i;
 188
 189         /* Disable all tables */
 190         for (i = 0; i < 7; i++)
 191                 WREG32(VM_CONTEXT0_CNTL + (i * 4), 0);
 192
 193         /* Disable L2 cache */
 194         WREG32(VM_L2_CNTL, ENABLE_L2_FRAGMENT_PROCESSING |
 195                                 EFFECTIVE_L2_QUEUE_SIZE(7));
 196         WREG32(VM_L2_CNTL3, BANK_SELECT_0(0) | BANK_SELECT_1(1));
 197         /* Setup L1 TLB control */
 198         tmp = EFFECTIVE_L1_TLB_SIZE(5) | EFFECTIVE_L1_QUEUE_SIZE(5) |
 199                 ENABLE_WAIT_L2_QUERY;
 200         WREG32(MC_VM_L1_TLB_MCD_RD_A_CNTL, tmp);
 201         WREG32(MC_VM_L1_TLB_MCD_WR_A_CNTL, tmp);
 202         WREG32(MC_VM_L1_TLB_MCD_RD_B_CNTL, tmp);
 203         WREG32(MC_VM_L1_TLB_MCD_WR_B_CNTL, tmp);
 204         WREG32(MC_VM_L1_TLB_MCB_RD_GFX_CNTL, tmp);
 205         WREG32(MC_VM_L1_TLB_MCB_WR_GFX_CNTL, tmp);
 206         WREG32(MC_VM_L1_TLB_MCB_RD_PDMA_CNTL, tmp);
 207         WREG32(MC_VM_L1_TLB_MCB_WR_PDMA_CNTL, tmp);
 208         WREG32(MC_VM_L1_TLB_MCB_RD_SEM_CNTL, tmp);
 209         WREG32(MC_VM_L1_TLB_MCB_WR_SEM_CNTL, tmp);
 210         WREG32(MC_VM_L1_TLB_MCB_RD_SYS_CNTL, tmp);
 211         WREG32(MC_VM_L1_TLB_MCB_WR_SYS_CNTL, tmp);
 212         WREG32(MC_VM_L1_TLB_MCB_RD_HDP_CNTL, tmp);
 213         WREG32(MC_VM_L1_TLB_MCB_WR_HDP_CNTL, tmp);
 214         if (rdev->gart.table.vram.robj) {
 215                 radeon_object_kunmap(rdev->gart.table.vram.robj);
 216                 radeon_object_unpin(rdev->gart.table.vram.robj);
 217         }
 218 }
 219
 220 void r600_pcie_gart_fini(struct radeon_device *rdev)
 221 {
 222         r600_pcie_gart_disable(rdev);
 223         radeon_gart_table_vram_free(rdev);
 224         radeon_gart_fini(rdev);
 225 }
 226
 227 void r600_agp_enable(struct radeon_device *rdev)
 228 {
 229         u32 tmp;
 230         int i;
 231
 232         /* Setup L2 cache */
 233         WREG32(VM_L2_CNTL, ENABLE_L2_CACHE | ENABLE_L2_FRAGMENT_PROCESSING |
 234                                 ENABLE_L2_PTE_CACHE_LRU_UPDATE_BY_WRITE |
 235                                 EFFECTIVE_L2_QUEUE_SIZE(7));
 236         WREG32(VM_L2_CNTL2, 0);
 237         WREG32(VM_L2_CNTL3, BANK_SELECT_0(0) | BANK_SELECT_1(1));
 238         /* Setup TLB control */
 239         tmp = ENABLE_L1_TLB | ENABLE_L1_FRAGMENT_PROCESSING |
 240                 SYSTEM_ACCESS_MODE_NOT_IN_SYS |
 241                 EFFECTIVE_L1_TLB_SIZE(5) | EFFECTIVE_L1_QUEUE_SIZE(5) |
 242                 ENABLE_WAIT_L2_QUERY;
 243         WREG32(MC_VM_L1_TLB_MCB_RD_SYS_CNTL, tmp);
 244         WREG32(MC_VM_L1_TLB_MCB_WR_SYS_CNTL, tmp);
 245         WREG32(MC_VM_L1_TLB_MCB_RD_HDP_CNTL, tmp | ENABLE_L1_STRICT_ORDERING);
 246         WREG32(MC_VM_L1_TLB_MCB_WR_HDP_CNTL, tmp);
 247         WREG32(MC_VM_L1_TLB_MCD_RD_A_CNTL, tmp);
 248         WREG32(MC_VM_L1_TLB_MCD_WR_A_CNTL, tmp);
 249         WREG32(MC_VM_L1_TLB_MCD_RD_B_CNTL, tmp);
 250         WREG32(MC_VM_L1_TLB_MCD_WR_B_CNTL, tmp);
 251         WREG32(MC_VM_L1_TLB_MCB_RD_GFX_CNTL, tmp);
 252         WREG32(MC_VM_L1_TLB_MCB_WR_GFX_CNTL, tmp);
 253         WREG32(MC_VM_L1_TLB_MCB_RD_PDMA_CNTL, tmp);
 254         WREG32(MC_VM_L1_TLB_MCB_WR_PDMA_CNTL, tmp);
 255         WREG32(MC_VM_L1_TLB_MCB_RD_SEM_CNTL, tmp | ENABLE_SEMAPHORE_MODE);
 256         WREG32(MC_VM_L1_TLB_MCB_WR_SEM_CNTL, tmp | ENABLE_SEMAPHORE_MODE);
 257         for (i = 0; i < 7; i++)
 258                 WREG32(VM_CONTEXT0_CNTL + (i * 4), 0);
 259 }
 260
 261 int r600_mc_wait_for_idle(struct radeon_device *rdev)
 262 {
 263         unsigned i;
 264         u32 tmp;
 265
 266         for (i = 0; i < rdev->usec_timeout; i++) {
 267                 /* read MC_STATUS */
 268                 tmp = RREG32(R_000E50_SRBM_STATUS) & 0x3F00;
 269                 if (!tmp)
 270                         return 0;
 271                 udelay(1);
 272         }
 273         return -1;
 274 }
 275
 276 static void r600_mc_program(struct radeon_device *rdev)
 277 {
 278         struct rv515_mc_save save;
 279         u32 tmp;
 280         int i, j;
 281
 282         /* Initialize HDP */
 283         for (i = 0, j = 0; i < 32; i++, j += 0x18) {
 284                 WREG32((0x2c14 + j), 0x00000000);
 285                 WREG32((0x2c18 + j), 0x00000000);
 286                 WREG32((0x2c1c + j), 0x00000000);
 287                 WREG32((0x2c20 + j), 0x00000000);
 288                 WREG32((0x2c24 + j), 0x00000000);
 289         }
 290         WREG32(HDP_REG_COHERENCY_FLUSH_CNTL, 0);
 291
 292         rv515_mc_stop(rdev, &save);
 293         if (r600_mc_wait_for_idle(rdev)) {
 294                 dev_warn(rdev->dev, "Wait for MC idle timedout !\n");
 295         }
 296         /* Lockout access through VGA aperture (doesn't exist before R600) */
 297         WREG32(VGA_HDP_CONTROL, VGA_MEMORY_DISABLE);
 298         /* Update configuration */
 299         if (rdev->flags & RADEON_IS_AGP) {
 300                 if (rdev->mc.vram_start < rdev->mc.gtt_start) {
 301                         /* VRAM before AGP */
 302                         WREG32(MC_VM_SYSTEM_APERTURE_LOW_ADDR,
 303                                 rdev->mc.vram_start >> 12);
 304                         WREG32(MC_VM_SYSTEM_APERTURE_HIGH_ADDR,
 305                                 rdev->mc.gtt_end >> 12);
 306                 } else {
 307                         /* VRAM after AGP */
 308                         WREG32(MC_VM_SYSTEM_APERTURE_LOW_ADDR,
 309                                 rdev->mc.gtt_start >> 12);
 310                         WREG32(MC_VM_SYSTEM_APERTURE_HIGH_ADDR,
 311                                 rdev->mc.vram_end >> 12);
 312                 }
 313         } else {
 314                 WREG32(MC_VM_SYSTEM_APERTURE_LOW_ADDR, rdev->mc.vram_start >> 12);
 315                 WREG32(MC_VM_SYSTEM_APERTURE_HIGH_ADDR, rdev->mc.vram_end >> 12);
 316         }
 317         WREG32(MC_VM_SYSTEM_APERTURE_DEFAULT_ADDR, 0);
 318         tmp = ((rdev->mc.vram_end >> 24) & 0xFFFF) << 16;
 319         tmp |= ((rdev->mc.vram_start >> 24) & 0xFFFF);
 320         WREG32(MC_VM_FB_LOCATION, tmp);
 321         WREG32(HDP_NONSURFACE_BASE, (rdev->mc.vram_start >> 8));
 322         WREG32(HDP_NONSURFACE_INFO, (2 << 7));
 323         WREG32(HDP_NONSURFACE_SIZE, rdev->mc.mc_vram_size | 0x3FF);
 324         if (rdev->flags & RADEON_IS_AGP) {
 325                 WREG32(MC_VM_AGP_TOP, rdev->mc.gtt_end >> 22);
 326                 WREG32(MC_VM_AGP_BOT, rdev->mc.gtt_start >> 22);
 327                 WREG32(MC_VM_AGP_BASE, rdev->mc.agp_base >> 22);
 328         } else {
 329                 WREG32(MC_VM_AGP_BASE, 0);
 330                 WREG32(MC_VM_AGP_TOP, 0x0FFFFFFF);
 331                 WREG32(MC_VM_AGP_BOT, 0x0FFFFFFF);
 332         }
 333         if (r600_mc_wait_for_idle(rdev)) {
 334                 dev_warn(rdev->dev, "Wait for MC idle timedout !\n");
 335         }
 336         rv515_mc_resume(rdev, &save);
 337         /* we need to own VRAM, so turn off the VGA renderer here
 338          * to stop it overwriting our objects */
 339         rv515_vga_render_disable(rdev);
 340 }
 341
 342 int r600_mc_init(struct radeon_device *rdev)
 343 {
 344         fixed20_12 a;
 345         u32 tmp;
 346         int chansize, numchan;
 347         int r;
 348
 349         /* Get VRAM informations */
 350         rdev->mc.vram_is_ddr = true;
 351         tmp = RREG32(RAMCFG);
 352         if (tmp & CHANSIZE_OVERRIDE) {
 353                 chansize = 16;
 354         } else if (tmp & CHANSIZE_MASK) {
 355                 chansize = 64;
 356         } else {
 357                 chansize = 32;
 358         }
 359         tmp = RREG32(CHMAP);
 360         switch ((tmp & NOOFCHAN_MASK) >> NOOFCHAN_SHIFT) {
 361         case 0:
 362         default:
 363                 numchan = 1;
 364                 break;
 365         case 1:
 366                 numchan = 2;
 367                 break;
 368         case 2:
 369                 numchan = 4;
 370                 break;
 371         case 3:
 372                 numchan = 8;
 373                 break;
 374         }
 375         rdev->mc.vram_width = numchan * chansize;
 376         /* Could aper size report 0 ? */
 377         rdev->mc.aper_base = drm_get_resource_start(rdev->ddev, 0);
 378         rdev->mc.aper_size = drm_get_resource_len(rdev->ddev, 0);
 379         /* Setup GPU memory space */
 380         rdev->mc.mc_vram_size = RREG32(CONFIG_MEMSIZE);
 381         rdev->mc.real_vram_size = RREG32(CONFIG_MEMSIZE);
 382
 383         if (rdev->mc.mc_vram_size > rdev->mc.aper_size)
 384                 rdev->mc.mc_vram_size = rdev->mc.aper_size;
 385
 386         if (rdev->mc.real_vram_size > rdev->mc.aper_size)
 387                 rdev->mc.real_vram_size = rdev->mc.aper_size;
 388
 389         if (rdev->flags & RADEON_IS_AGP) {
 390                 r = radeon_agp_init(rdev);
 391                 if (r)
 392                         return r;
 393                 /* gtt_size is setup by radeon_agp_init */
 394                 rdev->mc.gtt_location = rdev->mc.agp_base;
 395                 tmp = 0xFFFFFFFFUL - rdev->mc.agp_base - rdev->mc.gtt_size;
 396                 /* Try to put vram before or after AGP because we
 397                  * we want SYSTEM_APERTURE to cover both VRAM and
 398                  * AGP so that GPU can catch out of VRAM/AGP access
 399                  */
 400                 if (rdev->mc.gtt_location > rdev->mc.mc_vram_size) {
 401                         /* Enought place before */
 402                         rdev->mc.vram_location = rdev->mc.gtt_location -
 403                                                         rdev->mc.mc_vram_size;
 404                 } else if (tmp > rdev->mc.mc_vram_size) {
 405                         /* Enought place after */
 406                         rdev->mc.vram_location = rdev->mc.gtt_location +
 407                                                         rdev->mc.gtt_size;
 408                 } else {
 409                         /* Try to setup VRAM then AGP might not
 410                          * not work on some card
 411                          */
 412                         rdev->mc.vram_location = 0x00000000UL;
 413                         rdev->mc.gtt_location = rdev->mc.mc_vram_size;
 414                 }
 415         } else {
 416                 rdev->mc.gtt_size = radeon_gart_size * 1024 * 1024;
 417                 rdev->mc.vram_location = (RREG32(MC_VM_FB_LOCATION) &
 418                                                         0xFFFF) << 24;
 419                 tmp = rdev->mc.vram_location + rdev->mc.mc_vram_size;
 420                 if ((0xFFFFFFFFUL - tmp) >= rdev->mc.gtt_size) {
 421                         /* Enough place after vram */
 422                         rdev->mc.gtt_location = tmp;
 423                 } else if (rdev->mc.vram_location >= rdev->mc.gtt_size) {
 424                         /* Enough place before vram */
 425                         rdev->mc.gtt_location = 0;
 426                 } else {
 427                         /* Not enough place after or before shrink
 428                          * gart size
 429                          */
 430                         if (rdev->mc.vram_location > (0xFFFFFFFFUL - tmp)) {
 431                                 rdev->mc.gtt_location = 0;
 432                                 rdev->mc.gtt_size = rdev->mc.vram_location;
 433                         } else {
 434                                 rdev->mc.gtt_location = tmp;
 435                                 rdev->mc.gtt_size = 0xFFFFFFFFUL - tmp;
 436                         }
 437                 }
 438                 rdev->mc.gtt_location = rdev->mc.mc_vram_size;
 439         }
 440         rdev->mc.vram_start = rdev->mc.vram_location;
 441         rdev->mc.vram_end = rdev->mc.vram_location + rdev->mc.mc_vram_size - 1;
 442         rdev->mc.gtt_start = rdev->mc.gtt_location;
 443         rdev->mc.gtt_end = rdev->mc.gtt_location + rdev->mc.gtt_size - 1;
 444         /* FIXME: we should enforce default clock in case GPU is not in
 445          * default setup
 446          */
 447         a.full = rfixed_const(100);
 448         rdev->pm.sclk.full = rfixed_const(rdev->clock.default_sclk);
 449         rdev->pm.sclk.full = rfixed_div(rdev->pm.sclk, a);
 450         return 0;
 451 }
 452
 453 /* We doesn't check that the GPU really needs a reset we simply do the
 454  * reset, it's up to the caller to determine if the GPU needs one. We
 455  * might add an helper function to check that.
 456  */
 457 int r600_gpu_soft_reset(struct radeon_device *rdev)
 458 {
 459         struct rv515_mc_save save;
 460         u32 grbm_busy_mask = S_008010_VC_BUSY(1) | S_008010_VGT_BUSY_NO_DMA(1) |
 461                                 S_008010_VGT_BUSY(1) | S_008010_TA03_BUSY(1) |
 462                                 S_008010_TC_BUSY(1) | S_008010_SX_BUSY(1) |
 463                                 S_008010_SH_BUSY(1) | S_008010_SPI03_BUSY(1) |
 464                                 S_008010_SMX_BUSY(1) | S_008010_SC_BUSY(1) |
 465                                 S_008010_PA_BUSY(1) | S_008010_DB03_BUSY(1) |
 466                                 S_008010_CR_BUSY(1) | S_008010_CB03_BUSY(1) |
 467                                 S_008010_GUI_ACTIVE(1);
 468         u32 grbm2_busy_mask = S_008014_SPI0_BUSY(1) | S_008014_SPI1_BUSY(1) |
 469                         S_008014_SPI2_BUSY(1) | S_008014_SPI3_BUSY(1) |
 470                         S_008014_TA0_BUSY(1) | S_008014_TA1_BUSY(1) |
 471                         S_008014_TA2_BUSY(1) | S_008014_TA3_BUSY(1) |
 472                         S_008014_DB0_BUSY(1) | S_008014_DB1_BUSY(1) |
 473                         S_008014_DB2_BUSY(1) | S_008014_DB3_BUSY(1) |
 474                         S_008014_CB0_BUSY(1) | S_008014_CB1_BUSY(1) |
 475                         S_008014_CB2_BUSY(1) | S_008014_CB3_BUSY(1);
 476         u32 srbm_reset = 0;
 477         u32 tmp;
 478
 479         dev_info(rdev->dev, "GPU softreset \n");
 480         dev_info(rdev->dev, "  R_008010_GRBM_STATUS=0x%08X\n",
 481                 RREG32(R_008010_GRBM_STATUS));
 482         dev_info(rdev->dev, "  R_008014_GRBM_STATUS2=0x%08X\n",
 483                 RREG32(R_008014_GRBM_STATUS2));
 484         dev_info(rdev->dev, "  R_000E50_SRBM_STATUS=0x%08X\n",
 485                 RREG32(R_000E50_SRBM_STATUS));
 486         rv515_mc_stop(rdev, &save);
 487         if (r600_mc_wait_for_idle(rdev)) {
 488                 dev_warn(rdev->dev, "Wait for MC idle timedout !\n");
 489         }
 490         /* Disable CP parsing/prefetching */
 491         WREG32(R_0086D8_CP_ME_CNTL, S_0086D8_CP_ME_HALT(0xff));
 492         /* Check if any of the rendering block is busy and reset it */
 493         if ((RREG32(R_008010_GRBM_STATUS) & grbm_busy_mask) ||
 494             (RREG32(R_008014_GRBM_STATUS2) & grbm2_busy_mask)) {
 495                 tmp = S_008020_SOFT_RESET_CR(1) |
 496                         S_008020_SOFT_RESET_DB(1) |
 497                         S_008020_SOFT_RESET_CB(1) |
 498                         S_008020_SOFT_RESET_PA(1) |
 499                         S_008020_SOFT_RESET_SC(1) |
 500                         S_008020_SOFT_RESET_SMX(1) |
 501                         S_008020_SOFT_RESET_SPI(1) |
 502                         S_008020_SOFT_RESET_SX(1) |
 503                         S_008020_SOFT_RESET_SH(1) |
 504                         S_008020_SOFT_RESET_TC(1) |
 505                         S_008020_SOFT_RESET_TA(1) |
 506                         S_008020_SOFT_RESET_VC(1) |
 507                         S_008020_SOFT_RESET_VGT(1);
 508                 dev_info(rdev->dev, "  R_008020_GRBM_SOFT_RESET=0x%08X\n", tmp);
 509                 WREG32(R_008020_GRBM_SOFT_RESET, tmp);
 510                 (void)RREG32(R_008020_GRBM_SOFT_RESET);
 511                 udelay(50);
 512                 WREG32(R_008020_GRBM_SOFT_RESET, 0);
 513                 (void)RREG32(R_008020_GRBM_SOFT_RESET);
 514         }
 515         /* Reset CP (we always reset CP) */
 516         tmp = S_008020_SOFT_RESET_CP(1);
 517         dev_info(rdev->dev, "R_008020_GRBM_SOFT_RESET=0x%08X\n", tmp);
 518         WREG32(R_008020_GRBM_SOFT_RESET, tmp);
 519         (void)RREG32(R_008020_GRBM_SOFT_RESET);
 520         udelay(50);
 521         WREG32(R_008020_GRBM_SOFT_RESET, 0);
 522         (void)RREG32(R_008020_GRBM_SOFT_RESET);
 523         /* Reset others GPU block if necessary */
 524         if (G_000E50_RLC_BUSY(RREG32(R_000E50_SRBM_STATUS)))
 525                 srbm_reset |= S_000E60_SOFT_RESET_RLC(1);
 526         if (G_000E50_GRBM_RQ_PENDING(RREG32(R_000E50_SRBM_STATUS)))
 527                 srbm_reset |= S_000E60_SOFT_RESET_GRBM(1);
 528         if (G_000E50_HI_RQ_PENDING(RREG32(R_000E50_SRBM_STATUS)))
 529                 srbm_reset |= S_000E60_SOFT_RESET_IH(1);
 530         if (G_000E50_VMC_BUSY(RREG32(R_000E50_SRBM_STATUS)))
 531                 srbm_reset |= S_000E60_SOFT_RESET_VMC(1);
 532         if (G_000E50_MCB_BUSY(RREG32(R_000E50_SRBM_STATUS)))
 533                 srbm_reset |= S_000E60_SOFT_RESET_MC(1);
 534         if (G_000E50_MCDZ_BUSY(RREG32(R_000E50_SRBM_STATUS)))
 535                 srbm_reset |= S_000E60_SOFT_RESET_MC(1);
 536         if (G_000E50_MCDY_BUSY(RREG32(R_000E50_SRBM_STATUS)))
 537                 srbm_reset |= S_000E60_SOFT_RESET_MC(1);
 538         if (G_000E50_MCDX_BUSY(RREG32(R_000E50_SRBM_STATUS)))
 539                 srbm_reset |= S_000E60_SOFT_RESET_MC(1);
 540         if (G_000E50_MCDW_BUSY(RREG32(R_000E50_SRBM_STATUS)))
 541                 srbm_reset |= S_000E60_SOFT_RESET_MC(1);
 542         if (G_000E50_RLC_BUSY(RREG32(R_000E50_SRBM_STATUS)))
 543                 srbm_reset |= S_000E60_SOFT_RESET_RLC(1);
 544         if (G_000E50_SEM_BUSY(RREG32(R_000E50_SRBM_STATUS)))
 545                 srbm_reset |= S_000E60_SOFT_RESET_SEM(1);
 546         if (G_000E50_BIF_BUSY(RREG32(R_000E50_SRBM_STATUS)))
 547                 srbm_reset |= S_000E60_SOFT_RESET_BIF(1);
 548         dev_info(rdev->dev, "  R_000E60_SRBM_SOFT_RESET=0x%08X\n", srbm_reset);
 549         WREG32(R_000E60_SRBM_SOFT_RESET, srbm_reset);
 550         (void)RREG32(R_000E60_SRBM_SOFT_RESET);
 551         udelay(50);
 552         WREG32(R_000E60_SRBM_SOFT_RESET, 0);
 553         (void)RREG32(R_000E60_SRBM_SOFT_RESET);
 554         WREG32(R_000E60_SRBM_SOFT_RESET, srbm_reset);
 555         (void)RREG32(R_000E60_SRBM_SOFT_RESET);
 556         udelay(50);
 557         WREG32(R_000E60_SRBM_SOFT_RESET, 0);
 558         (void)RREG32(R_000E60_SRBM_SOFT_RESET);
 559         /* Wait a little for things to settle down */
 560         udelay(50);
 561         dev_info(rdev->dev, "  R_008010_GRBM_STATUS=0x%08X\n",
 562                 RREG32(R_008010_GRBM_STATUS));
 563         dev_info(rdev->dev, "  R_008014_GRBM_STATUS2=0x%08X\n",
 564                 RREG32(R_008014_GRBM_STATUS2));
 565         dev_info(rdev->dev, "  R_000E50_SRBM_STATUS=0x%08X\n",
 566                 RREG32(R_000E50_SRBM_STATUS));
 567         /* After reset we need to reinit the asic as GPU often endup in an
 568          * incoherent state.
 569          */
 570         atom_asic_init(rdev->mode_info.atom_context);
 571         rv515_mc_resume(rdev, &save);
 572         return 0;
 573 }
 574
 575 int r600_gpu_reset(struct radeon_device *rdev)
 576 {
 577         return r600_gpu_soft_reset(rdev);
 578 }
 579
 580 static u32 r600_get_tile_pipe_to_backend_map(u32 num_tile_pipes,
 581                                              u32 num_backends,
 582                                              u32 backend_disable_mask)
 583 {
 584         u32 backend_map = 0;
 585         u32 enabled_backends_mask;
 586         u32 enabled_backends_count;
 587         u32 cur_pipe;
 588         u32 swizzle_pipe[R6XX_MAX_PIPES];
 589         u32 cur_backend;
 590         u32 i;
 591
 592         if (num_tile_pipes > R6XX_MAX_PIPES)
 593                 num_tile_pipes = R6XX_MAX_PIPES;
 594         if (num_tile_pipes < 1)
 595                 num_tile_pipes = 1;
 596         if (num_backends > R6XX_MAX_BACKENDS)
 597                 num_backends = R6XX_MAX_BACKENDS;
 598         if (num_backends < 1)
 599                 num_backends = 1;
 600
 601         enabled_backends_mask = 0;
 602         enabled_backends_count = 0;
 603         for (i = 0; i < R6XX_MAX_BACKENDS; ++i) {
 604                 if (((backend_disable_mask >> i) & 1) == 0) {
 605                         enabled_backends_mask |= (1 << i);
 606                         ++enabled_backends_count;
 607                 }
 608                 if (enabled_backends_count == num_backends)
 609                         break;
 610         }
 611
 612         if (enabled_backends_count == 0) {
 613                 enabled_backends_mask = 1;
 614                 enabled_backends_count = 1;
 615         }
 616
 617         if (enabled_backends_count != num_backends)
 618                 num_backends = enabled_backends_count;
 619
 620         memset((uint8_t *)&swizzle_pipe[0], 0, sizeof(u32) * R6XX_MAX_PIPES);
 621         switch (num_tile_pipes) {
 622         case 1:
 623                 swizzle_pipe[0] = 0;
 624                 break;
 625         case 2:
 626                 swizzle_pipe[0] = 0;
 627                 swizzle_pipe[1] = 1;
 628                 break;
 629         case 3:
 630                 swizzle_pipe[0] = 0;
 631                 swizzle_pipe[1] = 1;
 632                 swizzle_pipe[2] = 2;
 633                 break;
 634         case 4:
 635                 swizzle_pipe[0] = 0;
 636                 swizzle_pipe[1] = 1;
 637                 swizzle_pipe[2] = 2;
 638                 swizzle_pipe[3] = 3;
 639                 break;
 640         case 5:
 641                 swizzle_pipe[0] = 0;
 642                 swizzle_pipe[1] = 1;
 643                 swizzle_pipe[2] = 2;
 644                 swizzle_pipe[3] = 3;
 645                 swizzle_pipe[4] = 4;
 646                 break;
 647         case 6:
 648                 swizzle_pipe[0] = 0;
 649                 swizzle_pipe[1] = 2;
 650                 swizzle_pipe[2] = 4;
 651                 swizzle_pipe[3] = 5;
 652                 swizzle_pipe[4] = 1;
 653                 swizzle_pipe[5] = 3;
 654                 break;
 655         case 7:
 656                 swizzle_pipe[0] = 0;
 657                 swizzle_pipe[1] = 2;
 658                 swizzle_pipe[2] = 4;
 659                 swizzle_pipe[3] = 6;
 660                 swizzle_pipe[4] = 1;
 661                 swizzle_pipe[5] = 3;
 662                 swizzle_pipe[6] = 5;
 663                 break;
 664         case 8:
 665                 swizzle_pipe[0] = 0;
 666                 swizzle_pipe[1] = 2;
 667                 swizzle_pipe[2] = 4;
 668                 swizzle_pipe[3] = 6;
 669                 swizzle_pipe[4] = 1;
 670                 swizzle_pipe[5] = 3;
 671                 swizzle_pipe[6] = 5;
 672                 swizzle_pipe[7] = 7;
 673                 break;
 674         }
 675
 676         cur_backend = 0;
 677         for (cur_pipe = 0; cur_pipe < num_tile_pipes; ++cur_pipe) {
 678                 while (((1 << cur_backend) & enabled_backends_mask) == 0)
 679                         cur_backend = (cur_backend + 1) % R6XX_MAX_BACKENDS;
 680
 681                 backend_map |= (u32)(((cur_backend & 3) << (swizzle_pipe[cur_pipe] * 2)));
 682
 683                 cur_backend = (cur_backend + 1) % R6XX_MAX_BACKENDS;
 684         }
 685
 686         return backend_map;
 687 }
 688
 689 int r600_count_pipe_bits(uint32_t val)
 690 {
 691         int i, ret = 0;
 692
 693         for (i = 0; i < 32; i++) {
 694                 ret += val & 1;
 695                 val >>= 1;
 696         }
 697         return ret;
 698 }
 699
 700 void r600_gpu_init(struct radeon_device *rdev)
 701 {
 702         u32 tiling_config;
 703         u32 ramcfg;
 704         u32 tmp;
 705         int i, j;
 706         u32 sq_config;
 707         u32 sq_gpr_resource_mgmt_1 = 0;
 708         u32 sq_gpr_resource_mgmt_2 = 0;
 709         u32 sq_thread_resource_mgmt = 0;
 710         u32 sq_stack_resource_mgmt_1 = 0;
 711         u32 sq_stack_resource_mgmt_2 = 0;
 712
 713         /* FIXME: implement */
 714         switch (rdev->family) {
 715         case CHIP_R600:
 716                 rdev->config.r600.max_pipes = 4;
 717                 rdev->config.r600.max_tile_pipes = 8;
 718                 rdev->config.r600.max_simds = 4;
 719                 rdev->config.r600.max_backends = 4;
 720                 rdev->config.r600.max_gprs = 256;
 721                 rdev->config.r600.max_threads = 192;
 722                 rdev->config.r600.max_stack_entries = 256;
 723                 rdev->config.r600.max_hw_contexts = 8;
 724                 rdev->config.r600.max_gs_threads = 16;
 725                 rdev->config.r600.sx_max_export_size = 128;
 726                 rdev->config.r600.sx_max_export_pos_size = 16;
 727                 rdev->config.r600.sx_max_export_smx_size = 128;
 728                 rdev->config.r600.sq_num_cf_insts = 2;
 729                 break;
 730         case CHIP_RV630:
 731         case CHIP_RV635:
 732                 rdev->config.r600.max_pipes = 2;
 733                 rdev->config.r600.max_tile_pipes = 2;
 734                 rdev->config.r600.max_simds = 3;
 735                 rdev->config.r600.max_backends = 1;
 736                 rdev->config.r600.max_gprs = 128;
 737                 rdev->config.r600.max_threads = 192;
 738                 rdev->config.r600.max_stack_entries = 128;
 739                 rdev->config.r600.max_hw_contexts = 8;
 740                 rdev->config.r600.max_gs_threads = 4;
 741                 rdev->config.r600.sx_max_export_size = 128;
 742                 rdev->config.r600.sx_max_export_pos_size = 16;
 743                 rdev->config.r600.sx_max_export_smx_size = 128;
 744                 rdev->config.r600.sq_num_cf_insts = 2;
 745                 break;
 746         case CHIP_RV610:
 747         case CHIP_RV620:
 748         case CHIP_RS780:
 749         case CHIP_RS880:
 750                 rdev->config.r600.max_pipes = 1;
 751                 rdev->config.r600.max_tile_pipes = 1;
 752                 rdev->config.r600.max_simds = 2;
 753                 rdev->config.r600.max_backends = 1;
 754                 rdev->config.r600.max_gprs = 128;
 755                 rdev->config.r600.max_threads = 192;
 756                 rdev->config.r600.max_stack_entries = 128;
 757                 rdev->config.r600.max_hw_contexts = 4;
 758                 rdev->config.r600.max_gs_threads = 4;
 759                 rdev->config.r600.sx_max_export_size = 128;
 760                 rdev->config.r600.sx_max_export_pos_size = 16;
 761                 rdev->config.r600.sx_max_export_smx_size = 128;
 762                 rdev->config.r600.sq_num_cf_insts = 1;
 763                 break;
 764         case CHIP_RV670:
 765                 rdev->config.r600.max_pipes = 4;
 766                 rdev->config.r600.max_tile_pipes = 4;
 767                 rdev->config.r600.max_simds = 4;
 768                 rdev->config.r600.max_backends = 4;
 769                 rdev->config.r600.max_gprs = 192;
 770                 rdev->config.r600.max_threads = 192;
 771                 rdev->config.r600.max_stack_entries = 256;
 772                 rdev->config.r600.max_hw_contexts = 8;
 773                 rdev->config.r600.max_gs_threads = 16;
 774                 rdev->config.r600.sx_max_export_size = 128;
 775                 rdev->config.r600.sx_max_export_pos_size = 16;
 776                 rdev->config.r600.sx_max_export_smx_size = 128;
 777                 rdev->config.r600.sq_num_cf_insts = 2;
 778                 break;
 779         default:
 780                 break;
 781         }
 782
 783         /* Initialize HDP */
 784         for (i = 0, j = 0; i < 32; i++, j += 0x18) {
 785                 WREG32((0x2c14 + j), 0x00000000);
 786                 WREG32((0x2c18 + j), 0x00000000);
 787                 WREG32((0x2c1c + j), 0x00000000);
 788                 WREG32((0x2c20 + j), 0x00000000);
 789                 WREG32((0x2c24 + j), 0x00000000);
 790         }
 791
 792         WREG32(GRBM_CNTL, GRBM_READ_TIMEOUT(0xff));
 793
 794         /* Setup tiling */
 795         tiling_config = 0;
 796         ramcfg = RREG32(RAMCFG);
 797         switch (rdev->config.r600.max_tile_pipes) {
 798         case 1:
 799                 tiling_config |= PIPE_TILING(0);
 800                 break;
 801         case 2:
 802                 tiling_config |= PIPE_TILING(1);
 803                 break;
 804         case 4:
 805                 tiling_config |= PIPE_TILING(2);
 806                 break;
 807         case 8:
 808                 tiling_config |= PIPE_TILING(3);
 809                 break;
 810         default:
 811                 break;
 812         }
 813         tiling_config |= BANK_TILING((ramcfg & NOOFBANK_MASK) >> NOOFBANK_SHIFT);
 814         tiling_config |= GROUP_SIZE(0);
 815         tmp = (ramcfg & NOOFROWS_MASK) >> NOOFROWS_SHIFT;
 816         if (tmp > 3) {
 817                 tiling_config |= ROW_TILING(3);
 818                 tiling_config |= SAMPLE_SPLIT(3);
 819         } else {
 820                 tiling_config |= ROW_TILING(tmp);
 821                 tiling_config |= SAMPLE_SPLIT(tmp);
 822         }
 823         tiling_config |= BANK_SWAPS(1);
 824         tmp = r600_get_tile_pipe_to_backend_map(rdev->config.r600.max_tile_pipes,
 825                                                 rdev->config.r600.max_backends,
 826                                                 (0xff << rdev->config.r600.max_backends) & 0xff);
 827         tiling_config |= BACKEND_MAP(tmp);
 828         WREG32(GB_TILING_CONFIG, tiling_config);
 829         WREG32(DCP_TILING_CONFIG, tiling_config & 0xffff);
 830         WREG32(HDP_TILING_CONFIG, tiling_config & 0xffff);
 831
 832         tmp = BACKEND_DISABLE((R6XX_MAX_BACKENDS_MASK << rdev->config.r600.max_backends) & R6XX_MAX_BACKENDS_MASK);
 833         WREG32(CC_RB_BACKEND_DISABLE, tmp);
 834
 835         /* Setup pipes */
 836         tmp = INACTIVE_QD_PIPES((R6XX_MAX_PIPES_MASK << rdev->config.r600.max_pipes) & R6XX_MAX_PIPES_MASK);
 837         tmp |= INACTIVE_SIMDS((R6XX_MAX_SIMDS_MASK << rdev->config.r600.max_simds) & R6XX_MAX_SIMDS_MASK);
 838         WREG32(CC_GC_SHADER_PIPE_CONFIG, tmp);
 839         WREG32(GC_USER_SHADER_PIPE_CONFIG, tmp);
 840
 841         tmp = R6XX_MAX_BACKENDS - r600_count_pipe_bits(tmp & INACTIVE_QD_PIPES_MASK);
 842         WREG32(VGT_OUT_DEALLOC_CNTL, (tmp * 4) & DEALLOC_DIST_MASK);
 843         WREG32(VGT_VERTEX_REUSE_BLOCK_CNTL, ((tmp * 4) - 2) & VTX_REUSE_DEPTH_MASK);
 844
 845         /* Setup some CP states */
 846         WREG32(CP_QUEUE_THRESHOLDS, (ROQ_IB1_START(0x16) | ROQ_IB2_START(0x2b)));
 847         WREG32(CP_MEQ_THRESHOLDS, (MEQ_END(0x40) | ROQ_END(0x40)));
 848
 849         WREG32(TA_CNTL_AUX, (DISABLE_CUBE_ANISO | SYNC_GRADIENT |
 850                              SYNC_WALKER | SYNC_ALIGNER));
 851         /* Setup various GPU states */
 852         if (rdev->family == CHIP_RV670)
 853                 WREG32(ARB_GDEC_RD_CNTL, 0x00000021);
 854
 855         tmp = RREG32(SX_DEBUG_1);
 856         tmp |= SMX_EVENT_RELEASE;
 857         if ((rdev->family > CHIP_R600))
 858                 tmp |= ENABLE_NEW_SMX_ADDRESS;
 859         WREG32(SX_DEBUG_1, tmp);
 860
 861         if (((rdev->family) == CHIP_R600) ||
 862             ((rdev->family) == CHIP_RV630) ||
 863             ((rdev->family) == CHIP_RV610) ||
 864             ((rdev->family) == CHIP_RV620) ||
 865             ((rdev->family) == CHIP_RS780) ||
 866             ((rdev->family) == CHIP_RS880)) {
 867                 WREG32(DB_DEBUG, PREZ_MUST_WAIT_FOR_POSTZ_DONE);
 868         } else {
 869                 WREG32(DB_DEBUG, 0);
 870         }
 871         WREG32(DB_WATERMARKS, (DEPTH_FREE(4) | DEPTH_CACHELINE_FREE(16) |
 872                                DEPTH_FLUSH(16) | DEPTH_PENDING_FREE(4)));
 873
 874         WREG32(PA_SC_MULTI_CHIP_CNTL, 0);
 875         WREG32(VGT_NUM_INSTANCES, 0);
 876
 877         WREG32(SPI_CONFIG_CNTL, GPR_WRITE_PRIORITY(0));
 878         WREG32(SPI_CONFIG_CNTL_1, VTX_DONE_DELAY(0));
 879
 880         tmp = RREG32(SQ_MS_FIFO_SIZES);
 881         if (((rdev->family) == CHIP_RV610) ||
 882             ((rdev->family) == CHIP_RV620) ||
 883             ((rdev->family) == CHIP_RS780) ||
 884             ((rdev->family) == CHIP_RS880)) {
 885                 tmp = (CACHE_FIFO_SIZE(0xa) |
 886                        FETCH_FIFO_HIWATER(0xa) |
 887                        DONE_FIFO_HIWATER(0xe0) |
 888                        ALU_UPDATE_FIFO_HIWATER(0x8));
 889         } else if (((rdev->family) == CHIP_R600) ||
 890                    ((rdev->family) == CHIP_RV630)) {
 891                 tmp &= ~DONE_FIFO_HIWATER(0xff);
 892                 tmp |= DONE_FIFO_HIWATER(0x4);
 893         }
 894         WREG32(SQ_MS_FIFO_SIZES, tmp);
 895
 896         /* SQ_CONFIG, SQ_GPR_RESOURCE_MGMT, SQ_THREAD_RESOURCE_MGMT, SQ_STACK_RESOURCE_MGMT
 897          * should be adjusted as needed by the 2D/3D drivers.  This just sets default values
 898          */
 899         sq_config = RREG32(SQ_CONFIG);
 900         sq_config &= ~(PS_PRIO(3) |
 901                        VS_PRIO(3) |
 902                        GS_PRIO(3) |
 903                        ES_PRIO(3));
 904         sq_config |= (DX9_CONSTS |
 905                       VC_ENABLE |
 906                       PS_PRIO(0) |
 907                       VS_PRIO(1) |
 908                       GS_PRIO(2) |
 909                       ES_PRIO(3));
 910
 911         if ((rdev->family) == CHIP_R600) {
 912                 sq_gpr_resource_mgmt_1 = (NUM_PS_GPRS(124) |
 913                                           NUM_VS_GPRS(124) |
 914                                           NUM_CLAUSE_TEMP_GPRS(4));
 915                 sq_gpr_resource_mgmt_2 = (NUM_GS_GPRS(0) |
 916                                           NUM_ES_GPRS(0));
 917                 sq_thread_resource_mgmt = (NUM_PS_THREADS(136) |
 918                                            NUM_VS_THREADS(48) |
 919                                            NUM_GS_THREADS(4) |
 920                                            NUM_ES_THREADS(4));
 921                 sq_stack_resource_mgmt_1 = (NUM_PS_STACK_ENTRIES(128) |
 922                                             NUM_VS_STACK_ENTRIES(128));
 923                 sq_stack_resource_mgmt_2 = (NUM_GS_STACK_ENTRIES(0) |
 924                                             NUM_ES_STACK_ENTRIES(0));
 925         } else if (((rdev->family) == CHIP_RV610) ||
 926                    ((rdev->family) == CHIP_RV620) ||
 927                    ((rdev->family) == CHIP_RS780) ||
 928                    ((rdev->family) == CHIP_RS880)) {
 929                 /* no vertex cache */
 930                 sq_config &= ~VC_ENABLE;
 931
 932                 sq_gpr_resource_mgmt_1 = (NUM_PS_GPRS(44) |
 933                                           NUM_VS_GPRS(44) |
 934                                           NUM_CLAUSE_TEMP_GPRS(2));
 935                 sq_gpr_resource_mgmt_2 = (NUM_GS_GPRS(17) |
 936                                           NUM_ES_GPRS(17));
 937                 sq_thread_resource_mgmt = (NUM_PS_THREADS(79) |
 938                                            NUM_VS_THREADS(78) |
 939                                            NUM_GS_THREADS(4) |
 940                                            NUM_ES_THREADS(31));
 941                 sq_stack_resource_mgmt_1 = (NUM_PS_STACK_ENTRIES(40) |
 942                                             NUM_VS_STACK_ENTRIES(40));
 943                 sq_stack_resource_mgmt_2 = (NUM_GS_STACK_ENTRIES(32) |
 944                                             NUM_ES_STACK_ENTRIES(16));
 945         } else if (((rdev->family) == CHIP_RV630) ||
 946                    ((rdev->family) == CHIP_RV635)) {
 947                 sq_gpr_resource_mgmt_1 = (NUM_PS_GPRS(44) |
 948                                           NUM_VS_GPRS(44) |
 949                                           NUM_CLAUSE_TEMP_GPRS(2));
 950                 sq_gpr_resource_mgmt_2 = (NUM_GS_GPRS(18) |
 951                                           NUM_ES_GPRS(18));
 952                 sq_thread_resource_mgmt = (NUM_PS_THREADS(79) |
 953                                            NUM_VS_THREADS(78) |
 954                                            NUM_GS_THREADS(4) |
 955                                            NUM_ES_THREADS(31));
 956                 sq_stack_resource_mgmt_1 = (NUM_PS_STACK_ENTRIES(40) |
 957                                             NUM_VS_STACK_ENTRIES(40));
 958                 sq_stack_resource_mgmt_2 = (NUM_GS_STACK_ENTRIES(32) |
 959                                             NUM_ES_STACK_ENTRIES(16));
 960         } else if ((rdev->family) == CHIP_RV670) {
 961                 sq_gpr_resource_mgmt_1 = (NUM_PS_GPRS(44) |
 962                                           NUM_VS_GPRS(44) |
 963                                           NUM_CLAUSE_TEMP_GPRS(2));
 964                 sq_gpr_resource_mgmt_2 = (NUM_GS_GPRS(17) |
 965                                           NUM_ES_GPRS(17));
 966                 sq_thread_resource_mgmt = (NUM_PS_THREADS(79) |
 967                                            NUM_VS_THREADS(78) |
 968                                            NUM_GS_THREADS(4) |
 969                                            NUM_ES_THREADS(31));
 970                 sq_stack_resource_mgmt_1 = (NUM_PS_STACK_ENTRIES(64) |
 971                                             NUM_VS_STACK_ENTRIES(64));
 972                 sq_stack_resource_mgmt_2 = (NUM_GS_STACK_ENTRIES(64) |
 973                                             NUM_ES_STACK_ENTRIES(64));
 974         }
 975
 976         WREG32(SQ_CONFIG, sq_config);
 977         WREG32(SQ_GPR_RESOURCE_MGMT_1,  sq_gpr_resource_mgmt_1);
 978         WREG32(SQ_GPR_RESOURCE_MGMT_2,  sq_gpr_resource_mgmt_2);
 979         WREG32(SQ_THREAD_RESOURCE_MGMT, sq_thread_resource_mgmt);
 980         WREG32(SQ_STACK_RESOURCE_MGMT_1, sq_stack_resource_mgmt_1);
 981         WREG32(SQ_STACK_RESOURCE_MGMT_2, sq_stack_resource_mgmt_2);
 982
 983         if (((rdev->family) == CHIP_RV610) ||
 984             ((rdev->family) == CHIP_RV620) ||
 985             ((rdev->family) == CHIP_RS780) ||
 986             ((rdev->family) == CHIP_RS880)) {
 987                 WREG32(VGT_CACHE_INVALIDATION, CACHE_INVALIDATION(TC_ONLY));
 988         } else {
 989                 WREG32(VGT_CACHE_INVALIDATION, CACHE_INVALIDATION(VC_AND_TC));
 990         }
 991
 992         /* More default values. 2D/3D driver should adjust as needed */
 993         WREG32(PA_SC_AA_SAMPLE_LOCS_2S, (S0_X(0xc) | S0_Y(0x4) |
 994                                          S1_X(0x4) | S1_Y(0xc)));
 995         WREG32(PA_SC_AA_SAMPLE_LOCS_4S, (S0_X(0xe) | S0_Y(0xe) |
 996                                          S1_X(0x2) | S1_Y(0x2) |
 997                                          S2_X(0xa) | S2_Y(0x6) |
 998                                          S3_X(0x6) | S3_Y(0xa)));
 999         WREG32(PA_SC_AA_SAMPLE_LOCS_8S_WD0, (S0_X(0xe) | S0_Y(0xb) |
1000                                              S1_X(0x4) | S1_Y(0xc) |
1001                                              S2_X(0x1) | S2_Y(0x6) |
1002                                              S3_X(0xa) | S3_Y(0xe)));
1003         WREG32(PA_SC_AA_SAMPLE_LOCS_8S_WD1, (S4_X(0x6) | S4_Y(0x1) |
1004                                              S5_X(0x0) | S5_Y(0x0) |
1005                                              S6_X(0xb) | S6_Y(0x4) |
1006                                              S7_X(0x7) | S7_Y(0x8)));
1007
1008         WREG32(VGT_STRMOUT_EN, 0);
1009         tmp = rdev->config.r600.max_pipes * 16;
1010         switch (rdev->family) {
1011         case CHIP_RV610:
1012         case CHIP_RV620:
1013         case CHIP_RS780:
1014         case CHIP_RS880:
1015                 tmp += 32;
1016                 break;
1017         case CHIP_RV670:
1018                 tmp += 128;
1019                 break;
1020         default:
1021                 break;
1022         }
1023         if (tmp > 256) {
1024                 tmp = 256;
1025         }
1026         WREG32(VGT_ES_PER_GS, 128);
1027         WREG32(VGT_GS_PER_ES, tmp);
1028         WREG32(VGT_GS_PER_VS, 2);
1029         WREG32(VGT_GS_VERTEX_REUSE, 16);
1030
1031         /* more default values. 2D/3D driver should adjust as needed */
1032         WREG32(PA_SC_LINE_STIPPLE_STATE, 0);
1033         WREG32(VGT_STRMOUT_EN, 0);
1034         WREG32(SX_MISC, 0);
1035         WREG32(PA_SC_MODE_CNTL, 0);
1036         WREG32(PA_SC_AA_CONFIG, 0);
1037         WREG32(PA_SC_LINE_STIPPLE, 0);
1038         WREG32(SPI_INPUT_Z, 0);
1039         WREG32(SPI_PS_IN_CONTROL_0, NUM_INTERP(2));
1040         WREG32(CB_COLOR7_FRAG, 0);
1041
1042         /* Clear render buffer base addresses */
1043         WREG32(CB_COLOR0_BASE, 0);
1044         WREG32(CB_COLOR1_BASE, 0);
1045         WREG32(CB_COLOR2_BASE, 0);
1046         WREG32(CB_COLOR3_BASE, 0);
1047         WREG32(CB_COLOR4_BASE, 0);
1048         WREG32(CB_COLOR5_BASE, 0);
1049         WREG32(CB_COLOR6_BASE, 0);
1050         WREG32(CB_COLOR7_BASE, 0);
1051         WREG32(CB_COLOR7_FRAG, 0);
1052
1053         switch (rdev->family) {
1054         case CHIP_RV610:
1055         case CHIP_RV620:
1056         case CHIP_RS780:
1057         case CHIP_RS880:
1058                 tmp = TC_L2_SIZE(8);
1059                 break;
1060         case CHIP_RV630:
1061         case CHIP_RV635:
1062                 tmp = TC_L2_SIZE(4);
1063                 break;
1064         case CHIP_R600:
1065                 tmp = TC_L2_SIZE(0) | L2_DISABLE_LATE_HIT;
1066                 break;
1067         default:
1068                 tmp = TC_L2_SIZE(0);
1069                 break;
1070         }
1071         WREG32(TC_CNTL, tmp);
1072
1073         tmp = RREG32(HDP_HOST_PATH_CNTL);
1074         WREG32(HDP_HOST_PATH_CNTL, tmp);
1075
1076         tmp = RREG32(ARB_POP);
1077         tmp |= ENABLE_TC128;
1078         WREG32(ARB_POP, tmp);
1079
1080         WREG32(PA_SC_MULTI_CHIP_CNTL, 0);
1081         WREG32(PA_CL_ENHANCE, (CLIP_VTX_REORDER_ENA |
1082                                NUM_CLIP_SEQ(3)));
1083         WREG32(PA_SC_ENHANCE, FORCE_EOV_MAX_CLK_CNT(4095));
1084 }
1085
1086
1087 /*
1088  * Indirect registers accessor
1089  */
1090 u32 r600_pciep_rreg(struct radeon_device *rdev, u32 reg)
1091 {
1092         u32 r;
1093
1094         WREG32(PCIE_PORT_INDEX, ((reg) & 0xff));
1095         (void)RREG32(PCIE_PORT_INDEX);
1096         r = RREG32(PCIE_PORT_DATA);
1097         return r;
1098 }
1099
1100 void r600_pciep_wreg(struct radeon_device *rdev, u32 reg, u32 v)
1101 {
1102         WREG32(PCIE_PORT_INDEX, ((reg) & 0xff));
1103         (void)RREG32(PCIE_PORT_INDEX);
1104         WREG32(PCIE_PORT_DATA, (v));
1105         (void)RREG32(PCIE_PORT_DATA);
1106 }
1107
1108 void r600_hdp_flush(struct radeon_device *rdev)
1109 {
1110         WREG32(R_005480_HDP_MEM_COHERENCY_FLUSH_CNTL, 0x1);
1111 }
1112
1113 /*
1114  * CP & Ring
1115  */
1116 void r600_cp_stop(struct radeon_device *rdev)
1117 {
1118         WREG32(R_0086D8_CP_ME_CNTL, S_0086D8_CP_ME_HALT(1));
1119 }
1120
1121 int r600_init_microcode(struct radeon_device *rdev)
1122 {
1123         struct platform_device *pdev;
1124         const char *chip_name;
1125         const char *rlc_chip_name;
1126         size_t pfp_req_size, me_req_size, rlc_req_size;
1127         char fw_name[30];
1128         int err;
1129
1130         DRM_DEBUG("\n");
1131
1132         pdev = platform_device_register_simple("radeon_cp", 0, NULL, 0);
1133         err = IS_ERR(pdev);
1134         if (err) {
1135                 printk(KERN_ERR "radeon_cp: Failed to register firmware\n");
1136                 return -EINVAL;
1137         }
1138
1139         switch (rdev->family) {
1140         case CHIP_R600:
1141                 chip_name = "R600";
1142                 rlc_chip_name = "R600";
1143                 break;
1144         case CHIP_RV610:
1145                 chip_name = "RV610";
1146                 rlc_chip_name = "R600";
1147                 break;
1148         case CHIP_RV630:
1149                 chip_name = "RV630";
1150                 rlc_chip_name = "R600";
1151                 break;
1152         case CHIP_RV620:
1153                 chip_name = "RV620";
1154                 rlc_chip_name = "R600";
1155                 break;
1156         case CHIP_RV635:
1157                 chip_name = "RV635";
1158                 rlc_chip_name = "R600";
1159                 break;
1160         case CHIP_RV670:
1161                 chip_name = "RV670";
1162                 rlc_chip_name = "R600";
1163                 break;
1164         case CHIP_RS780:
1165         case CHIP_RS880:
1166                 chip_name = "RS780";
1167                 rlc_chip_name = "R600";
1168                 break;
1169         case CHIP_RV770:
1170                 chip_name = "RV770";
1171                 rlc_chip_name = "R700";
1172                 break;
1173         case CHIP_RV730:
1174         case CHIP_RV740:
1175                 chip_name = "RV730";
1176                 rlc_chip_name = "R700";
1177                 break;
1178         case CHIP_RV710:
1179                 chip_name = "RV710";
1180                 rlc_chip_name = "R700";
1181                 break;
1182         default: BUG();
1183         }
1184
1185         if (rdev->family >= CHIP_RV770) {
1186                 pfp_req_size = R700_PFP_UCODE_SIZE * 4;
1187                 me_req_size = R700_PM4_UCODE_SIZE * 4;
1188                 rlc_req_size = R700_RLC_UCODE_SIZE * 4;
1189         } else {
1190                 pfp_req_size = PFP_UCODE_SIZE * 4;
1191                 me_req_size = PM4_UCODE_SIZE * 12;
1192                 rlc_req_size = RLC_UCODE_SIZE * 4;
1193         }
1194
1195         DRM_INFO("Loading %s Microcode\n", chip_name);
1196
1197         snprintf(fw_name, sizeof(fw_name), "radeon/%s_pfp.bin", chip_name);
1198         err = request_firmware(&rdev->pfp_fw, fw_name, &pdev->dev);
1199         if (err)
1200                 goto out;
1201         if (rdev->pfp_fw->size != pfp_req_size) {
1202                 printk(KERN_ERR
1203                        "r600_cp: Bogus length %zu in firmware \"%s\"\n",
1204                        rdev->pfp_fw->size, fw_name);
1205                 err = -EINVAL;
1206                 goto out;
1207         }
1208
1209         snprintf(fw_name, sizeof(fw_name), "radeon/%s_me.bin", chip_name);
1210         err = request_firmware(&rdev->me_fw, fw_name, &pdev->dev);
1211         if (err)
1212                 goto out;
1213         if (rdev->me_fw->size != me_req_size) {
1214                 printk(KERN_ERR
1215                        "r600_cp: Bogus length %zu in firmware \"%s\"\n",
1216                        rdev->me_fw->size, fw_name);
1217                 err = -EINVAL;
1218         }
1219
1220         snprintf(fw_name, sizeof(fw_name), "radeon/%s_rlc.bin", rlc_chip_name);
1221         err = request_firmware(&rdev->rlc_fw, fw_name, &pdev->dev);
1222         if (err)
1223                 goto out;
1224         if (rdev->rlc_fw->size != rlc_req_size) {
1225                 printk(KERN_ERR
1226                        "r600_rlc: Bogus length %zu in firmware \"%s\"\n",
1227                        rdev->rlc_fw->size, fw_name);
1228                 err = -EINVAL;
1229         }
1230
1231 out:
1232         platform_device_unregister(pdev);
1233
1234         if (err) {
1235                 if (err != -EINVAL)
1236                         printk(KERN_ERR
1237                                "r600_cp: Failed to load firmware \"%s\"\n",
1238                                fw_name);
1239                 release_firmware(rdev->pfp_fw);
1240                 rdev->pfp_fw = NULL;
1241                 release_firmware(rdev->me_fw);
1242                 rdev->me_fw = NULL;
1243                 release_firmware(rdev->rlc_fw);
1244                 rdev->rlc_fw = NULL;
1245         }
1246         return err;
1247 }
1248
1249 static int r600_cp_load_microcode(struct radeon_device *rdev)
1250 {
1251         const __be32 *fw_data;
1252         int i;
1253
1254         if (!rdev->me_fw || !rdev->pfp_fw)
1255                 return -EINVAL;
1256
1257         r600_cp_stop(rdev);
1258
1259         WREG32(CP_RB_CNTL, RB_NO_UPDATE | RB_BLKSZ(15) | RB_BUFSZ(3));
1260
1261         /* Reset cp */
1262         WREG32(GRBM_SOFT_RESET, SOFT_RESET_CP);
1263         RREG32(GRBM_SOFT_RESET);
1264         mdelay(15);
1265         WREG32(GRBM_SOFT_RESET, 0);
1266
1267         WREG32(CP_ME_RAM_WADDR, 0);
1268
1269         fw_data = (const __be32 *)rdev->me_fw->data;
1270         WREG32(CP_ME_RAM_WADDR, 0);
1271         for (i = 0; i < PM4_UCODE_SIZE * 3; i++)
1272                 WREG32(CP_ME_RAM_DATA,
1273                        be32_to_cpup(fw_data++));
1274
1275         fw_data = (const __be32 *)rdev->pfp_fw->data;
1276         WREG32(CP_PFP_UCODE_ADDR, 0);
1277         for (i = 0; i < PFP_UCODE_SIZE; i++)
1278                 WREG32(CP_PFP_UCODE_DATA,
1279                        be32_to_cpup(fw_data++));
1280
1281         WREG32(CP_PFP_UCODE_ADDR, 0);
1282         WREG32(CP_ME_RAM_WADDR, 0);
1283         WREG32(CP_ME_RAM_RADDR, 0);
1284         return 0;
1285 }
1286
1287 int r600_cp_start(struct radeon_device *rdev)
1288 {
1289         int r;
1290         uint32_t cp_me;
1291
1292         r = radeon_ring_lock(rdev, 7);
1293         if (r) {
1294                 DRM_ERROR("radeon: cp failed to lock ring (%d).\n", r);
1295                 return r;
1296         }
1297         radeon_ring_write(rdev, PACKET3(PACKET3_ME_INITIALIZE, 5));
1298         radeon_ring_write(rdev, 0x1);
1299         if (rdev->family < CHIP_RV770) {
1300                 radeon_ring_write(rdev, 0x3);
1301                 radeon_ring_write(rdev, rdev->config.r600.max_hw_contexts - 1);
1302         } else {
1303                 radeon_ring_write(rdev, 0x0);
1304                 radeon_ring_write(rdev, rdev->config.rv770.max_hw_contexts - 1);
1305         }
1306         radeon_ring_write(rdev, PACKET3_ME_INITIALIZE_DEVICE_ID(1));
1307         radeon_ring_write(rdev, 0);
1308         radeon_ring_write(rdev, 0);
1309         radeon_ring_unlock_commit(rdev);
1310
1311         cp_me = 0xff;
1312         WREG32(R_0086D8_CP_ME_CNTL, cp_me);
1313         return 0;
1314 }
1315
1316 int r600_cp_resume(struct radeon_device *rdev)
1317 {
1318         u32 tmp;
1319         u32 rb_bufsz;
1320         int r;
1321
1322         /* Reset cp */
1323         WREG32(GRBM_SOFT_RESET, SOFT_RESET_CP);
1324         RREG32(GRBM_SOFT_RESET);
1325         mdelay(15);
1326         WREG32(GRBM_SOFT_RESET, 0);
1327
1328         /* Set ring buffer size */
1329         rb_bufsz = drm_order(rdev->cp.ring_size / 8);
1330         tmp = RB_NO_UPDATE | (drm_order(RADEON_GPU_PAGE_SIZE/8) << 8) | rb_bufsz;
1331 #ifdef __BIG_ENDIAN
1332         tmp |= BUF_SWAP_32BIT;
1333 #endif
1334         WREG32(CP_RB_CNTL, tmp);
1335         WREG32(CP_SEM_WAIT_TIMER, 0x4);
1336
1337         /* Set the write pointer delay */
1338         WREG32(CP_RB_WPTR_DELAY, 0);
1339
1340         /* Initialize the ring buffer's read and write pointers */
1341         WREG32(CP_RB_CNTL, tmp | RB_RPTR_WR_ENA);
1342         WREG32(CP_RB_RPTR_WR, 0);
1343         WREG32(CP_RB_WPTR, 0);
1344         WREG32(CP_RB_RPTR_ADDR, rdev->cp.gpu_addr & 0xFFFFFFFF);
1345         WREG32(CP_RB_RPTR_ADDR_HI, upper_32_bits(rdev->cp.gpu_addr));
1346         mdelay(1);
1347         WREG32(CP_RB_CNTL, tmp);
1348
1349         WREG32(CP_RB_BASE, rdev->cp.gpu_addr >> 8);
1350         WREG32(CP_DEBUG, (1 << 27) | (1 << 28));
1351
1352         rdev->cp.rptr = RREG32(CP_RB_RPTR);
1353         rdev->cp.wptr = RREG32(CP_RB_WPTR);
1354
1355         r600_cp_start(rdev);
1356         rdev->cp.ready = true;
1357         r = radeon_ring_test(rdev);
1358         if (r) {
1359                 rdev->cp.ready = false;
1360                 return r;
1361         }
1362         return 0;
1363 }
1364
1365 void r600_cp_commit(struct radeon_device *rdev)
1366 {
1367         WREG32(CP_RB_WPTR, rdev->cp.wptr);
1368         (void)RREG32(CP_RB_WPTR);
1369 }
1370
1371 void r600_ring_init(struct radeon_device *rdev, unsigned ring_size)
1372 {
1373         u32 rb_bufsz;
1374
1375         /* Align ring size */
1376         rb_bufsz = drm_order(ring_size / 8);
1377         ring_size = (1 << (rb_bufsz + 1)) * 4;
1378         rdev->cp.ring_size = ring_size;
1379         rdev->cp.align_mask = 16 - 1;
1380 }
1381
1382
1383 /*
1384  * GPU scratch registers helpers function.
1385  */
1386 void r600_scratch_init(struct radeon_device *rdev)
1387 {
1388         int i;
1389
1390         rdev->scratch.num_reg = 7;
1391         for (i = 0; i < rdev->scratch.num_reg; i++) {
1392                 rdev->scratch.free[i] = true;
1393                 rdev->scratch.reg[i] = SCRATCH_REG0 + (i * 4);
1394         }
1395 }
1396
1397 int r600_ring_test(struct radeon_device *rdev)
1398 {
1399         uint32_t scratch;
1400         uint32_t tmp = 0;
1401         unsigned i;
1402         int r;
1403
1404         r = radeon_scratch_get(rdev, &scratch);
1405         if (r) {
1406                 DRM_ERROR("radeon: cp failed to get scratch reg (%d).\n", r);
1407                 return r;
1408         }
1409         WREG32(scratch, 0xCAFEDEAD);
1410         r = radeon_ring_lock(rdev, 3);
1411         if (r) {
1412                 DRM_ERROR("radeon: cp failed to lock ring (%d).\n", r);
1413                 radeon_scratch_free(rdev, scratch);
1414                 return r;
1415         }
1416         radeon_ring_write(rdev, PACKET3(PACKET3_SET_CONFIG_REG, 1));
1417         radeon_ring_write(rdev, ((scratch - PACKET3_SET_CONFIG_REG_OFFSET) >> 2));
1418         radeon_ring_write(rdev, 0xDEADBEEF);
1419         radeon_ring_unlock_commit(rdev);
1420         for (i = 0; i < rdev->usec_timeout; i++) {
1421                 tmp = RREG32(scratch);
1422                 if (tmp == 0xDEADBEEF)
1423                         break;
1424                 DRM_UDELAY(1);
1425         }
1426         if (i < rdev->usec_timeout) {
1427                 DRM_INFO("ring test succeeded in %d usecs\n", i);
1428         } else {
1429                 DRM_ERROR("radeon: ring test failed (scratch(0x%04X)=0x%08X)\n",
1430                           scratch, tmp);
1431                 r = -EINVAL;
1432         }
1433         radeon_scratch_free(rdev, scratch);
1434         return r;
1435 }
1436
1437 void r600_wb_disable(struct radeon_device *rdev)
1438 {
1439         WREG32(SCRATCH_UMSK, 0);
1440         if (rdev->wb.wb_obj) {
1441                 radeon_object_kunmap(rdev->wb.wb_obj);
1442                 radeon_object_unpin(rdev->wb.wb_obj);
1443         }
1444 }
1445
1446 void r600_wb_fini(struct radeon_device *rdev)
1447 {
1448         r600_wb_disable(rdev);
1449         if (rdev->wb.wb_obj) {
1450                 radeon_object_unref(&rdev->wb.wb_obj);
1451                 rdev->wb.wb = NULL;
1452                 rdev->wb.wb_obj = NULL;
1453         }
1454 }
1455
1456 int r600_wb_enable(struct radeon_device *rdev)
1457 {
1458         int r;
1459
1460         if (rdev->wb.wb_obj == NULL) {
1461                 r = radeon_object_create(rdev, NULL, RADEON_GPU_PAGE_SIZE, true,
1462                                 RADEON_GEM_DOMAIN_GTT, false, &rdev->wb.wb_obj);
1463                 if (r) {
1464                         dev_warn(rdev->dev, "failed to create WB buffer (%d).\n", r);
1465                         return r;
1466                 }
1467                 r = radeon_object_pin(rdev->wb.wb_obj, RADEON_GEM_DOMAIN_GTT,
1468                                 &rdev->wb.gpu_addr);
1469                 if (r) {
1470                         dev_warn(rdev->dev, "failed to pin WB buffer (%d).\n", r);
1471                         r600_wb_fini(rdev);
1472                         return r;
1473                 }
1474                 r = radeon_object_kmap(rdev->wb.wb_obj, (void **)&rdev->wb.wb);
1475                 if (r) {
1476                         dev_warn(rdev->dev, "failed to map WB buffer (%d).\n", r);
1477                         r600_wb_fini(rdev);
1478                         return r;
1479                 }
1480         }
1481         WREG32(SCRATCH_ADDR, (rdev->wb.gpu_addr >> 8) & 0xFFFFFFFF);
1482         WREG32(CP_RB_RPTR_ADDR, (rdev->wb.gpu_addr + 1024) & 0xFFFFFFFC);
1483         WREG32(CP_RB_RPTR_ADDR_HI, upper_32_bits(rdev->wb.gpu_addr + 1024) & 0xFF);
1484         WREG32(SCRATCH_UMSK, 0xff);
1485         return 0;
1486 }
1487
1488 void r600_fence_ring_emit(struct radeon_device *rdev,
1489                           struct radeon_fence *fence)
1490 {
1491         /* Also consider EVENT_WRITE_EOP.  it handles the interrupts + timestamps + events */
1492         /* Emit fence sequence & fire IRQ */
1493         radeon_ring_write(rdev, PACKET3(PACKET3_SET_CONFIG_REG, 1));
1494         radeon_ring_write(rdev, ((rdev->fence_drv.scratch_reg - PACKET3_SET_CONFIG_REG_OFFSET) >> 2));
1495         radeon_ring_write(rdev, fence->seq);
1496         /* CP_INTERRUPT packet 3 no longer exists, use packet 0 */
1497         radeon_ring_write(rdev, PACKET0(CP_INT_STATUS, 0));
1498         radeon_ring_write(rdev, RB_INT_STAT);
1499 }
1500
1501 int r600_copy_dma(struct radeon_device *rdev,
1502                   uint64_t src_offset,
1503                   uint64_t dst_offset,
1504                   unsigned num_pages,
1505                   struct radeon_fence *fence)
1506 {
1507         /* FIXME: implement */
1508         return 0;
1509 }
1510
1511 int r600_copy_blit(struct radeon_device *rdev,
1512                    uint64_t src_offset, uint64_t dst_offset,
1513                    unsigned num_pages, struct radeon_fence *fence)
1514 {
1515         r600_blit_prepare_copy(rdev, num_pages * RADEON_GPU_PAGE_SIZE);
1516         r600_kms_blit_copy(rdev, src_offset, dst_offset, num_pages * RADEON_GPU_PAGE_SIZE);
1517         r600_blit_done_copy(rdev, fence);
1518         return 0;
1519 }
1520
1521 int r600_set_surface_reg(struct radeon_device *rdev, int reg,
1522                          uint32_t tiling_flags, uint32_t pitch,
1523                          uint32_t offset, uint32_t obj_size)
1524 {
1525         /* FIXME: implement */
1526         return 0;
1527 }
1528
1529 void r600_clear_surface_reg(struct radeon_device *rdev, int reg)
1530 {
1531         /* FIXME: implement */
1532 }
1533
1534
1535 bool r600_card_posted(struct radeon_device *rdev)
1536 {
1537         uint32_t reg;
1538
1539         /* first check CRTCs */
1540         reg = RREG32(D1CRTC_CONTROL) |
1541                 RREG32(D2CRTC_CONTROL);
1542         if (reg & CRTC_EN)
1543                 return true;
1544
1545         /* then check MEM_SIZE, in case the crtcs are off */
1546         if (RREG32(CONFIG_MEMSIZE))
1547                 return true;
1548
1549         return false;
1550 }
1551
1552 int r600_startup(struct radeon_device *rdev)
1553 {
1554         int r;
1555
1556         r600_mc_program(rdev);
1557         if (rdev->flags & RADEON_IS_AGP) {
1558                 r600_agp_enable(rdev);
1559         } else {
1560                 r = r600_pcie_gart_enable(rdev);
1561                 if (r)
1562                         return r;
1563         }
1564         r600_gpu_init(rdev);
1565
1566         r = radeon_object_pin(rdev->r600_blit.shader_obj, RADEON_GEM_DOMAIN_VRAM,
1567                               &rdev->r600_blit.shader_gpu_addr);
1568         if (r) {
1569                 DRM_ERROR("failed to pin blit object %d\n", r);
1570                 return r;
1571         }
1572
1573         /* Enable IRQ */
1574         rdev->irq.sw_int = true;
1575         r = r600_irq_init(rdev);
1576         if (r) {
1577                 DRM_ERROR("radeon: IH init failed (%d).\n", r);
1578                 radeon_irq_kms_fini(rdev);
1579                 return r;
1580         }
1581         r600_irq_set(rdev);
1582
1583         r = radeon_ring_init(rdev, rdev->cp.ring_size);
1584         if (r)
1585                 return r;
1586         r = r600_cp_load_microcode(rdev);
1587         if (r)
1588                 return r;
1589         r = r600_cp_resume(rdev);
1590         if (r)
1591                 return r;
1592         /* write back buffer are not vital so don't worry about failure */
1593         r600_wb_enable(rdev);
1594         return 0;
1595 }
1596
1597 void r600_vga_set_state(struct radeon_device *rdev, bool state)
1598 {
1599         uint32_t temp;
1600
1601         temp = RREG32(CONFIG_CNTL);
1602         if (state == false) {
1603                 temp &= ~(1<<0);
1604                 temp |= (1<<1);
1605         } else {
1606                 temp &= ~(1<<1);
1607         }
1608         WREG32(CONFIG_CNTL, temp);
1609 }
1610
1611 int r600_resume(struct radeon_device *rdev)
1612 {
1613         int r;
1614
1615         /* Do not reset GPU before posting, on r600 hw unlike on r500 hw,
1616          * posting will perform necessary task to bring back GPU into good
1617          * shape.
1618          */
1619         /* post card */
1620         atom_asic_init(rdev->mode_info.atom_context);
1621         /* Initialize clocks */
1622         r = radeon_clocks_init(rdev);
1623         if (r) {
1624                 return r;
1625         }
1626
1627         r = r600_startup(rdev);
1628         if (r) {
1629                 DRM_ERROR("r600 startup failed on resume\n");
1630                 return r;
1631         }
1632
1633         r = r600_ib_test(rdev);
1634         if (r) {
1635                 DRM_ERROR("radeon: failled testing IB (%d).\n", r);
1636                 return r;
1637         }
1638         return r;
1639 }
1640
1641 int r600_suspend(struct radeon_device *rdev)
1642 {
1643         /* FIXME: we should wait for ring to be empty */
1644         r600_cp_stop(rdev);
1645         rdev->cp.ready = false;
1646         r600_wb_disable(rdev);
1647         r600_pcie_gart_disable(rdev);
1648         /* unpin shaders bo */
1649         radeon_object_unpin(rdev->r600_blit.shader_obj);
1650         return 0;
1651 }
1652
1653 /* Plan is to move initialization in that function and use
1654  * helper function so that radeon_device_init pretty much
1655  * do nothing more than calling asic specific function. This
1656  * should also allow to remove a bunch of callback function
1657  * like vram_info.
1658  */
1659 int r600_init(struct radeon_device *rdev)
1660 {
1661         int r;
1662
1663         r = radeon_dummy_page_init(rdev);
1664         if (r)
1665                 return r;
1666         if (r600_debugfs_mc_info_init(rdev)) {
1667                 DRM_ERROR("Failed to register debugfs file for mc !\n");
1668         }
1669         /* This don't do much */
1670         r = radeon_gem_init(rdev);
1671         if (r)
1672                 return r;
1673         /* Read BIOS */
1674         if (!radeon_get_bios(rdev)) {
1675                 if (ASIC_IS_AVIVO(rdev))
1676                         return -EINVAL;
1677         }
1678         /* Must be an ATOMBIOS */
1679         if (!rdev->is_atom_bios) {
1680                 dev_err(rdev->dev, "Expecting atombios for R600 GPU\n");
1681                 return -EINVAL;
1682         }
1683         r = radeon_atombios_init(rdev);
1684         if (r)
1685                 return r;
1686         /* Post card if necessary */
1687         if (!r600_card_posted(rdev)) {
1688                 if (!rdev->bios) {
1689                         dev_err(rdev->dev, "Card not posted and no BIOS - ignoring\n");
1690                         return -EINVAL;
1691                 }
1692                 DRM_INFO("GPU not posted. posting now...\n");
1693                 atom_asic_init(rdev->mode_info.atom_context);
1694         }
1695         /* Initialize scratch registers */
1696         r600_scratch_init(rdev);
1697         /* Initialize surface registers */
1698         radeon_surface_init(rdev);
1699         /* Initialize clocks */
1700         radeon_get_clock_info(rdev->ddev);
1701         r = radeon_clocks_init(rdev);
1702         if (r)
1703                 return r;
1704         /* Initialize power management */
1705         radeon_pm_init(rdev);
1706         /* Fence driver */
1707         r = radeon_fence_driver_init(rdev);
1708         if (r)
1709                 return r;
1710         r = r600_mc_init(rdev);
1711         if (r)
1712                 return r;
1713         /* Memory manager */
1714         r = radeon_object_init(rdev);
1715         if (r)
1716                 return r;
1717
1718         r = radeon_irq_kms_init(rdev);
1719         if (r)
1720                 return r;
1721
1722         rdev->cp.ring_obj = NULL;
1723         r600_ring_init(rdev, 1024 * 1024);
1724
1725         rdev->ih.ring_obj = NULL;
1726         r600_ih_ring_init(rdev, 64 * 1024);
1727
1728         if (!rdev->me_fw || !rdev->pfp_fw || !rdev->rlc_fw) {
1729                 r = r600_init_microcode(rdev);
1730                 if (r) {
1731                         DRM_ERROR("Failed to load firmware!\n");
1732                         return r;
1733                 }
1734         }
1735
1736         r = r600_pcie_gart_init(rdev);
1737         if (r)
1738                 return r;
1739
1740         rdev->accel_working = true;
1741         r = r600_blit_init(rdev);
1742         if (r) {
1743                 DRM_ERROR("radeon: failled blitter (%d).\n", r);
1744                 return r;
1745         }
1746
1747         r = r600_startup(rdev);
1748         if (r) {
1749                 r600_suspend(rdev);
1750                 r600_wb_fini(rdev);
1751                 radeon_ring_fini(rdev);
1752                 r600_pcie_gart_fini(rdev);
1753                 rdev->accel_working = false;
1754         }
1755         if (rdev->accel_working) {
1756                 r = radeon_ib_pool_init(rdev);
1757                 if (r) {
1758                         DRM_ERROR("radeon: failled initializing IB pool (%d).\n", r);
1759                         rdev->accel_working = false;
1760                 }
1761                 r = r600_ib_test(rdev);
1762                 if (r) {
1763                         DRM_ERROR("radeon: failled testing IB (%d).\n", r);
1764                         rdev->accel_working = false;
1765                 }
1766         }
1767         return 0;
1768 }
1769
1770 void r600_fini(struct radeon_device *rdev)
1771 {
1772         /* Suspend operations */
1773         r600_suspend(rdev);
1774
1775         r600_blit_fini(rdev);
1776         r600_irq_fini(rdev);
1777         radeon_irq_kms_fini(rdev);
1778         radeon_ring_fini(rdev);
1779         r600_wb_fini(rdev);
1780         r600_pcie_gart_fini(rdev);
1781         radeon_gem_fini(rdev);
1782         radeon_fence_driver_fini(rdev);
1783         radeon_clocks_fini(rdev);
1784         if (rdev->flags & RADEON_IS_AGP)
1785                 radeon_agp_fini(rdev);
1786         radeon_object_fini(rdev);
1787         radeon_atombios_fini(rdev);
1788         kfree(rdev->bios);
1789         rdev->bios = NULL;
1790         radeon_dummy_page_fini(rdev);
1791 }
1792
1793
1794 /*
1795  * CS stuff
1796  */
1797 void r600_ring_ib_execute(struct radeon_device *rdev, struct radeon_ib *ib)
1798 {
1799         /* FIXME: implement */
1800         radeon_ring_write(rdev, PACKET3(PACKET3_INDIRECT_BUFFER, 2));
1801         radeon_ring_write(rdev, ib->gpu_addr & 0xFFFFFFFC);
1802         radeon_ring_write(rdev, upper_32_bits(ib->gpu_addr) & 0xFF);
1803         radeon_ring_write(rdev, ib->length_dw);
1804 }
1805
1806 int r600_ib_test(struct radeon_device *rdev)
1807 {
1808         struct radeon_ib *ib;
1809         uint32_t scratch;
1810         uint32_t tmp = 0;
1811         unsigned i;
1812         int r;
1813
1814         r = radeon_scratch_get(rdev, &scratch);
1815         if (r) {
1816                 DRM_ERROR("radeon: failed to get scratch reg (%d).\n", r);
1817                 return r;
1818         }
1819         WREG32(scratch, 0xCAFEDEAD);
1820         r = radeon_ib_get(rdev, &ib);
1821         if (r) {
1822                 DRM_ERROR("radeon: failed to get ib (%d).\n", r);
1823                 return r;
1824         }
1825         ib->ptr[0] = PACKET3(PACKET3_SET_CONFIG_REG, 1);
1826         ib->ptr[1] = ((scratch - PACKET3_SET_CONFIG_REG_OFFSET) >> 2);
1827         ib->ptr[2] = 0xDEADBEEF;
1828         ib->ptr[3] = PACKET2(0);
1829         ib->ptr[4] = PACKET2(0);
1830         ib->ptr[5] = PACKET2(0);
1831         ib->ptr[6] = PACKET2(0);
1832         ib->ptr[7] = PACKET2(0);
1833         ib->ptr[8] = PACKET2(0);
1834         ib->ptr[9] = PACKET2(0);
1835         ib->ptr[10] = PACKET2(0);
1836         ib->ptr[11] = PACKET2(0);
1837         ib->ptr[12] = PACKET2(0);
1838         ib->ptr[13] = PACKET2(0);
1839         ib->ptr[14] = PACKET2(0);
1840         ib->ptr[15] = PACKET2(0);
1841         ib->length_dw = 16;
1842         r = radeon_ib_schedule(rdev, ib);
1843         if (r) {
1844                 radeon_scratch_free(rdev, scratch);
1845                 radeon_ib_free(rdev, &ib);
1846                 DRM_ERROR("radeon: failed to schedule ib (%d).\n", r);
1847                 return r;
1848         }
1849         r = radeon_fence_wait(ib->fence, false);
1850         if (r) {
1851                 DRM_ERROR("radeon: fence wait failed (%d).\n", r);
1852                 return r;
1853         }
1854         for (i = 0; i < rdev->usec_timeout; i++) {
1855                 tmp = RREG32(scratch);
1856                 if (tmp == 0xDEADBEEF)
1857                         break;
1858                 DRM_UDELAY(1);
1859         }
1860         if (i < rdev->usec_timeout) {
1861                 DRM_INFO("ib test succeeded in %u usecs\n", i);
1862         } else {
1863                 DRM_ERROR("radeon: ib test failed (sracth(0x%04X)=0x%08X)\n",
1864                           scratch, tmp);
1865                 r = -EINVAL;
1866         }
1867         radeon_scratch_free(rdev, scratch);
1868         radeon_ib_free(rdev, &ib);
1869         return r;
1870 }
1871
1872 /*
1873  * Interrupts
1874  *
1875  * Interrupts use a ring buffer on r6xx/r7xx hardware.  It works pretty
1876  * the same as the CP ring buffer, but in reverse.  Rather than the CPU
1877  * writing to the ring and the GPU consuming, the GPU writes to the ring
1878  * and host consumes.  As the host irq handler processes interrupts, it
1879  * increments the rptr.  When the rptr catches up with the wptr, all the
1880  * current interrupts have been processed.
1881  */
1882
1883 void r600_ih_ring_init(struct radeon_device *rdev, unsigned ring_size)
1884 {
1885         u32 rb_bufsz;
1886
1887         /* Align ring size */
1888         rb_bufsz = drm_order(ring_size / 4);
1889         ring_size = (1 << rb_bufsz) * 4;
1890         rdev->ih.ring_size = ring_size;
1891         rdev->ih.align_mask = 4 - 1;
1892 }
1893
1894 static int r600_ih_ring_alloc(struct radeon_device *rdev, unsigned ring_size)
1895 {
1896         int r;
1897
1898         rdev->ih.ring_size = ring_size;
1899         /* Allocate ring buffer */
1900         if (rdev->ih.ring_obj == NULL) {
1901                 r = radeon_object_create(rdev, NULL, rdev->ih.ring_size,
1902                                          true,
1903                                          RADEON_GEM_DOMAIN_GTT,
1904                                          false,
1905                                          &rdev->ih.ring_obj);
1906                 if (r) {
1907                         DRM_ERROR("radeon: failed to create ih ring buffer (%d).\n", r);
1908                         return r;
1909                 }
1910                 r = radeon_object_pin(rdev->ih.ring_obj,
1911                                       RADEON_GEM_DOMAIN_GTT,
1912                                       &rdev->ih.gpu_addr);
1913                 if (r) {
1914                         DRM_ERROR("radeon: failed to pin ih ring buffer (%d).\n", r);
1915                         return r;
1916                 }
1917                 r = radeon_object_kmap(rdev->ih.ring_obj,
1918                                        (void **)&rdev->ih.ring);
1919                 if (r) {
1920                         DRM_ERROR("radeon: failed to map ih ring buffer (%d).\n", r);
1921                         return r;
1922                 }
1923         }
1924         rdev->ih.ptr_mask = (rdev->cp.ring_size / 4) - 1;
1925         rdev->ih.rptr = 0;
1926
1927         return 0;
1928 }
1929
1930 static void r600_ih_ring_fini(struct radeon_device *rdev)
1931 {
1932         if (rdev->ih.ring_obj) {
1933                 radeon_object_kunmap(rdev->ih.ring_obj);
1934                 radeon_object_unpin(rdev->ih.ring_obj);
1935                 radeon_object_unref(&rdev->ih.ring_obj);
1936                 rdev->ih.ring = NULL;
1937                 rdev->ih.ring_obj = NULL;
1938         }
1939 }
1940
1941 static void r600_rlc_stop(struct radeon_device *rdev)
1942 {
1943
1944         if (rdev->family >= CHIP_RV770) {
1945                 /* r7xx asics need to soft reset RLC before halting */
1946                 WREG32(SRBM_SOFT_RESET, SOFT_RESET_RLC);
1947                 RREG32(SRBM_SOFT_RESET);
1948                 udelay(15000);
1949                 WREG32(SRBM_SOFT_RESET, 0);
1950                 RREG32(SRBM_SOFT_RESET);
1951         }
1952
1953         WREG32(RLC_CNTL, 0);
1954 }
1955
1956 static void r600_rlc_start(struct radeon_device *rdev)
1957 {
1958         WREG32(RLC_CNTL, RLC_ENABLE);
1959 }
1960
1961 static int r600_rlc_init(struct radeon_device *rdev)
1962 {
1963         u32 i;
1964         const __be32 *fw_data;
1965
1966         if (!rdev->rlc_fw)
1967                 return -EINVAL;
1968
1969         r600_rlc_stop(rdev);
1970
1971         WREG32(RLC_HB_BASE, 0);
1972         WREG32(RLC_HB_CNTL, 0);
1973         WREG32(RLC_HB_RPTR, 0);
1974         WREG32(RLC_HB_WPTR, 0);
1975         WREG32(RLC_HB_WPTR_LSB_ADDR, 0);
1976         WREG32(RLC_HB_WPTR_MSB_ADDR, 0);
1977         WREG32(RLC_MC_CNTL, 0);
1978         WREG32(RLC_UCODE_CNTL, 0);
1979
1980         fw_data = (const __be32 *)rdev->rlc_fw->data;
1981         if (rdev->family >= CHIP_RV770) {
1982                 for (i = 0; i < R700_RLC_UCODE_SIZE; i++) {
1983                         WREG32(RLC_UCODE_ADDR, i);
1984                         WREG32(RLC_UCODE_DATA, be32_to_cpup(fw_data++));
1985                 }
1986         } else {
1987                 for (i = 0; i < RLC_UCODE_SIZE; i++) {
1988                         WREG32(RLC_UCODE_ADDR, i);
1989                         WREG32(RLC_UCODE_DATA, be32_to_cpup(fw_data++));
1990                 }
1991         }
1992         WREG32(RLC_UCODE_ADDR, 0);
1993
1994         r600_rlc_start(rdev);
1995
1996         return 0;
1997 }
1998
1999 static void r600_enable_interrupts(struct radeon_device *rdev)
2000 {
2001         u32 ih_cntl = RREG32(IH_CNTL);
2002         u32 ih_rb_cntl = RREG32(IH_RB_CNTL);
2003
2004         ih_cntl |= ENABLE_INTR;
2005         ih_rb_cntl |= IH_RB_ENABLE;
2006         WREG32(IH_CNTL, ih_cntl);
2007         WREG32(IH_RB_CNTL, ih_rb_cntl);
2008         rdev->ih.enabled = true;
2009 }
2010
2011 static void r600_disable_interrupts(struct radeon_device *rdev)
2012 {
2013         u32 ih_rb_cntl = RREG32(IH_RB_CNTL);
2014         u32 ih_cntl = RREG32(IH_CNTL);
2015
2016         ih_rb_cntl &= ~IH_RB_ENABLE;
2017         ih_cntl &= ~ENABLE_INTR;
2018         WREG32(IH_RB_CNTL, ih_rb_cntl);
2019         WREG32(IH_CNTL, ih_cntl);
2020         /* set rptr, wptr to 0 */
2021         WREG32(IH_RB_RPTR, 0);
2022         WREG32(IH_RB_WPTR, 0);
2023         rdev->ih.enabled = false;
2024         rdev->ih.wptr = 0;
2025         rdev->ih.rptr = 0;
2026 }
2027
2028 int r600_irq_init(struct radeon_device *rdev)
2029 {
2030         int ret = 0;
2031         int rb_bufsz;
2032         u32 interrupt_cntl, ih_cntl, ih_rb_cntl;
2033
2034         /* allocate ring */
2035         ret = r600_ih_ring_alloc(rdev, rdev->ih.ring_size);
2036         if (ret)
2037                 return ret;
2038
2039         /* disable irqs */
2040         r600_disable_interrupts(rdev);
2041
2042         /* init rlc */
2043         ret = r600_rlc_init(rdev);
2044         if (ret) {
2045                 r600_ih_ring_fini(rdev);
2046                 return ret;
2047         }
2048
2049         /* setup interrupt control */
2050         /* set dummy read address to ring address */
2051         WREG32(INTERRUPT_CNTL2, rdev->ih.gpu_addr >> 8);
2052         interrupt_cntl = RREG32(INTERRUPT_CNTL);
2053         /* IH_DUMMY_RD_OVERRIDE=0 - dummy read disabled with msi, enabled without msi
2054          * IH_DUMMY_RD_OVERRIDE=1 - dummy read controlled by IH_DUMMY_RD_EN
2055          */
2056         interrupt_cntl &= ~IH_DUMMY_RD_OVERRIDE;
2057         /* IH_REQ_NONSNOOP_EN=1 if ring is in non-cacheable memory, e.g., vram */
2058         interrupt_cntl &= ~IH_REQ_NONSNOOP_EN;
2059         WREG32(INTERRUPT_CNTL, interrupt_cntl);
2060
2061         WREG32(IH_RB_BASE, rdev->ih.gpu_addr >> 8);
2062         rb_bufsz = drm_order(rdev->ih.ring_size / 4);
2063
2064         ih_rb_cntl = (IH_WPTR_OVERFLOW_ENABLE |
2065                       IH_WPTR_OVERFLOW_CLEAR |
2066                       (rb_bufsz << 1));
2067         /* WPTR writeback, not yet */
2068         /*ih_rb_cntl |= IH_WPTR_WRITEBACK_ENABLE;*/
2069         WREG32(IH_RB_WPTR_ADDR_LO, 0);
2070         WREG32(IH_RB_WPTR_ADDR_HI, 0);
2071
2072         WREG32(IH_RB_CNTL, ih_rb_cntl);
2073
2074         /* set rptr, wptr to 0 */
2075         WREG32(IH_RB_RPTR, 0);
2076         WREG32(IH_RB_WPTR, 0);
2077
2078         /* Default settings for IH_CNTL (disabled at first) */
2079         ih_cntl = MC_WRREQ_CREDIT(0x10) | MC_WR_CLEAN_CNT(0x10);
2080         /* RPTR_REARM only works if msi's are enabled */
2081         if (rdev->msi_enabled)
2082                 ih_cntl |= RPTR_REARM;
2083
2084 #ifdef __BIG_ENDIAN
2085         ih_cntl |= IH_MC_SWAP(IH_MC_SWAP_32BIT);
2086 #endif
2087         WREG32(IH_CNTL, ih_cntl);
2088
2089         /* force the active interrupt state to all disabled */
2090         WREG32(CP_INT_CNTL, 0);
2091         WREG32(GRBM_INT_CNTL, 0);
2092         WREG32(DxMODE_INT_MASK, 0);
2093
2094         /* enable irqs */
2095         r600_enable_interrupts(rdev);
2096
2097         return ret;
2098 }
2099
2100 void r600_irq_fini(struct radeon_device *rdev)
2101 {
2102         r600_disable_interrupts(rdev);
2103         r600_rlc_stop(rdev);
2104         r600_ih_ring_fini(rdev);
2105 }
2106
2107 int r600_irq_set(struct radeon_device *rdev)
2108 {
2109         uint32_t cp_int_cntl = CNTX_BUSY_INT_ENABLE | CNTX_EMPTY_INT_ENABLE;
2110         uint32_t mode_int = 0;
2111
2112         /* don't enable anything if the ih is disabled */
2113         if (!rdev->ih.enabled)
2114                 return 0;
2115
2116         if (rdev->irq.sw_int) {
2117                 DRM_DEBUG("r600_irq_set: sw int\n");
2118                 cp_int_cntl |= RB_INT_ENABLE;
2119         }
2120         if (rdev->irq.crtc_vblank_int[0]) {
2121                 DRM_DEBUG("r600_irq_set: vblank 0\n");
2122                 mode_int |= D1MODE_VBLANK_INT_MASK;
2123         }
2124         if (rdev->irq.crtc_vblank_int[1]) {
2125                 DRM_DEBUG("r600_irq_set: vblank 1\n");
2126                 mode_int |= D2MODE_VBLANK_INT_MASK;
2127         }
2128
2129         WREG32(CP_INT_CNTL, cp_int_cntl);
2130         WREG32(DxMODE_INT_MASK, mode_int);
2131
2132         return 0;
2133 }
2134
2135 static inline void r600_irq_ack(struct radeon_device *rdev, u32 disp_int)
2136 {
2137
2138         if (disp_int & LB_D1_VBLANK_INTERRUPT)
2139                 WREG32(D1MODE_VBLANK_STATUS, DxMODE_VBLANK_ACK);
2140         if (disp_int & LB_D1_VLINE_INTERRUPT)
2141                 WREG32(D1MODE_VLINE_STATUS, DxMODE_VLINE_ACK);
2142         if (disp_int & LB_D2_VBLANK_INTERRUPT)
2143                 WREG32(D2MODE_VBLANK_STATUS, DxMODE_VBLANK_ACK);
2144         if (disp_int & LB_D2_VLINE_INTERRUPT)
2145                 WREG32(D2MODE_VLINE_STATUS, DxMODE_VLINE_ACK);
2146
2147 }
2148
2149 void r600_irq_disable(struct radeon_device *rdev)
2150 {
2151         u32 disp_int;
2152
2153         r600_disable_interrupts(rdev);
2154         /* Wait and acknowledge irq */
2155         mdelay(1);
2156         if (ASIC_IS_DCE3(rdev))
2157                 disp_int = RREG32(DCE3_DISP_INTERRUPT_STATUS);
2158         else
2159                 disp_int = RREG32(DISP_INTERRUPT_STATUS);
2160         r600_irq_ack(rdev, disp_int);
2161 }
2162
2163 static inline u32 r600_get_ih_wptr(struct radeon_device *rdev)
2164 {
2165         u32 wptr, tmp;
2166
2167         /* XXX use writeback */
2168         wptr = RREG32(IH_RB_WPTR);
2169
2170         if (wptr & RB_OVERFLOW) {
2171                 WARN_ON(1);
2172                 /* XXX deal with overflow */
2173                 DRM_ERROR("IH RB overflow\n");
2174                 tmp = RREG32(IH_RB_CNTL);
2175                 tmp |= IH_WPTR_OVERFLOW_CLEAR;
2176                 WREG32(IH_RB_CNTL, tmp);
2177         }
2178         wptr = wptr & WPTR_OFFSET_MASK;
2179
2180         return wptr;
2181 }
2182
2183 /*        r600 IV Ring
2184  * Each IV ring entry is 128 bits:
2185  * [7:0]    - interrupt source id
2186  * [31:8]   - reserved
2187  * [59:32]  - interrupt source data
2188  * [127:60]  - reserved
2189  *
2190  * The basic interrupt vector entries
2191  * are decoded as follows:
2192  * src_id  src_data  description
2193  *      1         0  D1 Vblank
2194  *      1         1  D1 Vline
2195  *      5         0  D2 Vblank
2196  *      5         1  D2 Vline
2197  *     19         0  FP Hot plug detection A
2198  *     19         1  FP Hot plug detection B
2199  *     19         2  DAC A auto-detection
2200  *     19         3  DAC B auto-detection
2201  *    176         -  CP_INT RB
2202  *    177         -  CP_INT IB1
2203  *    178         -  CP_INT IB2
2204  *    181         -  EOP Interrupt
2205  *    233         -  GUI Idle
2206  *
2207  * Note, these are based on r600 and may need to be
2208  * adjusted or added to on newer asics
2209  */
2210
2211 int r600_irq_process(struct radeon_device *rdev)
2212 {
2213         u32 wptr = r600_get_ih_wptr(rdev);
2214         u32 rptr = rdev->ih.rptr;
2215         u32 src_id, src_data;
2216         u32 last_entry = rdev->ih.ring_size - 16;
2217         u32 ring_index, disp_int;
2218         unsigned long flags;
2219
2220         DRM_DEBUG("r600_irq_process start: rptr %d, wptr %d\n", rptr, wptr);
2221
2222         spin_lock_irqsave(&rdev->ih.lock, flags);
2223
2224         if (rptr == wptr) {
2225                 spin_unlock_irqrestore(&rdev->ih.lock, flags);
2226                 return IRQ_NONE;
2227         }
2228         if (rdev->shutdown) {
2229                 spin_unlock_irqrestore(&rdev->ih.lock, flags);
2230                 return IRQ_NONE;
2231         }
2232
2233 restart_ih:
2234         /* display interrupts */
2235         if (ASIC_IS_DCE3(rdev))
2236                 disp_int = RREG32(DCE3_DISP_INTERRUPT_STATUS);
2237         else
2238                 disp_int = RREG32(DISP_INTERRUPT_STATUS);
2239         r600_irq_ack(rdev, disp_int);
2240
2241         rdev->ih.wptr = wptr;
2242         while (rptr != wptr) {
2243                 /* wptr/rptr are in bytes! */
2244                 ring_index = rptr / 4;
2245                 src_id =  rdev->ih.ring[ring_index] & 0xff;
2246                 src_data = rdev->ih.ring[ring_index + 1] & 0xfffffff;
2247
2248                 switch (src_id) {
2249                 case 1: /* D1 vblank/vline */
2250                         switch (src_data) {
2251                         case 0: /* D1 vblank */
2252                                 if (disp_int & LB_D1_VBLANK_INTERRUPT) {
2253                                         drm_handle_vblank(rdev->ddev, 0);
2254                                         disp_int &= ~LB_D1_VBLANK_INTERRUPT;
2255                                         DRM_DEBUG("IH: D1 vblank\n");
2256                                 }
2257                                 break;
2258                         case 1: /* D1 vline */
2259                                 if (disp_int & LB_D1_VLINE_INTERRUPT) {
2260                                         disp_int &= ~LB_D1_VLINE_INTERRUPT;
2261                                         DRM_DEBUG("IH: D1 vline\n");
2262                                 }
2263                                 break;
2264                         default:
2265                                 DRM_ERROR("Unhandled interrupt: %d %d\n", src_id, src_data);
2266                                 break;
2267                         }
2268                         break;
2269                 case 5: /* D2 vblank/vline */
2270                         switch (src_data) {
2271                         case 0: /* D2 vblank */
2272                                 if (disp_int & LB_D2_VBLANK_INTERRUPT) {
2273                                         drm_handle_vblank(rdev->ddev, 1);
2274                                         disp_int &= ~LB_D2_VBLANK_INTERRUPT;
2275                                         DRM_DEBUG("IH: D2 vblank\n");
2276                                 }
2277                                 break;
2278                         case 1: /* D1 vline */
2279                                 if (disp_int & LB_D2_VLINE_INTERRUPT) {
2280                                         disp_int &= ~LB_D2_VLINE_INTERRUPT;
2281                                         DRM_DEBUG("IH: D2 vline\n");
2282                                 }
2283                                 break;
2284                         default:
2285                                 DRM_ERROR("Unhandled interrupt: %d %d\n", src_id, src_data);
2286                                 break;
2287                         }
2288                         break;
2289                 case 176: /* CP_INT in ring buffer */
2290                 case 177: /* CP_INT in IB1 */
2291                 case 178: /* CP_INT in IB2 */
2292                         DRM_DEBUG("IH: CP int: 0x%08x\n", src_data);
2293                         radeon_fence_process(rdev);
2294                         break;
2295                 case 181: /* CP EOP event */
2296                         DRM_DEBUG("IH: CP EOP\n");
2297                         break;
2298                 default:
2299                         DRM_ERROR("Unhandled interrupt: %d %d\n", src_id, src_data);
2300                         break;
2301                 }
2302
2303                 /* wptr/rptr are in bytes! */
2304                 if (rptr == last_entry)
2305                         rptr = 0;
2306                 else
2307                         rptr += 16;
2308         }
2309         /* make sure wptr hasn't changed while processing */
2310         wptr = r600_get_ih_wptr(rdev);
2311         if (wptr != rdev->ih.wptr)
2312                 goto restart_ih;
2313         rdev->ih.rptr = rptr;
2314         WREG32(IH_RB_RPTR, rdev->ih.rptr);
2315         spin_unlock_irqrestore(&rdev->ih.lock, flags);
2316         return IRQ_HANDLED;
2317 }
2318
2319 /*
2320  * Debugfs info
2321  */
2322 #if defined(CONFIG_DEBUG_FS)
2323
2324 static int r600_debugfs_cp_ring_info(struct seq_file *m, void *data)
2325 {
2326         struct drm_info_node *node = (struct drm_info_node *) m->private;
2327         struct drm_device *dev = node->minor->dev;
2328         struct radeon_device *rdev = dev->dev_private;
2329         uint32_t rdp, wdp;
2330         unsigned count, i, j;
2331
2332         radeon_ring_free_size(rdev);
2333         rdp = RREG32(CP_RB_RPTR);
2334         wdp = RREG32(CP_RB_WPTR);
2335         count = (rdp + rdev->cp.ring_size - wdp) & rdev->cp.ptr_mask;
2336         seq_printf(m, "CP_STAT 0x%08x\n", RREG32(CP_STAT));
2337         seq_printf(m, "CP_RB_WPTR 0x%08x\n", wdp);
2338         seq_printf(m, "CP_RB_RPTR 0x%08x\n", rdp);
2339         seq_printf(m, "%u free dwords in ring\n", rdev->cp.ring_free_dw);
2340         seq_printf(m, "%u dwords in ring\n", count);
2341         for (j = 0; j <= count; j++) {
2342                 i = (rdp + j) & rdev->cp.ptr_mask;
2343                 seq_printf(m, "r[%04d]=0x%08x\n", i, rdev->cp.ring[i]);
2344         }
2345         return 0;
2346 }
2347
2348 static int r600_debugfs_mc_info(struct seq_file *m, void *data)
2349 {
2350         struct drm_info_node *node = (struct drm_info_node *) m->private;
2351         struct drm_device *dev = node->minor->dev;
2352         struct radeon_device *rdev = dev->dev_private;
2353
2354         DREG32_SYS(m, rdev, R_000E50_SRBM_STATUS);
2355         DREG32_SYS(m, rdev, VM_L2_STATUS);
2356         return 0;
2357 }
2358
2359 static struct drm_info_list r600_mc_info_list[] = {
2360         {"r600_mc_info", r600_debugfs_mc_info, 0, NULL},
2361         {"r600_ring_info", r600_debugfs_cp_ring_info, 0, NULL},
2362 };
2363 #endif
2364
2365 int r600_debugfs_mc_info_init(struct radeon_device *rdev)
2366 {
2367 #if defined(CONFIG_DEBUG_FS)
2368         return radeon_debugfs_add_files(rdev, r600_mc_info_list, ARRAY_SIZE(r600_mc_info_list));
2369 #else
2370         return 0;
2371 #endif
2372 }