Pileus Git - ~andy/linux/blob - drivers/gpu/drm/radeon/si.c

   1 /*
   2  * Copyright 2011 Advanced Micro Devices, Inc.
   3  *
   4  * Permission is hereby granted, free of charge, to any person obtaining a
   5  * copy of this software and associated documentation files (the "Software"),
   6  * to deal in the Software without restriction, including without limitation
   7  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
   8  * and/or sell copies of the Software, and to permit persons to whom the
   9  * Software is furnished to do so, subject to the following conditions:
  10  *
  11  * The above copyright notice and this permission notice shall be included in
  12  * all copies or substantial portions of the Software.
  13  *
  14  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  15  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  16  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
  17  * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
  18  * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
  19  * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
  20  * OTHER DEALINGS IN THE SOFTWARE.
  21  *
  22  * Authors: Alex Deucher
  23  */
  24 #include <linux/firmware.h>
  25 #include <linux/platform_device.h>
  26 #include <linux/slab.h>
  27 #include <linux/module.h>
  28 #include <drm/drmP.h>
  29 #include "radeon.h"
  30 #include "radeon_asic.h"
  31 #include <drm/radeon_drm.h>
  32 #include "sid.h"
  33 #include "atom.h"
  34 #include "si_blit_shaders.h"
  35
  36 #define SI_PFP_UCODE_SIZE 2144
  37 #define SI_PM4_UCODE_SIZE 2144
  38 #define SI_CE_UCODE_SIZE 2144
  39 #define SI_RLC_UCODE_SIZE 2048
  40 #define SI_MC_UCODE_SIZE 7769
  41 #define OLAND_MC_UCODE_SIZE 7863
  42
  43 MODULE_FIRMWARE("radeon/TAHITI_pfp.bin");
  44 MODULE_FIRMWARE("radeon/TAHITI_me.bin");
  45 MODULE_FIRMWARE("radeon/TAHITI_ce.bin");
  46 MODULE_FIRMWARE("radeon/TAHITI_mc.bin");
  47 MODULE_FIRMWARE("radeon/TAHITI_rlc.bin");
  48 MODULE_FIRMWARE("radeon/PITCAIRN_pfp.bin");
  49 MODULE_FIRMWARE("radeon/PITCAIRN_me.bin");
  50 MODULE_FIRMWARE("radeon/PITCAIRN_ce.bin");
  51 MODULE_FIRMWARE("radeon/PITCAIRN_mc.bin");
  52 MODULE_FIRMWARE("radeon/PITCAIRN_rlc.bin");
  53 MODULE_FIRMWARE("radeon/VERDE_pfp.bin");
  54 MODULE_FIRMWARE("radeon/VERDE_me.bin");
  55 MODULE_FIRMWARE("radeon/VERDE_ce.bin");
  56 MODULE_FIRMWARE("radeon/VERDE_mc.bin");
  57 MODULE_FIRMWARE("radeon/VERDE_rlc.bin");
  58 MODULE_FIRMWARE("radeon/OLAND_pfp.bin");
  59 MODULE_FIRMWARE("radeon/OLAND_me.bin");
  60 MODULE_FIRMWARE("radeon/OLAND_ce.bin");
  61 MODULE_FIRMWARE("radeon/OLAND_mc.bin");
  62 MODULE_FIRMWARE("radeon/OLAND_rlc.bin");
  63
  64 extern int r600_ih_ring_alloc(struct radeon_device *rdev);
  65 extern void r600_ih_ring_fini(struct radeon_device *rdev);
  66 extern void evergreen_fix_pci_max_read_req_size(struct radeon_device *rdev);
  67 extern void evergreen_mc_stop(struct radeon_device *rdev, struct evergreen_mc_save *save);
  68 extern void evergreen_mc_resume(struct radeon_device *rdev, struct evergreen_mc_save *save);
  69 extern u32 evergreen_get_number_of_dram_channels(struct radeon_device *rdev);
  70 extern void evergreen_print_gpu_status_regs(struct radeon_device *rdev);
  71 extern bool evergreen_is_display_hung(struct radeon_device *rdev);
  72
  73 #define PCIE_BUS_CLK                10000
  74 #define TCLK                        (PCIE_BUS_CLK / 10)
  75
  76 /**
  77  * si_get_xclk - get the xclk
  78  *
  79  * @rdev: radeon_device pointer
  80  *
  81  * Returns the reference clock used by the gfx engine
  82  * (SI).
  83  */
  84 u32 si_get_xclk(struct radeon_device *rdev)
  85 {
  86         u32 reference_clock = rdev->clock.spll.reference_freq;
  87         u32 tmp;
  88
  89         tmp = RREG32(CG_CLKPIN_CNTL_2);
  90         if (tmp & MUX_TCLK_TO_XCLK)
  91                 return TCLK;
  92
  93         tmp = RREG32(CG_CLKPIN_CNTL);
  94         if (tmp & XTALIN_DIVIDE)
  95                 return reference_clock / 4;
  96
  97         return reference_clock;
  98 }
  99
 100 /* get temperature in millidegrees */
 101 int si_get_temp(struct radeon_device *rdev)
 102 {
 103         u32 temp;
 104         int actual_temp = 0;
 105
 106         temp = (RREG32(CG_MULT_THERMAL_STATUS) & CTF_TEMP_MASK) >>
 107                 CTF_TEMP_SHIFT;
 108
 109         if (temp & 0x200)
 110                 actual_temp = 255;
 111         else
 112                 actual_temp = temp & 0x1ff;
 113
 114         actual_temp = (actual_temp * 1000);
 115
 116         return actual_temp;
 117 }
 118
 119 #define TAHITI_IO_MC_REGS_SIZE 36
 120
 121 static const u32 tahiti_io_mc_regs[TAHITI_IO_MC_REGS_SIZE][2] = {
 122         {0x0000006f, 0x03044000},
 123         {0x00000070, 0x0480c018},
 124         {0x00000071, 0x00000040},
 125         {0x00000072, 0x01000000},
 126         {0x00000074, 0x000000ff},
 127         {0x00000075, 0x00143400},
 128         {0x00000076, 0x08ec0800},
 129         {0x00000077, 0x040000cc},
 130         {0x00000079, 0x00000000},
 131         {0x0000007a, 0x21000409},
 132         {0x0000007c, 0x00000000},
 133         {0x0000007d, 0xe8000000},
 134         {0x0000007e, 0x044408a8},
 135         {0x0000007f, 0x00000003},
 136         {0x00000080, 0x00000000},
 137         {0x00000081, 0x01000000},
 138         {0x00000082, 0x02000000},
 139         {0x00000083, 0x00000000},
 140         {0x00000084, 0xe3f3e4f4},
 141         {0x00000085, 0x00052024},
 142         {0x00000087, 0x00000000},
 143         {0x00000088, 0x66036603},
 144         {0x00000089, 0x01000000},
 145         {0x0000008b, 0x1c0a0000},
 146         {0x0000008c, 0xff010000},
 147         {0x0000008e, 0xffffefff},
 148         {0x0000008f, 0xfff3efff},
 149         {0x00000090, 0xfff3efbf},
 150         {0x00000094, 0x00101101},
 151         {0x00000095, 0x00000fff},
 152         {0x00000096, 0x00116fff},
 153         {0x00000097, 0x60010000},
 154         {0x00000098, 0x10010000},
 155         {0x00000099, 0x00006000},
 156         {0x0000009a, 0x00001000},
 157         {0x0000009f, 0x00a77400}
 158 };
 159
 160 static const u32 pitcairn_io_mc_regs[TAHITI_IO_MC_REGS_SIZE][2] = {
 161         {0x0000006f, 0x03044000},
 162         {0x00000070, 0x0480c018},
 163         {0x00000071, 0x00000040},
 164         {0x00000072, 0x01000000},
 165         {0x00000074, 0x000000ff},
 166         {0x00000075, 0x00143400},
 167         {0x00000076, 0x08ec0800},
 168         {0x00000077, 0x040000cc},
 169         {0x00000079, 0x00000000},
 170         {0x0000007a, 0x21000409},
 171         {0x0000007c, 0x00000000},
 172         {0x0000007d, 0xe8000000},
 173         {0x0000007e, 0x044408a8},
 174         {0x0000007f, 0x00000003},
 175         {0x00000080, 0x00000000},
 176         {0x00000081, 0x01000000},
 177         {0x00000082, 0x02000000},
 178         {0x00000083, 0x00000000},
 179         {0x00000084, 0xe3f3e4f4},
 180         {0x00000085, 0x00052024},
 181         {0x00000087, 0x00000000},
 182         {0x00000088, 0x66036603},
 183         {0x00000089, 0x01000000},
 184         {0x0000008b, 0x1c0a0000},
 185         {0x0000008c, 0xff010000},
 186         {0x0000008e, 0xffffefff},
 187         {0x0000008f, 0xfff3efff},
 188         {0x00000090, 0xfff3efbf},
 189         {0x00000094, 0x00101101},
 190         {0x00000095, 0x00000fff},
 191         {0x00000096, 0x00116fff},
 192         {0x00000097, 0x60010000},
 193         {0x00000098, 0x10010000},
 194         {0x00000099, 0x00006000},
 195         {0x0000009a, 0x00001000},
 196         {0x0000009f, 0x00a47400}
 197 };
 198
 199 static const u32 verde_io_mc_regs[TAHITI_IO_MC_REGS_SIZE][2] = {
 200         {0x0000006f, 0x03044000},
 201         {0x00000070, 0x0480c018},
 202         {0x00000071, 0x00000040},
 203         {0x00000072, 0x01000000},
 204         {0x00000074, 0x000000ff},
 205         {0x00000075, 0x00143400},
 206         {0x00000076, 0x08ec0800},
 207         {0x00000077, 0x040000cc},
 208         {0x00000079, 0x00000000},
 209         {0x0000007a, 0x21000409},
 210         {0x0000007c, 0x00000000},
 211         {0x0000007d, 0xe8000000},
 212         {0x0000007e, 0x044408a8},
 213         {0x0000007f, 0x00000003},
 214         {0x00000080, 0x00000000},
 215         {0x00000081, 0x01000000},
 216         {0x00000082, 0x02000000},
 217         {0x00000083, 0x00000000},
 218         {0x00000084, 0xe3f3e4f4},
 219         {0x00000085, 0x00052024},
 220         {0x00000087, 0x00000000},
 221         {0x00000088, 0x66036603},
 222         {0x00000089, 0x01000000},
 223         {0x0000008b, 0x1c0a0000},
 224         {0x0000008c, 0xff010000},
 225         {0x0000008e, 0xffffefff},
 226         {0x0000008f, 0xfff3efff},
 227         {0x00000090, 0xfff3efbf},
 228         {0x00000094, 0x00101101},
 229         {0x00000095, 0x00000fff},
 230         {0x00000096, 0x00116fff},
 231         {0x00000097, 0x60010000},
 232         {0x00000098, 0x10010000},
 233         {0x00000099, 0x00006000},
 234         {0x0000009a, 0x00001000},
 235         {0x0000009f, 0x00a37400}
 236 };
 237
 238 static const u32 oland_io_mc_regs[TAHITI_IO_MC_REGS_SIZE][2] = {
 239         {0x0000006f, 0x03044000},
 240         {0x00000070, 0x0480c018},
 241         {0x00000071, 0x00000040},
 242         {0x00000072, 0x01000000},
 243         {0x00000074, 0x000000ff},
 244         {0x00000075, 0x00143400},
 245         {0x00000076, 0x08ec0800},
 246         {0x00000077, 0x040000cc},
 247         {0x00000079, 0x00000000},
 248         {0x0000007a, 0x21000409},
 249         {0x0000007c, 0x00000000},
 250         {0x0000007d, 0xe8000000},
 251         {0x0000007e, 0x044408a8},
 252         {0x0000007f, 0x00000003},
 253         {0x00000080, 0x00000000},
 254         {0x00000081, 0x01000000},
 255         {0x00000082, 0x02000000},
 256         {0x00000083, 0x00000000},
 257         {0x00000084, 0xe3f3e4f4},
 258         {0x00000085, 0x00052024},
 259         {0x00000087, 0x00000000},
 260         {0x00000088, 0x66036603},
 261         {0x00000089, 0x01000000},
 262         {0x0000008b, 0x1c0a0000},
 263         {0x0000008c, 0xff010000},
 264         {0x0000008e, 0xffffefff},
 265         {0x0000008f, 0xfff3efff},
 266         {0x00000090, 0xfff3efbf},
 267         {0x00000094, 0x00101101},
 268         {0x00000095, 0x00000fff},
 269         {0x00000096, 0x00116fff},
 270         {0x00000097, 0x60010000},
 271         {0x00000098, 0x10010000},
 272         {0x00000099, 0x00006000},
 273         {0x0000009a, 0x00001000},
 274         {0x0000009f, 0x00a17730}
 275 };
 276
 277 /* ucode loading */
 278 static int si_mc_load_microcode(struct radeon_device *rdev)
 279 {
 280         const __be32 *fw_data;
 281         u32 running, blackout = 0;
 282         u32 *io_mc_regs;
 283         int i, ucode_size, regs_size;
 284
 285         if (!rdev->mc_fw)
 286                 return -EINVAL;
 287
 288         switch (rdev->family) {
 289         case CHIP_TAHITI:
 290                 io_mc_regs = (u32 *)&tahiti_io_mc_regs;
 291                 ucode_size = SI_MC_UCODE_SIZE;
 292                 regs_size = TAHITI_IO_MC_REGS_SIZE;
 293                 break;
 294         case CHIP_PITCAIRN:
 295                 io_mc_regs = (u32 *)&pitcairn_io_mc_regs;
 296                 ucode_size = SI_MC_UCODE_SIZE;
 297                 regs_size = TAHITI_IO_MC_REGS_SIZE;
 298                 break;
 299         case CHIP_VERDE:
 300         default:
 301                 io_mc_regs = (u32 *)&verde_io_mc_regs;
 302                 ucode_size = SI_MC_UCODE_SIZE;
 303                 regs_size = TAHITI_IO_MC_REGS_SIZE;
 304                 break;
 305         case CHIP_OLAND:
 306                 io_mc_regs = (u32 *)&oland_io_mc_regs;
 307                 ucode_size = OLAND_MC_UCODE_SIZE;
 308                 regs_size = TAHITI_IO_MC_REGS_SIZE;
 309                 break;
 310         }
 311
 312         running = RREG32(MC_SEQ_SUP_CNTL) & RUN_MASK;
 313
 314         if (running == 0) {
 315                 if (running) {
 316                         blackout = RREG32(MC_SHARED_BLACKOUT_CNTL);
 317                         WREG32(MC_SHARED_BLACKOUT_CNTL, blackout | 1);
 318                 }
 319
 320                 /* reset the engine and set to writable */
 321                 WREG32(MC_SEQ_SUP_CNTL, 0x00000008);
 322                 WREG32(MC_SEQ_SUP_CNTL, 0x00000010);
 323
 324                 /* load mc io regs */
 325                 for (i = 0; i < regs_size; i++) {
 326                         WREG32(MC_SEQ_IO_DEBUG_INDEX, io_mc_regs[(i << 1)]);
 327                         WREG32(MC_SEQ_IO_DEBUG_DATA, io_mc_regs[(i << 1) + 1]);
 328                 }
 329                 /* load the MC ucode */
 330                 fw_data = (const __be32 *)rdev->mc_fw->data;
 331                 for (i = 0; i < ucode_size; i++)
 332                         WREG32(MC_SEQ_SUP_PGM, be32_to_cpup(fw_data++));
 333
 334                 /* put the engine back into the active state */
 335                 WREG32(MC_SEQ_SUP_CNTL, 0x00000008);
 336                 WREG32(MC_SEQ_SUP_CNTL, 0x00000004);
 337                 WREG32(MC_SEQ_SUP_CNTL, 0x00000001);
 338
 339                 /* wait for training to complete */
 340                 for (i = 0; i < rdev->usec_timeout; i++) {
 341                         if (RREG32(MC_SEQ_TRAIN_WAKEUP_CNTL) & TRAIN_DONE_D0)
 342                                 break;
 343                         udelay(1);
 344                 }
 345                 for (i = 0; i < rdev->usec_timeout; i++) {
 346                         if (RREG32(MC_SEQ_TRAIN_WAKEUP_CNTL) & TRAIN_DONE_D1)
 347                                 break;
 348                         udelay(1);
 349                 }
 350
 351                 if (running)
 352                         WREG32(MC_SHARED_BLACKOUT_CNTL, blackout);
 353         }
 354
 355         return 0;
 356 }
 357
 358 static int si_init_microcode(struct radeon_device *rdev)
 359 {
 360         struct platform_device *pdev;
 361         const char *chip_name;
 362         const char *rlc_chip_name;
 363         size_t pfp_req_size, me_req_size, ce_req_size, rlc_req_size, mc_req_size;
 364         char fw_name[30];
 365         int err;
 366
 367         DRM_DEBUG("\n");
 368
 369         pdev = platform_device_register_simple("radeon_cp", 0, NULL, 0);
 370         err = IS_ERR(pdev);
 371         if (err) {
 372                 printk(KERN_ERR "radeon_cp: Failed to register firmware\n");
 373                 return -EINVAL;
 374         }
 375
 376         switch (rdev->family) {
 377         case CHIP_TAHITI:
 378                 chip_name = "TAHITI";
 379                 rlc_chip_name = "TAHITI";
 380                 pfp_req_size = SI_PFP_UCODE_SIZE * 4;
 381                 me_req_size = SI_PM4_UCODE_SIZE * 4;
 382                 ce_req_size = SI_CE_UCODE_SIZE * 4;
 383                 rlc_req_size = SI_RLC_UCODE_SIZE * 4;
 384                 mc_req_size = SI_MC_UCODE_SIZE * 4;
 385                 break;
 386         case CHIP_PITCAIRN:
 387                 chip_name = "PITCAIRN";
 388                 rlc_chip_name = "PITCAIRN";
 389                 pfp_req_size = SI_PFP_UCODE_SIZE * 4;
 390                 me_req_size = SI_PM4_UCODE_SIZE * 4;
 391                 ce_req_size = SI_CE_UCODE_SIZE * 4;
 392                 rlc_req_size = SI_RLC_UCODE_SIZE * 4;
 393                 mc_req_size = SI_MC_UCODE_SIZE * 4;
 394                 break;
 395         case CHIP_VERDE:
 396                 chip_name = "VERDE";
 397                 rlc_chip_name = "VERDE";
 398                 pfp_req_size = SI_PFP_UCODE_SIZE * 4;
 399                 me_req_size = SI_PM4_UCODE_SIZE * 4;
 400                 ce_req_size = SI_CE_UCODE_SIZE * 4;
 401                 rlc_req_size = SI_RLC_UCODE_SIZE * 4;
 402                 mc_req_size = SI_MC_UCODE_SIZE * 4;
 403                 break;
 404         case CHIP_OLAND:
 405                 chip_name = "OLAND";
 406                 rlc_chip_name = "OLAND";
 407                 pfp_req_size = SI_PFP_UCODE_SIZE * 4;
 408                 me_req_size = SI_PM4_UCODE_SIZE * 4;
 409                 ce_req_size = SI_CE_UCODE_SIZE * 4;
 410                 rlc_req_size = SI_RLC_UCODE_SIZE * 4;
 411                 mc_req_size = OLAND_MC_UCODE_SIZE * 4;
 412                 break;
 413         default: BUG();
 414         }
 415
 416         DRM_INFO("Loading %s Microcode\n", chip_name);
 417
 418         snprintf(fw_name, sizeof(fw_name), "radeon/%s_pfp.bin", chip_name);
 419         err = request_firmware(&rdev->pfp_fw, fw_name, &pdev->dev);
 420         if (err)
 421                 goto out;
 422         if (rdev->pfp_fw->size != pfp_req_size) {
 423                 printk(KERN_ERR
 424                        "si_cp: Bogus length %zu in firmware \"%s\"\n",
 425                        rdev->pfp_fw->size, fw_name);
 426                 err = -EINVAL;
 427                 goto out;
 428         }
 429
 430         snprintf(fw_name, sizeof(fw_name), "radeon/%s_me.bin", chip_name);
 431         err = request_firmware(&rdev->me_fw, fw_name, &pdev->dev);
 432         if (err)
 433                 goto out;
 434         if (rdev->me_fw->size != me_req_size) {
 435                 printk(KERN_ERR
 436                        "si_cp: Bogus length %zu in firmware \"%s\"\n",
 437                        rdev->me_fw->size, fw_name);
 438                 err = -EINVAL;
 439         }
 440
 441         snprintf(fw_name, sizeof(fw_name), "radeon/%s_ce.bin", chip_name);
 442         err = request_firmware(&rdev->ce_fw, fw_name, &pdev->dev);
 443         if (err)
 444                 goto out;
 445         if (rdev->ce_fw->size != ce_req_size) {
 446                 printk(KERN_ERR
 447                        "si_cp: Bogus length %zu in firmware \"%s\"\n",
 448                        rdev->ce_fw->size, fw_name);
 449                 err = -EINVAL;
 450         }
 451
 452         snprintf(fw_name, sizeof(fw_name), "radeon/%s_rlc.bin", rlc_chip_name);
 453         err = request_firmware(&rdev->rlc_fw, fw_name, &pdev->dev);
 454         if (err)
 455                 goto out;
 456         if (rdev->rlc_fw->size != rlc_req_size) {
 457                 printk(KERN_ERR
 458                        "si_rlc: Bogus length %zu in firmware \"%s\"\n",
 459                        rdev->rlc_fw->size, fw_name);
 460                 err = -EINVAL;
 461         }
 462
 463         snprintf(fw_name, sizeof(fw_name), "radeon/%s_mc.bin", chip_name);
 464         err = request_firmware(&rdev->mc_fw, fw_name, &pdev->dev);
 465         if (err)
 466                 goto out;
 467         if (rdev->mc_fw->size != mc_req_size) {
 468                 printk(KERN_ERR
 469                        "si_mc: Bogus length %zu in firmware \"%s\"\n",
 470                        rdev->mc_fw->size, fw_name);
 471                 err = -EINVAL;
 472         }
 473
 474 out:
 475         platform_device_unregister(pdev);
 476
 477         if (err) {
 478                 if (err != -EINVAL)
 479                         printk(KERN_ERR
 480                                "si_cp: Failed to load firmware \"%s\"\n",
 481                                fw_name);
 482                 release_firmware(rdev->pfp_fw);
 483                 rdev->pfp_fw = NULL;
 484                 release_firmware(rdev->me_fw);
 485                 rdev->me_fw = NULL;
 486                 release_firmware(rdev->ce_fw);
 487                 rdev->ce_fw = NULL;
 488                 release_firmware(rdev->rlc_fw);
 489                 rdev->rlc_fw = NULL;
 490                 release_firmware(rdev->mc_fw);
 491                 rdev->mc_fw = NULL;
 492         }
 493         return err;
 494 }
 495
 496 /* watermark setup */
 497 static u32 dce6_line_buffer_adjust(struct radeon_device *rdev,
 498                                    struct radeon_crtc *radeon_crtc,
 499                                    struct drm_display_mode *mode,
 500                                    struct drm_display_mode *other_mode)
 501 {
 502         u32 tmp;
 503         /*
 504          * Line Buffer Setup
 505          * There are 3 line buffers, each one shared by 2 display controllers.
 506          * DC_LB_MEMORY_SPLIT controls how that line buffer is shared between
 507          * the display controllers.  The paritioning is done via one of four
 508          * preset allocations specified in bits 21:20:
 509          *  0 - half lb
 510          *  2 - whole lb, other crtc must be disabled
 511          */
 512         /* this can get tricky if we have two large displays on a paired group
 513          * of crtcs.  Ideally for multiple large displays we'd assign them to
 514          * non-linked crtcs for maximum line buffer allocation.
 515          */
 516         if (radeon_crtc->base.enabled && mode) {
 517                 if (other_mode)
 518                         tmp = 0; /* 1/2 */
 519                 else
 520                         tmp = 2; /* whole */
 521         } else
 522                 tmp = 0;
 523
 524         WREG32(DC_LB_MEMORY_SPLIT + radeon_crtc->crtc_offset,
 525                DC_LB_MEMORY_CONFIG(tmp));
 526
 527         if (radeon_crtc->base.enabled && mode) {
 528                 switch (tmp) {
 529                 case 0:
 530                 default:
 531                         return 4096 * 2;
 532                 case 2:
 533                         return 8192 * 2;
 534                 }
 535         }
 536
 537         /* controller not enabled, so no lb used */
 538         return 0;
 539 }
 540
 541 static u32 si_get_number_of_dram_channels(struct radeon_device *rdev)
 542 {
 543         u32 tmp = RREG32(MC_SHARED_CHMAP);
 544
 545         switch ((tmp & NOOFCHAN_MASK) >> NOOFCHAN_SHIFT) {
 546         case 0:
 547         default:
 548                 return 1;
 549         case 1:
 550                 return 2;
 551         case 2:
 552                 return 4;
 553         case 3:
 554                 return 8;
 555         case 4:
 556                 return 3;
 557         case 5:
 558                 return 6;
 559         case 6:
 560                 return 10;
 561         case 7:
 562                 return 12;
 563         case 8:
 564                 return 16;
 565         }
 566 }
 567
 568 struct dce6_wm_params {
 569         u32 dram_channels; /* number of dram channels */
 570         u32 yclk;          /* bandwidth per dram data pin in kHz */
 571         u32 sclk;          /* engine clock in kHz */
 572         u32 disp_clk;      /* display clock in kHz */
 573         u32 src_width;     /* viewport width */
 574         u32 active_time;   /* active display time in ns */
 575         u32 blank_time;    /* blank time in ns */
 576         bool interlaced;    /* mode is interlaced */
 577         fixed20_12 vsc;    /* vertical scale ratio */
 578         u32 num_heads;     /* number of active crtcs */
 579         u32 bytes_per_pixel; /* bytes per pixel display + overlay */
 580         u32 lb_size;       /* line buffer allocated to pipe */
 581         u32 vtaps;         /* vertical scaler taps */
 582 };
 583
 584 static u32 dce6_dram_bandwidth(struct dce6_wm_params *wm)
 585 {
 586         /* Calculate raw DRAM Bandwidth */
 587         fixed20_12 dram_efficiency; /* 0.7 */
 588         fixed20_12 yclk, dram_channels, bandwidth;
 589         fixed20_12 a;
 590
 591         a.full = dfixed_const(1000);
 592         yclk.full = dfixed_const(wm->yclk);
 593         yclk.full = dfixed_div(yclk, a);
 594         dram_channels.full = dfixed_const(wm->dram_channels * 4);
 595         a.full = dfixed_const(10);
 596         dram_efficiency.full = dfixed_const(7);
 597         dram_efficiency.full = dfixed_div(dram_efficiency, a);
 598         bandwidth.full = dfixed_mul(dram_channels, yclk);
 599         bandwidth.full = dfixed_mul(bandwidth, dram_efficiency);
 600
 601         return dfixed_trunc(bandwidth);
 602 }
 603
 604 static u32 dce6_dram_bandwidth_for_display(struct dce6_wm_params *wm)
 605 {
 606         /* Calculate DRAM Bandwidth and the part allocated to display. */
 607         fixed20_12 disp_dram_allocation; /* 0.3 to 0.7 */
 608         fixed20_12 yclk, dram_channels, bandwidth;
 609         fixed20_12 a;
 610
 611         a.full = dfixed_const(1000);
 612         yclk.full = dfixed_const(wm->yclk);
 613         yclk.full = dfixed_div(yclk, a);
 614         dram_channels.full = dfixed_const(wm->dram_channels * 4);
 615         a.full = dfixed_const(10);
 616         disp_dram_allocation.full = dfixed_const(3); /* XXX worse case value 0.3 */
 617         disp_dram_allocation.full = dfixed_div(disp_dram_allocation, a);
 618         bandwidth.full = dfixed_mul(dram_channels, yclk);
 619         bandwidth.full = dfixed_mul(bandwidth, disp_dram_allocation);
 620
 621         return dfixed_trunc(bandwidth);
 622 }
 623
 624 static u32 dce6_data_return_bandwidth(struct dce6_wm_params *wm)
 625 {
 626         /* Calculate the display Data return Bandwidth */
 627         fixed20_12 return_efficiency; /* 0.8 */
 628         fixed20_12 sclk, bandwidth;
 629         fixed20_12 a;
 630
 631         a.full = dfixed_const(1000);
 632         sclk.full = dfixed_const(wm->sclk);
 633         sclk.full = dfixed_div(sclk, a);
 634         a.full = dfixed_const(10);
 635         return_efficiency.full = dfixed_const(8);
 636         return_efficiency.full = dfixed_div(return_efficiency, a);
 637         a.full = dfixed_const(32);
 638         bandwidth.full = dfixed_mul(a, sclk);
 639         bandwidth.full = dfixed_mul(bandwidth, return_efficiency);
 640
 641         return dfixed_trunc(bandwidth);
 642 }
 643
 644 static u32 dce6_get_dmif_bytes_per_request(struct dce6_wm_params *wm)
 645 {
 646         return 32;
 647 }
 648
 649 static u32 dce6_dmif_request_bandwidth(struct dce6_wm_params *wm)
 650 {
 651         /* Calculate the DMIF Request Bandwidth */
 652         fixed20_12 disp_clk_request_efficiency; /* 0.8 */
 653         fixed20_12 disp_clk, sclk, bandwidth;
 654         fixed20_12 a, b1, b2;
 655         u32 min_bandwidth;
 656
 657         a.full = dfixed_const(1000);
 658         disp_clk.full = dfixed_const(wm->disp_clk);
 659         disp_clk.full = dfixed_div(disp_clk, a);
 660         a.full = dfixed_const(dce6_get_dmif_bytes_per_request(wm) / 2);
 661         b1.full = dfixed_mul(a, disp_clk);
 662
 663         a.full = dfixed_const(1000);
 664         sclk.full = dfixed_const(wm->sclk);
 665         sclk.full = dfixed_div(sclk, a);
 666         a.full = dfixed_const(dce6_get_dmif_bytes_per_request(wm));
 667         b2.full = dfixed_mul(a, sclk);
 668
 669         a.full = dfixed_const(10);
 670         disp_clk_request_efficiency.full = dfixed_const(8);
 671         disp_clk_request_efficiency.full = dfixed_div(disp_clk_request_efficiency, a);
 672
 673         min_bandwidth = min(dfixed_trunc(b1), dfixed_trunc(b2));
 674
 675         a.full = dfixed_const(min_bandwidth);
 676         bandwidth.full = dfixed_mul(a, disp_clk_request_efficiency);
 677
 678         return dfixed_trunc(bandwidth);
 679 }
 680
 681 static u32 dce6_available_bandwidth(struct dce6_wm_params *wm)
 682 {
 683         /* Calculate the Available bandwidth. Display can use this temporarily but not in average. */
 684         u32 dram_bandwidth = dce6_dram_bandwidth(wm);
 685         u32 data_return_bandwidth = dce6_data_return_bandwidth(wm);
 686         u32 dmif_req_bandwidth = dce6_dmif_request_bandwidth(wm);
 687
 688         return min(dram_bandwidth, min(data_return_bandwidth, dmif_req_bandwidth));
 689 }
 690
 691 static u32 dce6_average_bandwidth(struct dce6_wm_params *wm)
 692 {
 693         /* Calculate the display mode Average Bandwidth
 694          * DisplayMode should contain the source and destination dimensions,
 695          * timing, etc.
 696          */
 697         fixed20_12 bpp;
 698         fixed20_12 line_time;
 699         fixed20_12 src_width;
 700         fixed20_12 bandwidth;
 701         fixed20_12 a;
 702
 703         a.full = dfixed_const(1000);
 704         line_time.full = dfixed_const(wm->active_time + wm->blank_time);
 705         line_time.full = dfixed_div(line_time, a);
 706         bpp.full = dfixed_const(wm->bytes_per_pixel);
 707         src_width.full = dfixed_const(wm->src_width);
 708         bandwidth.full = dfixed_mul(src_width, bpp);
 709         bandwidth.full = dfixed_mul(bandwidth, wm->vsc);
 710         bandwidth.full = dfixed_div(bandwidth, line_time);
 711
 712         return dfixed_trunc(bandwidth);
 713 }
 714
 715 static u32 dce6_latency_watermark(struct dce6_wm_params *wm)
 716 {
 717         /* First calcualte the latency in ns */
 718         u32 mc_latency = 2000; /* 2000 ns. */
 719         u32 available_bandwidth = dce6_available_bandwidth(wm);
 720         u32 worst_chunk_return_time = (512 * 8 * 1000) / available_bandwidth;
 721         u32 cursor_line_pair_return_time = (128 * 4 * 1000) / available_bandwidth;
 722         u32 dc_latency = 40000000 / wm->disp_clk; /* dc pipe latency */
 723         u32 other_heads_data_return_time = ((wm->num_heads + 1) * worst_chunk_return_time) +
 724                 (wm->num_heads * cursor_line_pair_return_time);
 725         u32 latency = mc_latency + other_heads_data_return_time + dc_latency;
 726         u32 max_src_lines_per_dst_line, lb_fill_bw, line_fill_time;
 727         u32 tmp, dmif_size = 12288;
 728         fixed20_12 a, b, c;
 729
 730         if (wm->num_heads == 0)
 731                 return 0;
 732
 733         a.full = dfixed_const(2);
 734         b.full = dfixed_const(1);
 735         if ((wm->vsc.full > a.full) ||
 736             ((wm->vsc.full > b.full) && (wm->vtaps >= 3)) ||
 737             (wm->vtaps >= 5) ||
 738             ((wm->vsc.full >= a.full) && wm->interlaced))
 739                 max_src_lines_per_dst_line = 4;
 740         else
 741                 max_src_lines_per_dst_line = 2;
 742
 743         a.full = dfixed_const(available_bandwidth);
 744         b.full = dfixed_const(wm->num_heads);
 745         a.full = dfixed_div(a, b);
 746
 747         b.full = dfixed_const(mc_latency + 512);
 748         c.full = dfixed_const(wm->disp_clk);
 749         b.full = dfixed_div(b, c);
 750
 751         c.full = dfixed_const(dmif_size);
 752         b.full = dfixed_div(c, b);
 753
 754         tmp = min(dfixed_trunc(a), dfixed_trunc(b));
 755
 756         b.full = dfixed_const(1000);
 757         c.full = dfixed_const(wm->disp_clk);
 758         b.full = dfixed_div(c, b);
 759         c.full = dfixed_const(wm->bytes_per_pixel);
 760         b.full = dfixed_mul(b, c);
 761
 762         lb_fill_bw = min(tmp, dfixed_trunc(b));
 763
 764         a.full = dfixed_const(max_src_lines_per_dst_line * wm->src_width * wm->bytes_per_pixel);
 765         b.full = dfixed_const(1000);
 766         c.full = dfixed_const(lb_fill_bw);
 767         b.full = dfixed_div(c, b);
 768         a.full = dfixed_div(a, b);
 769         line_fill_time = dfixed_trunc(a);
 770
 771         if (line_fill_time < wm->active_time)
 772                 return latency;
 773         else
 774                 return latency + (line_fill_time - wm->active_time);
 775
 776 }
 777
 778 static bool dce6_average_bandwidth_vs_dram_bandwidth_for_display(struct dce6_wm_params *wm)
 779 {
 780         if (dce6_average_bandwidth(wm) <=
 781             (dce6_dram_bandwidth_for_display(wm) / wm->num_heads))
 782                 return true;
 783         else
 784                 return false;
 785 };
 786
 787 static bool dce6_average_bandwidth_vs_available_bandwidth(struct dce6_wm_params *wm)
 788 {
 789         if (dce6_average_bandwidth(wm) <=
 790             (dce6_available_bandwidth(wm) / wm->num_heads))
 791                 return true;
 792         else
 793                 return false;
 794 };
 795
 796 static bool dce6_check_latency_hiding(struct dce6_wm_params *wm)
 797 {
 798         u32 lb_partitions = wm->lb_size / wm->src_width;
 799         u32 line_time = wm->active_time + wm->blank_time;
 800         u32 latency_tolerant_lines;
 801         u32 latency_hiding;
 802         fixed20_12 a;
 803
 804         a.full = dfixed_const(1);
 805         if (wm->vsc.full > a.full)
 806                 latency_tolerant_lines = 1;
 807         else {
 808                 if (lb_partitions <= (wm->vtaps + 1))
 809                         latency_tolerant_lines = 1;
 810                 else
 811                         latency_tolerant_lines = 2;
 812         }
 813
 814         latency_hiding = (latency_tolerant_lines * line_time + wm->blank_time);
 815
 816         if (dce6_latency_watermark(wm) <= latency_hiding)
 817                 return true;
 818         else
 819                 return false;
 820 }
 821
 822 static void dce6_program_watermarks(struct radeon_device *rdev,
 823                                          struct radeon_crtc *radeon_crtc,
 824                                          u32 lb_size, u32 num_heads)
 825 {
 826         struct drm_display_mode *mode = &radeon_crtc->base.mode;
 827         struct dce6_wm_params wm;
 828         u32 pixel_period;
 829         u32 line_time = 0;
 830         u32 latency_watermark_a = 0, latency_watermark_b = 0;
 831         u32 priority_a_mark = 0, priority_b_mark = 0;
 832         u32 priority_a_cnt = PRIORITY_OFF;
 833         u32 priority_b_cnt = PRIORITY_OFF;
 834         u32 tmp, arb_control3;
 835         fixed20_12 a, b, c;
 836
 837         if (radeon_crtc->base.enabled && num_heads && mode) {
 838                 pixel_period = 1000000 / (u32)mode->clock;
 839                 line_time = min((u32)mode->crtc_htotal * pixel_period, (u32)65535);
 840                 priority_a_cnt = 0;
 841                 priority_b_cnt = 0;
 842
 843                 wm.yclk = rdev->pm.current_mclk * 10;
 844                 wm.sclk = rdev->pm.current_sclk * 10;
 845                 wm.disp_clk = mode->clock;
 846                 wm.src_width = mode->crtc_hdisplay;
 847                 wm.active_time = mode->crtc_hdisplay * pixel_period;
 848                 wm.blank_time = line_time - wm.active_time;
 849                 wm.interlaced = false;
 850                 if (mode->flags & DRM_MODE_FLAG_INTERLACE)
 851                         wm.interlaced = true;
 852                 wm.vsc = radeon_crtc->vsc;
 853                 wm.vtaps = 1;
 854                 if (radeon_crtc->rmx_type != RMX_OFF)
 855                         wm.vtaps = 2;
 856                 wm.bytes_per_pixel = 4; /* XXX: get this from fb config */
 857                 wm.lb_size = lb_size;
 858                 if (rdev->family == CHIP_ARUBA)
 859                         wm.dram_channels = evergreen_get_number_of_dram_channels(rdev);
 860                 else
 861                         wm.dram_channels = si_get_number_of_dram_channels(rdev);
 862                 wm.num_heads = num_heads;
 863
 864                 /* set for high clocks */
 865                 latency_watermark_a = min(dce6_latency_watermark(&wm), (u32)65535);
 866                 /* set for low clocks */
 867                 /* wm.yclk = low clk; wm.sclk = low clk */
 868                 latency_watermark_b = min(dce6_latency_watermark(&wm), (u32)65535);
 869
 870                 /* possibly force display priority to high */
 871                 /* should really do this at mode validation time... */
 872                 if (!dce6_average_bandwidth_vs_dram_bandwidth_for_display(&wm) ||
 873                     !dce6_average_bandwidth_vs_available_bandwidth(&wm) ||
 874                     !dce6_check_latency_hiding(&wm) ||
 875                     (rdev->disp_priority == 2)) {
 876                         DRM_DEBUG_KMS("force priority to high\n");
 877                         priority_a_cnt |= PRIORITY_ALWAYS_ON;
 878                         priority_b_cnt |= PRIORITY_ALWAYS_ON;
 879                 }
 880
 881                 a.full = dfixed_const(1000);
 882                 b.full = dfixed_const(mode->clock);
 883                 b.full = dfixed_div(b, a);
 884                 c.full = dfixed_const(latency_watermark_a);
 885                 c.full = dfixed_mul(c, b);
 886                 c.full = dfixed_mul(c, radeon_crtc->hsc);
 887                 c.full = dfixed_div(c, a);
 888                 a.full = dfixed_const(16);
 889                 c.full = dfixed_div(c, a);
 890                 priority_a_mark = dfixed_trunc(c);
 891                 priority_a_cnt |= priority_a_mark & PRIORITY_MARK_MASK;
 892
 893                 a.full = dfixed_const(1000);
 894                 b.full = dfixed_const(mode->clock);
 895                 b.full = dfixed_div(b, a);
 896                 c.full = dfixed_const(latency_watermark_b);
 897                 c.full = dfixed_mul(c, b);
 898                 c.full = dfixed_mul(c, radeon_crtc->hsc);
 899                 c.full = dfixed_div(c, a);
 900                 a.full = dfixed_const(16);
 901                 c.full = dfixed_div(c, a);
 902                 priority_b_mark = dfixed_trunc(c);
 903                 priority_b_cnt |= priority_b_mark & PRIORITY_MARK_MASK;
 904         }
 905
 906         /* select wm A */
 907         arb_control3 = RREG32(DPG_PIPE_ARBITRATION_CONTROL3 + radeon_crtc->crtc_offset);
 908         tmp = arb_control3;
 909         tmp &= ~LATENCY_WATERMARK_MASK(3);
 910         tmp |= LATENCY_WATERMARK_MASK(1);
 911         WREG32(DPG_PIPE_ARBITRATION_CONTROL3 + radeon_crtc->crtc_offset, tmp);
 912         WREG32(DPG_PIPE_LATENCY_CONTROL + radeon_crtc->crtc_offset,
 913                (LATENCY_LOW_WATERMARK(latency_watermark_a) |
 914                 LATENCY_HIGH_WATERMARK(line_time)));
 915         /* select wm B */
 916         tmp = RREG32(DPG_PIPE_ARBITRATION_CONTROL3 + radeon_crtc->crtc_offset);
 917         tmp &= ~LATENCY_WATERMARK_MASK(3);
 918         tmp |= LATENCY_WATERMARK_MASK(2);
 919         WREG32(DPG_PIPE_ARBITRATION_CONTROL3 + radeon_crtc->crtc_offset, tmp);
 920         WREG32(DPG_PIPE_LATENCY_CONTROL + radeon_crtc->crtc_offset,
 921                (LATENCY_LOW_WATERMARK(latency_watermark_b) |
 922                 LATENCY_HIGH_WATERMARK(line_time)));
 923         /* restore original selection */
 924         WREG32(DPG_PIPE_ARBITRATION_CONTROL3 + radeon_crtc->crtc_offset, arb_control3);
 925
 926         /* write the priority marks */
 927         WREG32(PRIORITY_A_CNT + radeon_crtc->crtc_offset, priority_a_cnt);
 928         WREG32(PRIORITY_B_CNT + radeon_crtc->crtc_offset, priority_b_cnt);
 929
 930 }
 931
 932 void dce6_bandwidth_update(struct radeon_device *rdev)
 933 {
 934         struct drm_display_mode *mode0 = NULL;
 935         struct drm_display_mode *mode1 = NULL;
 936         u32 num_heads = 0, lb_size;
 937         int i;
 938
 939         radeon_update_display_priority(rdev);
 940
 941         for (i = 0; i < rdev->num_crtc; i++) {
 942                 if (rdev->mode_info.crtcs[i]->base.enabled)
 943                         num_heads++;
 944         }
 945         for (i = 0; i < rdev->num_crtc; i += 2) {
 946                 mode0 = &rdev->mode_info.crtcs[i]->base.mode;
 947                 mode1 = &rdev->mode_info.crtcs[i+1]->base.mode;
 948                 lb_size = dce6_line_buffer_adjust(rdev, rdev->mode_info.crtcs[i], mode0, mode1);
 949                 dce6_program_watermarks(rdev, rdev->mode_info.crtcs[i], lb_size, num_heads);
 950                 lb_size = dce6_line_buffer_adjust(rdev, rdev->mode_info.crtcs[i+1], mode1, mode0);
 951                 dce6_program_watermarks(rdev, rdev->mode_info.crtcs[i+1], lb_size, num_heads);
 952         }
 953 }
 954
 955 /*
 956  * Core functions
 957  */
 958 static void si_tiling_mode_table_init(struct radeon_device *rdev)
 959 {
 960         const u32 num_tile_mode_states = 32;
 961         u32 reg_offset, gb_tile_moden, split_equal_to_row_size;
 962
 963         switch (rdev->config.si.mem_row_size_in_kb) {
 964         case 1:
 965                 split_equal_to_row_size = ADDR_SURF_TILE_SPLIT_1KB;
 966                 break;
 967         case 2:
 968         default:
 969                 split_equal_to_row_size = ADDR_SURF_TILE_SPLIT_2KB;
 970                 break;
 971         case 4:
 972                 split_equal_to_row_size = ADDR_SURF_TILE_SPLIT_4KB;
 973                 break;
 974         }
 975
 976         if ((rdev->family == CHIP_TAHITI) ||
 977             (rdev->family == CHIP_PITCAIRN)) {
 978                 for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++) {
 979                         switch (reg_offset) {
 980                         case 0:  /* non-AA compressed depth or any compressed stencil */
 981                                 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
 982                                                  MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) |
 983                                                  PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
 984                                                  TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) |
 985                                                  NUM_BANKS(ADDR_SURF_16_BANK) |
 986                                                  BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
 987                                                  BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
 988                                                  MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
 989                                 break;
 990                         case 1:  /* 2xAA/4xAA compressed depth only */
 991                                 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
 992                                                  MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) |
 993                                                  PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
 994                                                  TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) |
 995                                                  NUM_BANKS(ADDR_SURF_16_BANK) |
 996                                                  BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
 997                                                  BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
 998                                                  MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
 999                                 break;
1000                         case 2:  /* 8xAA compressed depth only */
1001                                 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1002                                                  MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) |
1003                                                  PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
1004                                                  TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
1005                                                  NUM_BANKS(ADDR_SURF_16_BANK) |
1006                                                  BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1007                                                  BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1008                                                  MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
1009                                 break;
1010                         case 3:  /* 2xAA/4xAA compressed depth with stencil (for depth buffer) */
1011                                 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1012                                                  MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) |
1013                                                  PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
1014                                                  TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) |
1015                                                  NUM_BANKS(ADDR_SURF_16_BANK) |
1016                                                  BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1017                                                  BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1018                                                  MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
1019                                 break;
1020                         case 4:  /* Maps w/ a dimension less than the 2D macro-tile dimensions (for mipmapped depth textures) */
1021                                 gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
1022                                                  MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) |
1023                                                  PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
1024                                                  TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) |
1025                                                  NUM_BANKS(ADDR_SURF_16_BANK) |
1026                                                  BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1027                                                  BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1028                                                  MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
1029                                 break;
1030                         case 5:  /* Uncompressed 16bpp depth - and stencil buffer allocated with it */
1031                                 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1032                                                  MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) |
1033                                                  PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
1034                                                  TILE_SPLIT(split_equal_to_row_size) |
1035                                                  NUM_BANKS(ADDR_SURF_16_BANK) |
1036                                                  BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1037                                                  BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1038                                                  MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
1039                                 break;
1040                         case 6:  /* Uncompressed 32bpp depth - and stencil buffer allocated with it */
1041                                 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1042                                                  MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) |
1043                                                  PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
1044                                                  TILE_SPLIT(split_equal_to_row_size) |
1045                                                  NUM_BANKS(ADDR_SURF_16_BANK) |
1046                                                  BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1047                                                  BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1048                                                  MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1));
1049                                 break;
1050                         case 7:  /* Uncompressed 8bpp stencil without depth (drivers typically do not use) */
1051                                 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1052                                                  MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) |
1053                                                  PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
1054                                                  TILE_SPLIT(split_equal_to_row_size) |
1055                                                  NUM_BANKS(ADDR_SURF_16_BANK) |
1056                                                  BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1057                                                  BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1058                                                  MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
1059                                 break;
1060                         case 8:  /* 1D and 1D Array Surfaces */
1061                                 gb_tile_moden = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) |
1062                                                  MICRO_TILE_MODE(ADDR_SURF_DISPLAY_MICRO_TILING) |
1063                                                  PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
1064                                                  TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) |
1065                                                  NUM_BANKS(ADDR_SURF_16_BANK) |
1066                                                  BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1067                                                  BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1068                                                  MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
1069                                 break;
1070                         case 9:  /* Displayable maps. */
1071                                 gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
1072                                                  MICRO_TILE_MODE(ADDR_SURF_DISPLAY_MICRO_TILING) |
1073                                                  PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
1074                                                  TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) |
1075                                                  NUM_BANKS(ADDR_SURF_16_BANK) |
1076                                                  BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1077                                                  BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1078                                                  MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
1079                                 break;
1080                         case 10:  /* Display 8bpp. */
1081                                 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1082                                                  MICRO_TILE_MODE(ADDR_SURF_DISPLAY_MICRO_TILING) |
1083                                                  PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
1084                                                  TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
1085                                                  NUM_BANKS(ADDR_SURF_16_BANK) |
1086                                                  BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1087                                                  BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1088                                                  MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
1089                                 break;
1090                         case 11:  /* Display 16bpp. */
1091                                 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1092                                                  MICRO_TILE_MODE(ADDR_SURF_DISPLAY_MICRO_TILING) |
1093                                                  PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
1094                                                  TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
1095                                                  NUM_BANKS(ADDR_SURF_16_BANK) |
1096                                                  BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1097                                                  BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1098                                                  MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
1099                                 break;
1100                         case 12:  /* Display 32bpp. */
1101                                 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1102                                                  MICRO_TILE_MODE(ADDR_SURF_DISPLAY_MICRO_TILING) |
1103                                                  PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
1104                                                  TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) |
1105                                                  NUM_BANKS(ADDR_SURF_16_BANK) |
1106                                                  BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1107                                                  BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1108                                                  MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1));
1109                                 break;
1110                         case 13:  /* Thin. */
1111                                 gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
1112                                                  MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) |
1113                                                  PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
1114                                                  TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) |
1115                                                  NUM_BANKS(ADDR_SURF_16_BANK) |
1116                                                  BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1117                                                  BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1118                                                  MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
1119                                 break;
1120                         case 14:  /* Thin 8 bpp. */
1121                                 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1122                                                  MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) |
1123                                                  PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
1124                                                  TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
1125                                                  NUM_BANKS(ADDR_SURF_16_BANK) |
1126                                                  BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1127                                                  BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1128                                                  MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1));
1129                                 break;
1130                         case 15:  /* Thin 16 bpp. */
1131                                 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1132                                                  MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) |
1133                                                  PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
1134                                                  TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
1135                                                  NUM_BANKS(ADDR_SURF_16_BANK) |
1136                                                  BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1137                                                  BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1138                                                  MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1));
1139                                 break;
1140                         case 16:  /* Thin 32 bpp. */
1141                                 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1142                                                  MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) |
1143                                                  PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
1144                                                  TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) |
1145                                                  NUM_BANKS(ADDR_SURF_16_BANK) |
1146                                                  BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1147                                                  BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1148                                                  MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1));
1149                                 break;
1150                         case 17:  /* Thin 64 bpp. */
1151                                 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1152                                                  MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) |
1153                                                  PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
1154                                                  TILE_SPLIT(split_equal_to_row_size) |
1155                                                  NUM_BANKS(ADDR_SURF_16_BANK) |
1156                                                  BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1157                                                  BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1158                                                  MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1));
1159                                 break;
1160                         case 21:  /* 8 bpp PRT. */
1161                                 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1162                                                  MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) |
1163                                                  PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
1164                                                  TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
1165                                                  NUM_BANKS(ADDR_SURF_16_BANK) |
1166                                                  BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
1167                                                  BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1168                                                  MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
1169                                 break;
1170                         case 22:  /* 16 bpp PRT */
1171                                 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1172                                                  MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) |
1173                                                  PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
1174                                                  TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
1175                                                  NUM_BANKS(ADDR_SURF_16_BANK) |
1176                                                  BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1177                                                  BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1178                                                  MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4));
1179                                 break;
1180                         case 23:  /* 32 bpp PRT */
1181                                 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1182                                                  MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) |
1183                                                  PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
1184                                                  TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
1185                                                  NUM_BANKS(ADDR_SURF_16_BANK) |
1186                                                  BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1187                                                  BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1188                                                  MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
1189                                 break;
1190                         case 24:  /* 64 bpp PRT */
1191                                 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1192                                                  MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) |
1193                                                  PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
1194                                                  TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) |
1195                                                  NUM_BANKS(ADDR_SURF_16_BANK) |
1196                                                  BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1197                                                  BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1198                                                  MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
1199                                 break;
1200                         case 25:  /* 128 bpp PRT */
1201                                 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1202                                                  MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) |
1203                                                  PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
1204                                                  TILE_SPLIT(ADDR_SURF_TILE_SPLIT_1KB) |
1205                                                  NUM_BANKS(ADDR_SURF_8_BANK) |
1206                                                  BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1207                                                  BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1208                                                  MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1));
1209                                 break;
1210                         default:
1211                                 gb_tile_moden = 0;
1212                                 break;
1213                         }
1214                         WREG32(GB_TILE_MODE0 + (reg_offset * 4), gb_tile_moden);
1215                 }
1216         } else if ((rdev->family == CHIP_VERDE) ||
1217                    (rdev->family == CHIP_OLAND)) {
1218                 for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++) {
1219                         switch (reg_offset) {
1220                         case 0:  /* non-AA compressed depth or any compressed stencil */
1221                                 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1222                                                  MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) |
1223                                                  PIPE_CONFIG(ADDR_SURF_P4_8x16) |
1224                                                  TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) |
1225                                                  NUM_BANKS(ADDR_SURF_16_BANK) |
1226                                                  BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1227                                                  BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1228                                                  MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4));
1229                                 break;
1230                         case 1:  /* 2xAA/4xAA compressed depth only */
1231                                 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1232                                                  MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) |
1233                                                  PIPE_CONFIG(ADDR_SURF_P4_8x16) |
1234                                                  TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) |
1235                                                  NUM_BANKS(ADDR_SURF_16_BANK) |
1236                                                  BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1237                                                  BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1238                                                  MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4));
1239                                 break;
1240                         case 2:  /* 8xAA compressed depth only */
1241                                 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1242                                                  MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) |
1243                                                  PIPE_CONFIG(ADDR_SURF_P4_8x16) |
1244                                                  TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
1245                                                  NUM_BANKS(ADDR_SURF_16_BANK) |
1246                                                  BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1247                                                  BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1248                                                  MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4));
1249                                 break;
1250                         case 3:  /* 2xAA/4xAA compressed depth with stencil (for depth buffer) */
1251                                 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1252                                                  MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) |
1253                                                  PIPE_CONFIG(ADDR_SURF_P4_8x16) |
1254                                                  TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) |
1255                                                  NUM_BANKS(ADDR_SURF_16_BANK) |
1256                                                  BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1257                                                  BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1258                                                  MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4));
1259                                 break;
1260                         case 4:  /* Maps w/ a dimension less than the 2D macro-tile dimensions (for mipmapped depth textures) */
1261                                 gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
1262                                                  MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) |
1263                                                  PIPE_CONFIG(ADDR_SURF_P4_8x16) |
1264                                                  TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) |
1265                                                  NUM_BANKS(ADDR_SURF_16_BANK) |
1266                                                  BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1267                                                  BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1268                                                  MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
1269                                 break;
1270                         case 5:  /* Uncompressed 16bpp depth - and stencil buffer allocated with it */
1271                                 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1272                                                  MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) |
1273                                                  PIPE_CONFIG(ADDR_SURF_P4_8x16) |
1274                                                  TILE_SPLIT(split_equal_to_row_size) |
1275                                                  NUM_BANKS(ADDR_SURF_16_BANK) |
1276                                                  BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1277                                                  BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1278                                                  MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
1279                                 break;
1280                         case 6:  /* Uncompressed 32bpp depth - and stencil buffer allocated with it */
1281                                 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1282                                                  MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) |
1283                                                  PIPE_CONFIG(ADDR_SURF_P4_8x16) |
1284                                                  TILE_SPLIT(split_equal_to_row_size) |
1285                                                  NUM_BANKS(ADDR_SURF_16_BANK) |
1286                                                  BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1287                                                  BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1288                                                  MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
1289                                 break;
1290                         case 7:  /* Uncompressed 8bpp stencil without depth (drivers typically do not use) */
1291                                 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1292                                                  MICRO_TILE_MODE(ADDR_SURF_DEPTH_MICRO_TILING) |
1293                                                  PIPE_CONFIG(ADDR_SURF_P4_8x16) |
1294                                                  TILE_SPLIT(split_equal_to_row_size) |
1295                                                  NUM_BANKS(ADDR_SURF_16_BANK) |
1296                                                  BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1297                                                  BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1298                                                  MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4));
1299                                 break;
1300                         case 8:  /* 1D and 1D Array Surfaces */
1301                                 gb_tile_moden = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) |
1302                                                  MICRO_TILE_MODE(ADDR_SURF_DISPLAY_MICRO_TILING) |
1303                                                  PIPE_CONFIG(ADDR_SURF_P4_8x16) |
1304                                                  TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) |
1305                                                  NUM_BANKS(ADDR_SURF_16_BANK) |
1306                                                  BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1307                                                  BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1308                                                  MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
1309                                 break;
1310                         case 9:  /* Displayable maps. */
1311                                 gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
1312                                                  MICRO_TILE_MODE(ADDR_SURF_DISPLAY_MICRO_TILING) |
1313                                                  PIPE_CONFIG(ADDR_SURF_P4_8x16) |
1314                                                  TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) |
1315                                                  NUM_BANKS(ADDR_SURF_16_BANK) |
1316                                                  BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1317                                                  BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1318                                                  MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
1319                                 break;
1320                         case 10:  /* Display 8bpp. */
1321                                 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1322                                                  MICRO_TILE_MODE(ADDR_SURF_DISPLAY_MICRO_TILING) |
1323                                                  PIPE_CONFIG(ADDR_SURF_P4_8x16) |
1324                                                  TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
1325                                                  NUM_BANKS(ADDR_SURF_16_BANK) |
1326                                                  BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1327                                                  BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1328                                                  MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4));
1329                                 break;
1330                         case 11:  /* Display 16bpp. */
1331                                 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1332                                                  MICRO_TILE_MODE(ADDR_SURF_DISPLAY_MICRO_TILING) |
1333                                                  PIPE_CONFIG(ADDR_SURF_P4_8x16) |
1334                                                  TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
1335                                                  NUM_BANKS(ADDR_SURF_16_BANK) |
1336                                                  BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1337                                                  BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1338                                                  MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
1339                                 break;
1340                         case 12:  /* Display 32bpp. */
1341                                 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1342                                                  MICRO_TILE_MODE(ADDR_SURF_DISPLAY_MICRO_TILING) |
1343                                                  PIPE_CONFIG(ADDR_SURF_P4_8x16) |
1344                                                  TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) |
1345                                                  NUM_BANKS(ADDR_SURF_16_BANK) |
1346                                                  BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1347                                                  BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1348                                                  MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
1349                                 break;
1350                         case 13:  /* Thin. */
1351                                 gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
1352                                                  MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) |
1353                                                  PIPE_CONFIG(ADDR_SURF_P4_8x16) |
1354                                                  TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) |
1355                                                  NUM_BANKS(ADDR_SURF_16_BANK) |
1356                                                  BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1357                                                  BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1358                                                  MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
1359                                 break;
1360                         case 14:  /* Thin 8 bpp. */
1361                                 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1362                                                  MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) |
1363                                                  PIPE_CONFIG(ADDR_SURF_P4_8x16) |
1364                                                  TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
1365                                                  NUM_BANKS(ADDR_SURF_16_BANK) |
1366                                                  BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1367                                                  BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1368                                                  MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
1369                                 break;
1370                         case 15:  /* Thin 16 bpp. */
1371                                 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1372                                                  MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) |
1373                                                  PIPE_CONFIG(ADDR_SURF_P4_8x16) |
1374                                                  TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
1375                                                  NUM_BANKS(ADDR_SURF_16_BANK) |
1376                                                  BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1377                                                  BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1378                                                  MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
1379                                 break;
1380                         case 16:  /* Thin 32 bpp. */
1381                                 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1382                                                  MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) |
1383                                                  PIPE_CONFIG(ADDR_SURF_P4_8x16) |
1384                                                  TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) |
1385                                                  NUM_BANKS(ADDR_SURF_16_BANK) |
1386                                                  BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1387                                                  BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1388                                                  MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
1389                                 break;
1390                         case 17:  /* Thin 64 bpp. */
1391                                 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1392                                                  MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) |
1393                                                  PIPE_CONFIG(ADDR_SURF_P4_8x16) |
1394                                                  TILE_SPLIT(split_equal_to_row_size) |
1395                                                  NUM_BANKS(ADDR_SURF_16_BANK) |
1396                                                  BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1397                                                  BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1398                                                  MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
1399                                 break;
1400                         case 21:  /* 8 bpp PRT. */
1401                                 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1402                                                  MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) |
1403                                                  PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
1404                                                  TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
1405                                                  NUM_BANKS(ADDR_SURF_16_BANK) |
1406                                                  BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
1407                                                  BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1408                                                  MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
1409                                 break;
1410                         case 22:  /* 16 bpp PRT */
1411                                 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1412                                                  MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) |
1413                                                  PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
1414                                                  TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
1415                                                  NUM_BANKS(ADDR_SURF_16_BANK) |
1416                                                  BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1417                                                  BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
1418                                                  MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4));
1419                                 break;
1420                         case 23:  /* 32 bpp PRT */
1421                                 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1422                                                  MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) |
1423                                                  PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
1424                                                  TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
1425                                                  NUM_BANKS(ADDR_SURF_16_BANK) |
1426                                                  BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1427                                                  BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
1428                                                  MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
1429                                 break;
1430                         case 24:  /* 64 bpp PRT */
1431                                 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1432                                                  MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) |
1433                                                  PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
1434                                                  TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) |
1435                                                  NUM_BANKS(ADDR_SURF_16_BANK) |
1436                                                  BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1437                                                  BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1438                                                  MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2));
1439                                 break;
1440                         case 25:  /* 128 bpp PRT */
1441                                 gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
1442                                                  MICRO_TILE_MODE(ADDR_SURF_THIN_MICRO_TILING) |
1443                                                  PIPE_CONFIG(ADDR_SURF_P8_32x32_8x16) |
1444                                                  TILE_SPLIT(ADDR_SURF_TILE_SPLIT_1KB) |
1445                                                  NUM_BANKS(ADDR_SURF_8_BANK) |
1446                                                  BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
1447                                                  BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
1448                                                  MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1));
1449                                 break;
1450                         default:
1451                                 gb_tile_moden = 0;
1452                                 break;
1453                         }
1454                         WREG32(GB_TILE_MODE0 + (reg_offset * 4), gb_tile_moden);
1455                 }
1456         } else
1457                 DRM_ERROR("unknown asic: 0x%x\n", rdev->family);
1458 }
1459
1460 static void si_select_se_sh(struct radeon_device *rdev,
1461                             u32 se_num, u32 sh_num)
1462 {
1463         u32 data = INSTANCE_BROADCAST_WRITES;
1464
1465         if ((se_num == 0xffffffff) && (sh_num == 0xffffffff))
1466                 data = SH_BROADCAST_WRITES | SE_BROADCAST_WRITES;
1467         else if (se_num == 0xffffffff)
1468                 data |= SE_BROADCAST_WRITES | SH_INDEX(sh_num);
1469         else if (sh_num == 0xffffffff)
1470                 data |= SH_BROADCAST_WRITES | SE_INDEX(se_num);
1471         else
1472                 data |= SH_INDEX(sh_num) | SE_INDEX(se_num);
1473         WREG32(GRBM_GFX_INDEX, data);
1474 }
1475
1476 static u32 si_create_bitmask(u32 bit_width)
1477 {
1478         u32 i, mask = 0;
1479
1480         for (i = 0; i < bit_width; i++) {
1481                 mask <<= 1;
1482                 mask |= 1;
1483         }
1484         return mask;
1485 }
1486
1487 static u32 si_get_cu_enabled(struct radeon_device *rdev, u32 cu_per_sh)
1488 {
1489         u32 data, mask;
1490
1491         data = RREG32(CC_GC_SHADER_ARRAY_CONFIG);
1492         if (data & 1)
1493                 data &= INACTIVE_CUS_MASK;
1494         else
1495                 data = 0;
1496         data |= RREG32(GC_USER_SHADER_ARRAY_CONFIG);
1497
1498         data >>= INACTIVE_CUS_SHIFT;
1499
1500         mask = si_create_bitmask(cu_per_sh);
1501
1502         return ~data & mask;
1503 }
1504
1505 static void si_setup_spi(struct radeon_device *rdev,
1506                          u32 se_num, u32 sh_per_se,
1507                          u32 cu_per_sh)
1508 {
1509         int i, j, k;
1510         u32 data, mask, active_cu;
1511
1512         for (i = 0; i < se_num; i++) {
1513                 for (j = 0; j < sh_per_se; j++) {
1514                         si_select_se_sh(rdev, i, j);
1515                         data = RREG32(SPI_STATIC_THREAD_MGMT_3);
1516                         active_cu = si_get_cu_enabled(rdev, cu_per_sh);
1517
1518                         mask = 1;
1519                         for (k = 0; k < 16; k++) {
1520                                 mask <<= k;
1521                                 if (active_cu & mask) {
1522                                         data &= ~mask;
1523                                         WREG32(SPI_STATIC_THREAD_MGMT_3, data);
1524                                         break;
1525                                 }
1526                         }
1527                 }
1528         }
1529         si_select_se_sh(rdev, 0xffffffff, 0xffffffff);
1530 }
1531
1532 static u32 si_get_rb_disabled(struct radeon_device *rdev,
1533                               u32 max_rb_num, u32 se_num,
1534                               u32 sh_per_se)
1535 {
1536         u32 data, mask;
1537
1538         data = RREG32(CC_RB_BACKEND_DISABLE);
1539         if (data & 1)
1540                 data &= BACKEND_DISABLE_MASK;
1541         else
1542                 data = 0;
1543         data |= RREG32(GC_USER_RB_BACKEND_DISABLE);
1544
1545         data >>= BACKEND_DISABLE_SHIFT;
1546
1547         mask = si_create_bitmask(max_rb_num / se_num / sh_per_se);
1548
1549         return data & mask;
1550 }
1551
1552 static void si_setup_rb(struct radeon_device *rdev,
1553                         u32 se_num, u32 sh_per_se,
1554                         u32 max_rb_num)
1555 {
1556         int i, j;
1557         u32 data, mask;
1558         u32 disabled_rbs = 0;
1559         u32 enabled_rbs = 0;
1560
1561         for (i = 0; i < se_num; i++) {
1562                 for (j = 0; j < sh_per_se; j++) {
1563                         si_select_se_sh(rdev, i, j);
1564                         data = si_get_rb_disabled(rdev, max_rb_num, se_num, sh_per_se);
1565                         disabled_rbs |= data << ((i * sh_per_se + j) * TAHITI_RB_BITMAP_WIDTH_PER_SH);
1566                 }
1567         }
1568         si_select_se_sh(rdev, 0xffffffff, 0xffffffff);
1569
1570         mask = 1;
1571         for (i = 0; i < max_rb_num; i++) {
1572                 if (!(disabled_rbs & mask))
1573                         enabled_rbs |= mask;
1574                 mask <<= 1;
1575         }
1576
1577         for (i = 0; i < se_num; i++) {
1578                 si_select_se_sh(rdev, i, 0xffffffff);
1579                 data = 0;
1580                 for (j = 0; j < sh_per_se; j++) {
1581                         switch (enabled_rbs & 3) {
1582                         case 1:
1583                                 data |= (RASTER_CONFIG_RB_MAP_0 << (i * sh_per_se + j) * 2);
1584                                 break;
1585                         case 2:
1586                                 data |= (RASTER_CONFIG_RB_MAP_3 << (i * sh_per_se + j) * 2);
1587                                 break;
1588                         case 3:
1589                         default:
1590                                 data |= (RASTER_CONFIG_RB_MAP_2 << (i * sh_per_se + j) * 2);
1591                                 break;
1592                         }
1593                         enabled_rbs >>= 2;
1594                 }
1595                 WREG32(PA_SC_RASTER_CONFIG, data);
1596         }
1597         si_select_se_sh(rdev, 0xffffffff, 0xffffffff);
1598 }
1599
1600 static void si_gpu_init(struct radeon_device *rdev)
1601 {
1602         u32 gb_addr_config = 0;
1603         u32 mc_shared_chmap, mc_arb_ramcfg;
1604         u32 sx_debug_1;
1605         u32 hdp_host_path_cntl;
1606         u32 tmp;
1607         int i, j;
1608
1609         switch (rdev->family) {
1610         case CHIP_TAHITI:
1611                 rdev->config.si.max_shader_engines = 2;
1612                 rdev->config.si.max_tile_pipes = 12;
1613                 rdev->config.si.max_cu_per_sh = 8;
1614                 rdev->config.si.max_sh_per_se = 2;
1615                 rdev->config.si.max_backends_per_se = 4;
1616                 rdev->config.si.max_texture_channel_caches = 12;
1617                 rdev->config.si.max_gprs = 256;
1618                 rdev->config.si.max_gs_threads = 32;
1619                 rdev->config.si.max_hw_contexts = 8;
1620
1621                 rdev->config.si.sc_prim_fifo_size_frontend = 0x20;
1622                 rdev->config.si.sc_prim_fifo_size_backend = 0x100;
1623                 rdev->config.si.sc_hiz_tile_fifo_size = 0x30;
1624                 rdev->config.si.sc_earlyz_tile_fifo_size = 0x130;
1625                 gb_addr_config = TAHITI_GB_ADDR_CONFIG_GOLDEN;
1626                 break;
1627         case CHIP_PITCAIRN:
1628                 rdev->config.si.max_shader_engines = 2;
1629                 rdev->config.si.max_tile_pipes = 8;
1630                 rdev->config.si.max_cu_per_sh = 5;
1631                 rdev->config.si.max_sh_per_se = 2;
1632                 rdev->config.si.max_backends_per_se = 4;
1633                 rdev->config.si.max_texture_channel_caches = 8;
1634                 rdev->config.si.max_gprs = 256;
1635                 rdev->config.si.max_gs_threads = 32;
1636                 rdev->config.si.max_hw_contexts = 8;
1637
1638                 rdev->config.si.sc_prim_fifo_size_frontend = 0x20;
1639                 rdev->config.si.sc_prim_fifo_size_backend = 0x100;
1640                 rdev->config.si.sc_hiz_tile_fifo_size = 0x30;
1641                 rdev->config.si.sc_earlyz_tile_fifo_size = 0x130;
1642                 gb_addr_config = TAHITI_GB_ADDR_CONFIG_GOLDEN;
1643                 break;
1644         case CHIP_VERDE:
1645         default:
1646                 rdev->config.si.max_shader_engines = 1;
1647                 rdev->config.si.max_tile_pipes = 4;
1648                 rdev->config.si.max_cu_per_sh = 2;
1649                 rdev->config.si.max_sh_per_se = 2;
1650                 rdev->config.si.max_backends_per_se = 4;
1651                 rdev->config.si.max_texture_channel_caches = 4;
1652                 rdev->config.si.max_gprs = 256;
1653                 rdev->config.si.max_gs_threads = 32;
1654                 rdev->config.si.max_hw_contexts = 8;
1655
1656                 rdev->config.si.sc_prim_fifo_size_frontend = 0x20;
1657                 rdev->config.si.sc_prim_fifo_size_backend = 0x40;
1658                 rdev->config.si.sc_hiz_tile_fifo_size = 0x30;
1659                 rdev->config.si.sc_earlyz_tile_fifo_size = 0x130;
1660                 gb_addr_config = VERDE_GB_ADDR_CONFIG_GOLDEN;
1661                 break;
1662         case CHIP_OLAND:
1663                 rdev->config.si.max_shader_engines = 1;
1664                 rdev->config.si.max_tile_pipes = 4;
1665                 rdev->config.si.max_cu_per_sh = 6;
1666                 rdev->config.si.max_sh_per_se = 1;
1667                 rdev->config.si.max_backends_per_se = 2;
1668                 rdev->config.si.max_texture_channel_caches = 4;
1669                 rdev->config.si.max_gprs = 256;
1670                 rdev->config.si.max_gs_threads = 16;
1671                 rdev->config.si.max_hw_contexts = 8;
1672
1673                 rdev->config.si.sc_prim_fifo_size_frontend = 0x20;
1674                 rdev->config.si.sc_prim_fifo_size_backend = 0x40;
1675                 rdev->config.si.sc_hiz_tile_fifo_size = 0x30;
1676                 rdev->config.si.sc_earlyz_tile_fifo_size = 0x130;
1677                 gb_addr_config = VERDE_GB_ADDR_CONFIG_GOLDEN;
1678                 break;
1679         }
1680
1681         /* Initialize HDP */
1682         for (i = 0, j = 0; i < 32; i++, j += 0x18) {
1683                 WREG32((0x2c14 + j), 0x00000000);
1684                 WREG32((0x2c18 + j), 0x00000000);
1685                 WREG32((0x2c1c + j), 0x00000000);
1686                 WREG32((0x2c20 + j), 0x00000000);
1687                 WREG32((0x2c24 + j), 0x00000000);
1688         }
1689
1690         WREG32(GRBM_CNTL, GRBM_READ_TIMEOUT(0xff));
1691
1692         evergreen_fix_pci_max_read_req_size(rdev);
1693
1694         WREG32(BIF_FB_EN, FB_READ_EN | FB_WRITE_EN);
1695
1696         mc_shared_chmap = RREG32(MC_SHARED_CHMAP);
1697         mc_arb_ramcfg = RREG32(MC_ARB_RAMCFG);
1698
1699         rdev->config.si.num_tile_pipes = rdev->config.si.max_tile_pipes;
1700         rdev->config.si.mem_max_burst_length_bytes = 256;
1701         tmp = (mc_arb_ramcfg & NOOFCOLS_MASK) >> NOOFCOLS_SHIFT;
1702         rdev->config.si.mem_row_size_in_kb = (4 * (1 << (8 + tmp))) / 1024;
1703         if (rdev->config.si.mem_row_size_in_kb > 4)
1704                 rdev->config.si.mem_row_size_in_kb = 4;
1705         /* XXX use MC settings? */
1706         rdev->config.si.shader_engine_tile_size = 32;
1707         rdev->config.si.num_gpus = 1;
1708         rdev->config.si.multi_gpu_tile_size = 64;
1709
1710         /* fix up row size */
1711         gb_addr_config &= ~ROW_SIZE_MASK;
1712         switch (rdev->config.si.mem_row_size_in_kb) {
1713         case 1:
1714         default:
1715                 gb_addr_config |= ROW_SIZE(0);
1716                 break;
1717         case 2:
1718                 gb_addr_config |= ROW_SIZE(1);
1719                 break;
1720         case 4:
1721                 gb_addr_config |= ROW_SIZE(2);
1722                 break;
1723         }
1724
1725         /* setup tiling info dword.  gb_addr_config is not adequate since it does
1726          * not have bank info, so create a custom tiling dword.
1727          * bits 3:0   num_pipes
1728          * bits 7:4   num_banks
1729          * bits 11:8  group_size
1730          * bits 15:12 row_size
1731          */
1732         rdev->config.si.tile_config = 0;
1733         switch (rdev->config.si.num_tile_pipes) {
1734         case 1:
1735                 rdev->config.si.tile_config |= (0 << 0);
1736                 break;
1737         case 2:
1738                 rdev->config.si.tile_config |= (1 << 0);
1739                 break;
1740         case 4:
1741                 rdev->config.si.tile_config |= (2 << 0);
1742                 break;
1743         case 8:
1744         default:
1745                 /* XXX what about 12? */
1746                 rdev->config.si.tile_config |= (3 << 0);
1747                 break;
1748         }
1749         switch ((mc_arb_ramcfg & NOOFBANK_MASK) >> NOOFBANK_SHIFT) {
1750         case 0: /* four banks */
1751                 rdev->config.si.tile_config |= 0 << 4;
1752                 break;
1753         case 1: /* eight banks */
1754                 rdev->config.si.tile_config |= 1 << 4;
1755                 break;
1756         case 2: /* sixteen banks */
1757         default:
1758                 rdev->config.si.tile_config |= 2 << 4;
1759                 break;
1760         }
1761         rdev->config.si.tile_config |=
1762                 ((gb_addr_config & PIPE_INTERLEAVE_SIZE_MASK) >> PIPE_INTERLEAVE_SIZE_SHIFT) << 8;
1763         rdev->config.si.tile_config |=
1764                 ((gb_addr_config & ROW_SIZE_MASK) >> ROW_SIZE_SHIFT) << 12;
1765
1766         WREG32(GB_ADDR_CONFIG, gb_addr_config);
1767         WREG32(DMIF_ADDR_CONFIG, gb_addr_config);
1768         WREG32(HDP_ADDR_CONFIG, gb_addr_config);
1769         WREG32(DMA_TILING_CONFIG + DMA0_REGISTER_OFFSET, gb_addr_config);
1770         WREG32(DMA_TILING_CONFIG + DMA1_REGISTER_OFFSET, gb_addr_config);
1771
1772         si_tiling_mode_table_init(rdev);
1773
1774         si_setup_rb(rdev, rdev->config.si.max_shader_engines,
1775                     rdev->config.si.max_sh_per_se,
1776                     rdev->config.si.max_backends_per_se);
1777
1778         si_setup_spi(rdev, rdev->config.si.max_shader_engines,
1779                      rdev->config.si.max_sh_per_se,
1780                      rdev->config.si.max_cu_per_sh);
1781
1782
1783         /* set HW defaults for 3D engine */
1784         WREG32(CP_QUEUE_THRESHOLDS, (ROQ_IB1_START(0x16) |
1785                                      ROQ_IB2_START(0x2b)));
1786         WREG32(CP_MEQ_THRESHOLDS, MEQ1_START(0x30) | MEQ2_START(0x60));
1787
1788         sx_debug_1 = RREG32(SX_DEBUG_1);
1789         WREG32(SX_DEBUG_1, sx_debug_1);
1790
1791         WREG32(SPI_CONFIG_CNTL_1, VTX_DONE_DELAY(4));
1792
1793         WREG32(PA_SC_FIFO_SIZE, (SC_FRONTEND_PRIM_FIFO_SIZE(rdev->config.si.sc_prim_fifo_size_frontend) |
1794                                  SC_BACKEND_PRIM_FIFO_SIZE(rdev->config.si.sc_prim_fifo_size_backend) |
1795                                  SC_HIZ_TILE_FIFO_SIZE(rdev->config.si.sc_hiz_tile_fifo_size) |
1796                                  SC_EARLYZ_TILE_FIFO_SIZE(rdev->config.si.sc_earlyz_tile_fifo_size)));
1797
1798         WREG32(VGT_NUM_INSTANCES, 1);
1799
1800         WREG32(CP_PERFMON_CNTL, 0);
1801
1802         WREG32(SQ_CONFIG, 0);
1803
1804         WREG32(PA_SC_FORCE_EOV_MAX_CNTS, (FORCE_EOV_MAX_CLK_CNT(4095) |
1805                                           FORCE_EOV_MAX_REZ_CNT(255)));
1806
1807         WREG32(VGT_CACHE_INVALIDATION, CACHE_INVALIDATION(VC_AND_TC) |
1808                AUTO_INVLD_EN(ES_AND_GS_AUTO));
1809
1810         WREG32(VGT_GS_VERTEX_REUSE, 16);
1811         WREG32(PA_SC_LINE_STIPPLE_STATE, 0);
1812
1813         WREG32(CB_PERFCOUNTER0_SELECT0, 0);
1814         WREG32(CB_PERFCOUNTER0_SELECT1, 0);
1815         WREG32(CB_PERFCOUNTER1_SELECT0, 0);
1816         WREG32(CB_PERFCOUNTER1_SELECT1, 0);
1817         WREG32(CB_PERFCOUNTER2_SELECT0, 0);
1818         WREG32(CB_PERFCOUNTER2_SELECT1, 0);
1819         WREG32(CB_PERFCOUNTER3_SELECT0, 0);
1820         WREG32(CB_PERFCOUNTER3_SELECT1, 0);
1821
1822         tmp = RREG32(HDP_MISC_CNTL);
1823         tmp |= HDP_FLUSH_INVALIDATE_CACHE;
1824         WREG32(HDP_MISC_CNTL, tmp);
1825
1826         hdp_host_path_cntl = RREG32(HDP_HOST_PATH_CNTL);
1827         WREG32(HDP_HOST_PATH_CNTL, hdp_host_path_cntl);
1828
1829         WREG32(PA_CL_ENHANCE, CLIP_VTX_REORDER_ENA | NUM_CLIP_SEQ(3));
1830
1831         udelay(50);
1832 }
1833
1834 /*
1835  * GPU scratch registers helpers function.
1836  */
1837 static void si_scratch_init(struct radeon_device *rdev)
1838 {
1839         int i;
1840
1841         rdev->scratch.num_reg = 7;
1842         rdev->scratch.reg_base = SCRATCH_REG0;
1843         for (i = 0; i < rdev->scratch.num_reg; i++) {
1844                 rdev->scratch.free[i] = true;
1845                 rdev->scratch.reg[i] = rdev->scratch.reg_base + (i * 4);
1846         }
1847 }
1848
1849 void si_fence_ring_emit(struct radeon_device *rdev,
1850                         struct radeon_fence *fence)
1851 {
1852         struct radeon_ring *ring = &rdev->ring[fence->ring];
1853         u64 addr = rdev->fence_drv[fence->ring].gpu_addr;
1854
1855         /* flush read cache over gart */
1856         radeon_ring_write(ring, PACKET3(PACKET3_SET_CONFIG_REG, 1));
1857         radeon_ring_write(ring, (CP_COHER_CNTL2 - PACKET3_SET_CONFIG_REG_START) >> 2);
1858         radeon_ring_write(ring, 0);
1859         radeon_ring_write(ring, PACKET3(PACKET3_SURFACE_SYNC, 3));
1860         radeon_ring_write(ring, PACKET3_TCL1_ACTION_ENA |
1861                           PACKET3_TC_ACTION_ENA |
1862                           PACKET3_SH_KCACHE_ACTION_ENA |
1863                           PACKET3_SH_ICACHE_ACTION_ENA);
1864         radeon_ring_write(ring, 0xFFFFFFFF);
1865         radeon_ring_write(ring, 0);
1866         radeon_ring_write(ring, 10); /* poll interval */
1867         /* EVENT_WRITE_EOP - flush caches, send int */
1868         radeon_ring_write(ring, PACKET3(PACKET3_EVENT_WRITE_EOP, 4));
1869         radeon_ring_write(ring, EVENT_TYPE(CACHE_FLUSH_AND_INV_TS_EVENT) | EVENT_INDEX(5));
1870         radeon_ring_write(ring, addr & 0xffffffff);
1871         radeon_ring_write(ring, (upper_32_bits(addr) & 0xff) | DATA_SEL(1) | INT_SEL(2));
1872         radeon_ring_write(ring, fence->seq);
1873         radeon_ring_write(ring, 0);
1874 }
1875
1876 /*
1877  * IB stuff
1878  */
1879 void si_ring_ib_execute(struct radeon_device *rdev, struct radeon_ib *ib)
1880 {
1881         struct radeon_ring *ring = &rdev->ring[ib->ring];
1882         u32 header;
1883
1884         if (ib->is_const_ib) {
1885                 /* set switch buffer packet before const IB */
1886                 radeon_ring_write(ring, PACKET3(PACKET3_SWITCH_BUFFER, 0));
1887                 radeon_ring_write(ring, 0);
1888
1889                 header = PACKET3(PACKET3_INDIRECT_BUFFER_CONST, 2);
1890         } else {
1891                 u32 next_rptr;
1892                 if (ring->rptr_save_reg) {
1893                         next_rptr = ring->wptr + 3 + 4 + 8;
1894                         radeon_ring_write(ring, PACKET3(PACKET3_SET_CONFIG_REG, 1));
1895                         radeon_ring_write(ring, ((ring->rptr_save_reg -
1896                                                   PACKET3_SET_CONFIG_REG_START) >> 2));
1897                         radeon_ring_write(ring, next_rptr);
1898                 } else if (rdev->wb.enabled) {
1899                         next_rptr = ring->wptr + 5 + 4 + 8;
1900                         radeon_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
1901                         radeon_ring_write(ring, (1 << 8));
1902                         radeon_ring_write(ring, ring->next_rptr_gpu_addr & 0xfffffffc);
1903                         radeon_ring_write(ring, upper_32_bits(ring->next_rptr_gpu_addr) & 0xffffffff);
1904                         radeon_ring_write(ring, next_rptr);
1905                 }
1906
1907                 header = PACKET3(PACKET3_INDIRECT_BUFFER, 2);
1908         }
1909
1910         radeon_ring_write(ring, header);
1911         radeon_ring_write(ring,
1912 #ifdef __BIG_ENDIAN
1913                           (2 << 0) |
1914 #endif
1915                           (ib->gpu_addr & 0xFFFFFFFC));
1916         radeon_ring_write(ring, upper_32_bits(ib->gpu_addr) & 0xFFFF);
1917         radeon_ring_write(ring, ib->length_dw |
1918                           (ib->vm ? (ib->vm->id << 24) : 0));
1919
1920         if (!ib->is_const_ib) {
1921                 /* flush read cache over gart for this vmid */
1922                 radeon_ring_write(ring, PACKET3(PACKET3_SET_CONFIG_REG, 1));
1923                 radeon_ring_write(ring, (CP_COHER_CNTL2 - PACKET3_SET_CONFIG_REG_START) >> 2);
1924                 radeon_ring_write(ring, ib->vm ? ib->vm->id : 0);
1925                 radeon_ring_write(ring, PACKET3(PACKET3_SURFACE_SYNC, 3));
1926                 radeon_ring_write(ring, PACKET3_TCL1_ACTION_ENA |
1927                                   PACKET3_TC_ACTION_ENA |
1928                                   PACKET3_SH_KCACHE_ACTION_ENA |
1929                                   PACKET3_SH_ICACHE_ACTION_ENA);
1930                 radeon_ring_write(ring, 0xFFFFFFFF);
1931                 radeon_ring_write(ring, 0);
1932                 radeon_ring_write(ring, 10); /* poll interval */
1933         }
1934 }
1935
1936 /*
1937  * CP.
1938  */
1939 static void si_cp_enable(struct radeon_device *rdev, bool enable)
1940 {
1941         if (enable)
1942                 WREG32(CP_ME_CNTL, 0);
1943         else {
1944                 radeon_ttm_set_active_vram_size(rdev, rdev->mc.visible_vram_size);
1945                 WREG32(CP_ME_CNTL, (CP_ME_HALT | CP_PFP_HALT | CP_CE_HALT));
1946                 WREG32(SCRATCH_UMSK, 0);
1947                 rdev->ring[RADEON_RING_TYPE_GFX_INDEX].ready = false;
1948                 rdev->ring[CAYMAN_RING_TYPE_CP1_INDEX].ready = false;
1949                 rdev->ring[CAYMAN_RING_TYPE_CP2_INDEX].ready = false;
1950         }
1951         udelay(50);
1952 }
1953
1954 static int si_cp_load_microcode(struct radeon_device *rdev)
1955 {
1956         const __be32 *fw_data;
1957         int i;
1958
1959         if (!rdev->me_fw || !rdev->pfp_fw)
1960                 return -EINVAL;
1961
1962         si_cp_enable(rdev, false);
1963
1964         /* PFP */
1965         fw_data = (const __be32 *)rdev->pfp_fw->data;
1966         WREG32(CP_PFP_UCODE_ADDR, 0);
1967         for (i = 0; i < SI_PFP_UCODE_SIZE; i++)
1968                 WREG32(CP_PFP_UCODE_DATA, be32_to_cpup(fw_data++));
1969         WREG32(CP_PFP_UCODE_ADDR, 0);
1970
1971         /* CE */
1972         fw_data = (const __be32 *)rdev->ce_fw->data;
1973         WREG32(CP_CE_UCODE_ADDR, 0);
1974         for (i = 0; i < SI_CE_UCODE_SIZE; i++)
1975                 WREG32(CP_CE_UCODE_DATA, be32_to_cpup(fw_data++));
1976         WREG32(CP_CE_UCODE_ADDR, 0);
1977
1978         /* ME */
1979         fw_data = (const __be32 *)rdev->me_fw->data;
1980         WREG32(CP_ME_RAM_WADDR, 0);
1981         for (i = 0; i < SI_PM4_UCODE_SIZE; i++)
1982                 WREG32(CP_ME_RAM_DATA, be32_to_cpup(fw_data++));
1983         WREG32(CP_ME_RAM_WADDR, 0);
1984
1985         WREG32(CP_PFP_UCODE_ADDR, 0);
1986         WREG32(CP_CE_UCODE_ADDR, 0);
1987         WREG32(CP_ME_RAM_WADDR, 0);
1988         WREG32(CP_ME_RAM_RADDR, 0);
1989         return 0;
1990 }
1991
1992 static int si_cp_start(struct radeon_device *rdev)
1993 {
1994         struct radeon_ring *ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
1995         int r, i;
1996
1997         r = radeon_ring_lock(rdev, ring, 7 + 4);
1998         if (r) {
1999                 DRM_ERROR("radeon: cp failed to lock ring (%d).\n", r);
2000                 return r;
2001         }
2002         /* init the CP */
2003         radeon_ring_write(ring, PACKET3(PACKET3_ME_INITIALIZE, 5));
2004         radeon_ring_write(ring, 0x1);
2005         radeon_ring_write(ring, 0x0);
2006         radeon_ring_write(ring, rdev->config.si.max_hw_contexts - 1);
2007         radeon_ring_write(ring, PACKET3_ME_INITIALIZE_DEVICE_ID(1));
2008         radeon_ring_write(ring, 0);
2009         radeon_ring_write(ring, 0);
2010
2011         /* init the CE partitions */
2012         radeon_ring_write(ring, PACKET3(PACKET3_SET_BASE, 2));
2013         radeon_ring_write(ring, PACKET3_BASE_INDEX(CE_PARTITION_BASE));
2014         radeon_ring_write(ring, 0xc000);
2015         radeon_ring_write(ring, 0xe000);
2016         radeon_ring_unlock_commit(rdev, ring);
2017
2018         si_cp_enable(rdev, true);
2019
2020         r = radeon_ring_lock(rdev, ring, si_default_size + 10);
2021         if (r) {
2022                 DRM_ERROR("radeon: cp failed to lock ring (%d).\n", r);
2023                 return r;
2024         }
2025
2026         /* setup clear context state */
2027         radeon_ring_write(ring, PACKET3(PACKET3_PREAMBLE_CNTL, 0));
2028         radeon_ring_write(ring, PACKET3_PREAMBLE_BEGIN_CLEAR_STATE);
2029
2030         for (i = 0; i < si_default_size; i++)
2031                 radeon_ring_write(ring, si_default_state[i]);
2032
2033         radeon_ring_write(ring, PACKET3(PACKET3_PREAMBLE_CNTL, 0));
2034         radeon_ring_write(ring, PACKET3_PREAMBLE_END_CLEAR_STATE);
2035
2036         /* set clear context state */
2037         radeon_ring_write(ring, PACKET3(PACKET3_CLEAR_STATE, 0));
2038         radeon_ring_write(ring, 0);
2039
2040         radeon_ring_write(ring, PACKET3(PACKET3_SET_CONTEXT_REG, 2));
2041         radeon_ring_write(ring, 0x00000316);
2042         radeon_ring_write(ring, 0x0000000e); /* VGT_VERTEX_REUSE_BLOCK_CNTL */
2043         radeon_ring_write(ring, 0x00000010); /* VGT_OUT_DEALLOC_CNTL */
2044
2045         radeon_ring_unlock_commit(rdev, ring);
2046
2047         for (i = RADEON_RING_TYPE_GFX_INDEX; i <= CAYMAN_RING_TYPE_CP2_INDEX; ++i) {
2048                 ring = &rdev->ring[i];
2049                 r = radeon_ring_lock(rdev, ring, 2);
2050
2051                 /* clear the compute context state */
2052                 radeon_ring_write(ring, PACKET3_COMPUTE(PACKET3_CLEAR_STATE, 0));
2053                 radeon_ring_write(ring, 0);
2054
2055                 radeon_ring_unlock_commit(rdev, ring);
2056         }
2057
2058         return 0;
2059 }
2060
2061 static void si_cp_fini(struct radeon_device *rdev)
2062 {
2063         struct radeon_ring *ring;
2064         si_cp_enable(rdev, false);
2065
2066         ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
2067         radeon_ring_fini(rdev, ring);
2068         radeon_scratch_free(rdev, ring->rptr_save_reg);
2069
2070         ring = &rdev->ring[CAYMAN_RING_TYPE_CP1_INDEX];
2071         radeon_ring_fini(rdev, ring);
2072         radeon_scratch_free(rdev, ring->rptr_save_reg);
2073
2074         ring = &rdev->ring[CAYMAN_RING_TYPE_CP2_INDEX];
2075         radeon_ring_fini(rdev, ring);
2076         radeon_scratch_free(rdev, ring->rptr_save_reg);
2077 }
2078
2079 static int si_cp_resume(struct radeon_device *rdev)
2080 {
2081         struct radeon_ring *ring;
2082         u32 tmp;
2083         u32 rb_bufsz;
2084         int r;
2085
2086         /* Reset cp; if cp is reset, then PA, SH, VGT also need to be reset */
2087         WREG32(GRBM_SOFT_RESET, (SOFT_RESET_CP |
2088                                  SOFT_RESET_PA |
2089                                  SOFT_RESET_VGT |
2090                                  SOFT_RESET_SPI |
2091                                  SOFT_RESET_SX));
2092         RREG32(GRBM_SOFT_RESET);
2093         mdelay(15);
2094         WREG32(GRBM_SOFT_RESET, 0);
2095         RREG32(GRBM_SOFT_RESET);
2096
2097         WREG32(CP_SEM_WAIT_TIMER, 0x0);
2098         WREG32(CP_SEM_INCOMPLETE_TIMER_CNTL, 0x0);
2099
2100         /* Set the write pointer delay */
2101         WREG32(CP_RB_WPTR_DELAY, 0);
2102
2103         WREG32(CP_DEBUG, 0);
2104         WREG32(SCRATCH_ADDR, ((rdev->wb.gpu_addr + RADEON_WB_SCRATCH_OFFSET) >> 8) & 0xFFFFFFFF);
2105
2106         /* ring 0 - compute and gfx */
2107         /* Set ring buffer size */
2108         ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
2109         rb_bufsz = drm_order(ring->ring_size / 8);
2110         tmp = (drm_order(RADEON_GPU_PAGE_SIZE/8) << 8) | rb_bufsz;
2111 #ifdef __BIG_ENDIAN
2112         tmp |= BUF_SWAP_32BIT;
2113 #endif
2114         WREG32(CP_RB0_CNTL, tmp);
2115
2116         /* Initialize the ring buffer's read and write pointers */
2117         WREG32(CP_RB0_CNTL, tmp | RB_RPTR_WR_ENA);
2118         ring->wptr = 0;
2119         WREG32(CP_RB0_WPTR, ring->wptr);
2120
2121         /* set the wb address whether it's enabled or not */
2122         WREG32(CP_RB0_RPTR_ADDR, (rdev->wb.gpu_addr + RADEON_WB_CP_RPTR_OFFSET) & 0xFFFFFFFC);
2123         WREG32(CP_RB0_RPTR_ADDR_HI, upper_32_bits(rdev->wb.gpu_addr + RADEON_WB_CP_RPTR_OFFSET) & 0xFF);
2124
2125         if (rdev->wb.enabled)
2126                 WREG32(SCRATCH_UMSK, 0xff);
2127         else {
2128                 tmp |= RB_NO_UPDATE;
2129                 WREG32(SCRATCH_UMSK, 0);
2130         }
2131
2132         mdelay(1);
2133         WREG32(CP_RB0_CNTL, tmp);
2134
2135         WREG32(CP_RB0_BASE, ring->gpu_addr >> 8);
2136
2137         ring->rptr = RREG32(CP_RB0_RPTR);
2138
2139         /* ring1  - compute only */
2140         /* Set ring buffer size */
2141         ring = &rdev->ring[CAYMAN_RING_TYPE_CP1_INDEX];
2142         rb_bufsz = drm_order(ring->ring_size / 8);
2143         tmp = (drm_order(RADEON_GPU_PAGE_SIZE/8) << 8) | rb_bufsz;
2144 #ifdef __BIG_ENDIAN
2145         tmp |= BUF_SWAP_32BIT;
2146 #endif
2147         WREG32(CP_RB1_CNTL, tmp);
2148
2149         /* Initialize the ring buffer's read and write pointers */
2150         WREG32(CP_RB1_CNTL, tmp | RB_RPTR_WR_ENA);
2151         ring->wptr = 0;
2152         WREG32(CP_RB1_WPTR, ring->wptr);
2153
2154         /* set the wb address whether it's enabled or not */
2155         WREG32(CP_RB1_RPTR_ADDR, (rdev->wb.gpu_addr + RADEON_WB_CP1_RPTR_OFFSET) & 0xFFFFFFFC);
2156         WREG32(CP_RB1_RPTR_ADDR_HI, upper_32_bits(rdev->wb.gpu_addr + RADEON_WB_CP1_RPTR_OFFSET) & 0xFF);
2157
2158         mdelay(1);
2159         WREG32(CP_RB1_CNTL, tmp);
2160
2161         WREG32(CP_RB1_BASE, ring->gpu_addr >> 8);
2162
2163         ring->rptr = RREG32(CP_RB1_RPTR);
2164
2165         /* ring2 - compute only */
2166         /* Set ring buffer size */
2167         ring = &rdev->ring[CAYMAN_RING_TYPE_CP2_INDEX];
2168         rb_bufsz = drm_order(ring->ring_size / 8);
2169         tmp = (drm_order(RADEON_GPU_PAGE_SIZE/8) << 8) | rb_bufsz;
2170 #ifdef __BIG_ENDIAN
2171         tmp |= BUF_SWAP_32BIT;
2172 #endif
2173         WREG32(CP_RB2_CNTL, tmp);
2174
2175         /* Initialize the ring buffer's read and write pointers */
2176         WREG32(CP_RB2_CNTL, tmp | RB_RPTR_WR_ENA);
2177         ring->wptr = 0;
2178         WREG32(CP_RB2_WPTR, ring->wptr);
2179
2180         /* set the wb address whether it's enabled or not */
2181         WREG32(CP_RB2_RPTR_ADDR, (rdev->wb.gpu_addr + RADEON_WB_CP2_RPTR_OFFSET) & 0xFFFFFFFC);
2182         WREG32(CP_RB2_RPTR_ADDR_HI, upper_32_bits(rdev->wb.gpu_addr + RADEON_WB_CP2_RPTR_OFFSET) & 0xFF);
2183
2184         mdelay(1);
2185         WREG32(CP_RB2_CNTL, tmp);
2186
2187         WREG32(CP_RB2_BASE, ring->gpu_addr >> 8);
2188
2189         ring->rptr = RREG32(CP_RB2_RPTR);
2190
2191         /* start the rings */
2192         si_cp_start(rdev);
2193         rdev->ring[RADEON_RING_TYPE_GFX_INDEX].ready = true;
2194         rdev->ring[CAYMAN_RING_TYPE_CP1_INDEX].ready = true;
2195         rdev->ring[CAYMAN_RING_TYPE_CP2_INDEX].ready = true;
2196         r = radeon_ring_test(rdev, RADEON_RING_TYPE_GFX_INDEX, &rdev->ring[RADEON_RING_TYPE_GFX_INDEX]);
2197         if (r) {
2198                 rdev->ring[RADEON_RING_TYPE_GFX_INDEX].ready = false;
2199                 rdev->ring[CAYMAN_RING_TYPE_CP1_INDEX].ready = false;
2200                 rdev->ring[CAYMAN_RING_TYPE_CP2_INDEX].ready = false;
2201                 return r;
2202         }
2203         r = radeon_ring_test(rdev, CAYMAN_RING_TYPE_CP1_INDEX, &rdev->ring[CAYMAN_RING_TYPE_CP1_INDEX]);
2204         if (r) {
2205                 rdev->ring[CAYMAN_RING_TYPE_CP1_INDEX].ready = false;
2206         }
2207         r = radeon_ring_test(rdev, CAYMAN_RING_TYPE_CP2_INDEX, &rdev->ring[CAYMAN_RING_TYPE_CP2_INDEX]);
2208         if (r) {
2209                 rdev->ring[CAYMAN_RING_TYPE_CP2_INDEX].ready = false;
2210         }
2211
2212         return 0;
2213 }
2214
2215 static u32 si_gpu_check_soft_reset(struct radeon_device *rdev)
2216 {
2217         u32 reset_mask = 0;
2218         u32 tmp;
2219
2220         /* GRBM_STATUS */
2221         tmp = RREG32(GRBM_STATUS);
2222         if (tmp & (PA_BUSY | SC_BUSY |
2223                    BCI_BUSY | SX_BUSY |
2224                    TA_BUSY | VGT_BUSY |
2225                    DB_BUSY | CB_BUSY |
2226                    GDS_BUSY | SPI_BUSY |
2227                    IA_BUSY | IA_BUSY_NO_DMA))
2228                 reset_mask |= RADEON_RESET_GFX;
2229
2230         if (tmp & (CF_RQ_PENDING | PF_RQ_PENDING |
2231                    CP_BUSY | CP_COHERENCY_BUSY))
2232                 reset_mask |= RADEON_RESET_CP;
2233
2234         if (tmp & GRBM_EE_BUSY)
2235                 reset_mask |= RADEON_RESET_GRBM | RADEON_RESET_GFX | RADEON_RESET_CP;
2236
2237         /* GRBM_STATUS2 */
2238         tmp = RREG32(GRBM_STATUS2);
2239         if (tmp & (RLC_RQ_PENDING | RLC_BUSY))
2240                 reset_mask |= RADEON_RESET_RLC;
2241
2242         /* DMA_STATUS_REG 0 */
2243         tmp = RREG32(DMA_STATUS_REG + DMA0_REGISTER_OFFSET);
2244         if (!(tmp & DMA_IDLE))
2245                 reset_mask |= RADEON_RESET_DMA;
2246
2247         /* DMA_STATUS_REG 1 */
2248         tmp = RREG32(DMA_STATUS_REG + DMA1_REGISTER_OFFSET);
2249         if (!(tmp & DMA_IDLE))
2250                 reset_mask |= RADEON_RESET_DMA1;
2251
2252         /* SRBM_STATUS2 */
2253         tmp = RREG32(SRBM_STATUS2);
2254         if (tmp & DMA_BUSY)
2255                 reset_mask |= RADEON_RESET_DMA;
2256
2257         if (tmp & DMA1_BUSY)
2258                 reset_mask |= RADEON_RESET_DMA1;
2259
2260         /* SRBM_STATUS */
2261         tmp = RREG32(SRBM_STATUS);
2262
2263         if (tmp & IH_BUSY)
2264                 reset_mask |= RADEON_RESET_IH;
2265
2266         if (tmp & SEM_BUSY)
2267                 reset_mask |= RADEON_RESET_SEM;
2268
2269         if (tmp & GRBM_RQ_PENDING)
2270                 reset_mask |= RADEON_RESET_GRBM;
2271
2272         if (tmp & VMC_BUSY)
2273                 reset_mask |= RADEON_RESET_VMC;
2274
2275         if (tmp & (MCB_BUSY | MCB_NON_DISPLAY_BUSY |
2276                    MCC_BUSY | MCD_BUSY))
2277                 reset_mask |= RADEON_RESET_MC;
2278
2279         if (evergreen_is_display_hung(rdev))
2280                 reset_mask |= RADEON_RESET_DISPLAY;
2281
2282         /* VM_L2_STATUS */
2283         tmp = RREG32(VM_L2_STATUS);
2284         if (tmp & L2_BUSY)
2285                 reset_mask |= RADEON_RESET_VMC;
2286
2287         return reset_mask;
2288 }
2289
2290 static void si_gpu_soft_reset(struct radeon_device *rdev, u32 reset_mask)
2291 {
2292         struct evergreen_mc_save save;
2293         u32 grbm_soft_reset = 0, srbm_soft_reset = 0;
2294         u32 tmp;
2295
2296         if (reset_mask == 0)
2297                 return;
2298
2299         dev_info(rdev->dev, "GPU softreset: 0x%08X\n", reset_mask);
2300
2301         evergreen_print_gpu_status_regs(rdev);
2302         dev_info(rdev->dev, "  VM_CONTEXT1_PROTECTION_FAULT_ADDR   0x%08X\n",
2303                  RREG32(VM_CONTEXT1_PROTECTION_FAULT_ADDR));
2304         dev_info(rdev->dev, "  VM_CONTEXT1_PROTECTION_FAULT_STATUS 0x%08X\n",
2305                  RREG32(VM_CONTEXT1_PROTECTION_FAULT_STATUS));
2306
2307         /* Disable CP parsing/prefetching */
2308         WREG32(CP_ME_CNTL, CP_ME_HALT | CP_PFP_HALT | CP_CE_HALT);
2309
2310         if (reset_mask & RADEON_RESET_DMA) {
2311                 /* dma0 */
2312                 tmp = RREG32(DMA_RB_CNTL + DMA0_REGISTER_OFFSET);
2313                 tmp &= ~DMA_RB_ENABLE;
2314                 WREG32(DMA_RB_CNTL + DMA0_REGISTER_OFFSET, tmp);
2315         }
2316         if (reset_mask & RADEON_RESET_DMA1) {
2317                 /* dma1 */
2318                 tmp = RREG32(DMA_RB_CNTL + DMA1_REGISTER_OFFSET);
2319                 tmp &= ~DMA_RB_ENABLE;
2320                 WREG32(DMA_RB_CNTL + DMA1_REGISTER_OFFSET, tmp);
2321         }
2322
2323         udelay(50);
2324
2325         evergreen_mc_stop(rdev, &save);
2326         if (evergreen_mc_wait_for_idle(rdev)) {
2327                 dev_warn(rdev->dev, "Wait for MC idle timedout !\n");
2328         }
2329
2330         if (reset_mask & (RADEON_RESET_GFX | RADEON_RESET_COMPUTE | RADEON_RESET_CP)) {
2331                 grbm_soft_reset = SOFT_RESET_CB |
2332                         SOFT_RESET_DB |
2333                         SOFT_RESET_GDS |
2334                         SOFT_RESET_PA |
2335                         SOFT_RESET_SC |
2336                         SOFT_RESET_BCI |
2337                         SOFT_RESET_SPI |
2338                         SOFT_RESET_SX |
2339                         SOFT_RESET_TC |
2340                         SOFT_RESET_TA |
2341                         SOFT_RESET_VGT |
2342                         SOFT_RESET_IA;
2343         }
2344
2345         if (reset_mask & RADEON_RESET_CP) {
2346                 grbm_soft_reset |= SOFT_RESET_CP | SOFT_RESET_VGT;
2347
2348                 srbm_soft_reset |= SOFT_RESET_GRBM;
2349         }
2350
2351         if (reset_mask & RADEON_RESET_DMA)
2352                 srbm_soft_reset |= SOFT_RESET_DMA;
2353
2354         if (reset_mask & RADEON_RESET_DMA1)
2355                 srbm_soft_reset |= SOFT_RESET_DMA1;
2356
2357         if (reset_mask & RADEON_RESET_DISPLAY)
2358                 srbm_soft_reset |= SOFT_RESET_DC;
2359
2360         if (reset_mask & RADEON_RESET_RLC)
2361                 grbm_soft_reset |= SOFT_RESET_RLC;
2362
2363         if (reset_mask & RADEON_RESET_SEM)
2364                 srbm_soft_reset |= SOFT_RESET_SEM;
2365
2366         if (reset_mask & RADEON_RESET_IH)
2367                 srbm_soft_reset |= SOFT_RESET_IH;
2368
2369         if (reset_mask & RADEON_RESET_GRBM)
2370                 srbm_soft_reset |= SOFT_RESET_GRBM;
2371
2372         if (reset_mask & RADEON_RESET_VMC)
2373                 srbm_soft_reset |= SOFT_RESET_VMC;
2374
2375         if (reset_mask & RADEON_RESET_MC)
2376                 srbm_soft_reset |= SOFT_RESET_MC;
2377
2378         if (grbm_soft_reset) {
2379                 tmp = RREG32(GRBM_SOFT_RESET);
2380                 tmp |= grbm_soft_reset;
2381                 dev_info(rdev->dev, "GRBM_SOFT_RESET=0x%08X\n", tmp);
2382                 WREG32(GRBM_SOFT_RESET, tmp);
2383                 tmp = RREG32(GRBM_SOFT_RESET);
2384
2385                 udelay(50);
2386
2387                 tmp &= ~grbm_soft_reset;
2388                 WREG32(GRBM_SOFT_RESET, tmp);
2389                 tmp = RREG32(GRBM_SOFT_RESET);
2390         }
2391
2392         if (srbm_soft_reset) {
2393                 tmp = RREG32(SRBM_SOFT_RESET);
2394                 tmp |= srbm_soft_reset;
2395                 dev_info(rdev->dev, "SRBM_SOFT_RESET=0x%08X\n", tmp);
2396                 WREG32(SRBM_SOFT_RESET, tmp);
2397                 tmp = RREG32(SRBM_SOFT_RESET);
2398
2399                 udelay(50);
2400
2401                 tmp &= ~srbm_soft_reset;
2402                 WREG32(SRBM_SOFT_RESET, tmp);
2403                 tmp = RREG32(SRBM_SOFT_RESET);
2404         }
2405
2406         /* Wait a little for things to settle down */
2407         udelay(50);
2408
2409         evergreen_mc_resume(rdev, &save);
2410         udelay(50);
2411
2412         evergreen_print_gpu_status_regs(rdev);
2413 }
2414
2415 int si_asic_reset(struct radeon_device *rdev)
2416 {
2417         u32 reset_mask;
2418
2419         reset_mask = si_gpu_check_soft_reset(rdev);
2420
2421         if (reset_mask)
2422                 r600_set_bios_scratch_engine_hung(rdev, true);
2423
2424         si_gpu_soft_reset(rdev, reset_mask);
2425
2426         reset_mask = si_gpu_check_soft_reset(rdev);
2427
2428         if (!reset_mask)
2429                 r600_set_bios_scratch_engine_hung(rdev, false);
2430
2431         return 0;
2432 }
2433
2434 /**
2435  * si_gfx_is_lockup - Check if the GFX engine is locked up
2436  *
2437  * @rdev: radeon_device pointer
2438  * @ring: radeon_ring structure holding ring information
2439  *
2440  * Check if the GFX engine is locked up.
2441  * Returns true if the engine appears to be locked up, false if not.
2442  */
2443 bool si_gfx_is_lockup(struct radeon_device *rdev, struct radeon_ring *ring)
2444 {
2445         u32 reset_mask = si_gpu_check_soft_reset(rdev);
2446
2447         if (!(reset_mask & (RADEON_RESET_GFX |
2448                             RADEON_RESET_COMPUTE |
2449                             RADEON_RESET_CP))) {
2450                 radeon_ring_lockup_update(ring);
2451                 return false;
2452         }
2453         /* force CP activities */
2454         radeon_ring_force_activity(rdev, ring);
2455         return radeon_ring_test_lockup(rdev, ring);
2456 }
2457
2458 /**
2459  * si_dma_is_lockup - Check if the DMA engine is locked up
2460  *
2461  * @rdev: radeon_device pointer
2462  * @ring: radeon_ring structure holding ring information
2463  *
2464  * Check if the async DMA engine is locked up.
2465  * Returns true if the engine appears to be locked up, false if not.
2466  */
2467 bool si_dma_is_lockup(struct radeon_device *rdev, struct radeon_ring *ring)
2468 {
2469         u32 reset_mask = si_gpu_check_soft_reset(rdev);
2470         u32 mask;
2471
2472         if (ring->idx == R600_RING_TYPE_DMA_INDEX)
2473                 mask = RADEON_RESET_DMA;
2474         else
2475                 mask = RADEON_RESET_DMA1;
2476
2477         if (!(reset_mask & mask)) {
2478                 radeon_ring_lockup_update(ring);
2479                 return false;
2480         }
2481         /* force ring activities */
2482         radeon_ring_force_activity(rdev, ring);
2483         return radeon_ring_test_lockup(rdev, ring);
2484 }
2485
2486 /* MC */
2487 static void si_mc_program(struct radeon_device *rdev)
2488 {
2489         struct evergreen_mc_save save;
2490         u32 tmp;
2491         int i, j;
2492
2493         /* Initialize HDP */
2494         for (i = 0, j = 0; i < 32; i++, j += 0x18) {
2495                 WREG32((0x2c14 + j), 0x00000000);
2496                 WREG32((0x2c18 + j), 0x00000000);
2497                 WREG32((0x2c1c + j), 0x00000000);
2498                 WREG32((0x2c20 + j), 0x00000000);
2499                 WREG32((0x2c24 + j), 0x00000000);
2500         }
2501         WREG32(HDP_REG_COHERENCY_FLUSH_CNTL, 0);
2502
2503         evergreen_mc_stop(rdev, &save);
2504         if (radeon_mc_wait_for_idle(rdev)) {
2505                 dev_warn(rdev->dev, "Wait for MC idle timedout !\n");
2506         }
2507         /* Lockout access through VGA aperture*/
2508         WREG32(VGA_HDP_CONTROL, VGA_MEMORY_DISABLE);
2509         /* Update configuration */
2510         WREG32(MC_VM_SYSTEM_APERTURE_LOW_ADDR,
2511                rdev->mc.vram_start >> 12);
2512         WREG32(MC_VM_SYSTEM_APERTURE_HIGH_ADDR,
2513                rdev->mc.vram_end >> 12);
2514         WREG32(MC_VM_SYSTEM_APERTURE_DEFAULT_ADDR,
2515                rdev->vram_scratch.gpu_addr >> 12);
2516         tmp = ((rdev->mc.vram_end >> 24) & 0xFFFF) << 16;
2517         tmp |= ((rdev->mc.vram_start >> 24) & 0xFFFF);
2518         WREG32(MC_VM_FB_LOCATION, tmp);
2519         /* XXX double check these! */
2520         WREG32(HDP_NONSURFACE_BASE, (rdev->mc.vram_start >> 8));
2521         WREG32(HDP_NONSURFACE_INFO, (2 << 7) | (1 << 30));
2522         WREG32(HDP_NONSURFACE_SIZE, 0x3FFFFFFF);
2523         WREG32(MC_VM_AGP_BASE, 0);
2524         WREG32(MC_VM_AGP_TOP, 0x0FFFFFFF);
2525         WREG32(MC_VM_AGP_BOT, 0x0FFFFFFF);
2526         if (radeon_mc_wait_for_idle(rdev)) {
2527                 dev_warn(rdev->dev, "Wait for MC idle timedout !\n");
2528         }
2529         evergreen_mc_resume(rdev, &save);
2530         /* we need to own VRAM, so turn off the VGA renderer here
2531          * to stop it overwriting our objects */
2532         rv515_vga_render_disable(rdev);
2533 }
2534
2535 /* SI MC address space is 40 bits */
2536 static void si_vram_location(struct radeon_device *rdev,
2537                              struct radeon_mc *mc, u64 base)
2538 {
2539         mc->vram_start = base;
2540         if (mc->mc_vram_size > (0xFFFFFFFFFFULL - base + 1)) {
2541                 dev_warn(rdev->dev, "limiting VRAM to PCI aperture size\n");
2542                 mc->real_vram_size = mc->aper_size;
2543                 mc->mc_vram_size = mc->aper_size;
2544         }
2545         mc->vram_end = mc->vram_start + mc->mc_vram_size - 1;
2546         dev_info(rdev->dev, "VRAM: %lluM 0x%016llX - 0x%016llX (%lluM used)\n",
2547                         mc->mc_vram_size >> 20, mc->vram_start,
2548                         mc->vram_end, mc->real_vram_size >> 20);
2549 }
2550
2551 static void si_gtt_location(struct radeon_device *rdev, struct radeon_mc *mc)
2552 {
2553         u64 size_af, size_bf;
2554
2555         size_af = ((0xFFFFFFFFFFULL - mc->vram_end) + mc->gtt_base_align) & ~mc->gtt_base_align;
2556         size_bf = mc->vram_start & ~mc->gtt_base_align;
2557         if (size_bf > size_af) {
2558                 if (mc->gtt_size > size_bf) {
2559                         dev_warn(rdev->dev, "limiting GTT\n");
2560                         mc->gtt_size = size_bf;
2561                 }
2562                 mc->gtt_start = (mc->vram_start & ~mc->gtt_base_align) - mc->gtt_size;
2563         } else {
2564                 if (mc->gtt_size > size_af) {
2565                         dev_warn(rdev->dev, "limiting GTT\n");
2566                         mc->gtt_size = size_af;
2567                 }
2568                 mc->gtt_start = (mc->vram_end + 1 + mc->gtt_base_align) & ~mc->gtt_base_align;
2569         }
2570         mc->gtt_end = mc->gtt_start + mc->gtt_size - 1;
2571         dev_info(rdev->dev, "GTT: %lluM 0x%016llX - 0x%016llX\n",
2572                         mc->gtt_size >> 20, mc->gtt_start, mc->gtt_end);
2573 }
2574
2575 static void si_vram_gtt_location(struct radeon_device *rdev,
2576                                  struct radeon_mc *mc)
2577 {
2578         if (mc->mc_vram_size > 0xFFC0000000ULL) {
2579                 /* leave room for at least 1024M GTT */
2580                 dev_warn(rdev->dev, "limiting VRAM\n");
2581                 mc->real_vram_size = 0xFFC0000000ULL;
2582                 mc->mc_vram_size = 0xFFC0000000ULL;
2583         }
2584         si_vram_location(rdev, &rdev->mc, 0);
2585         rdev->mc.gtt_base_align = 0;
2586         si_gtt_location(rdev, mc);
2587 }
2588
2589 static int si_mc_init(struct radeon_device *rdev)
2590 {
2591         u32 tmp;
2592         int chansize, numchan;
2593
2594         /* Get VRAM informations */
2595         rdev->mc.vram_is_ddr = true;
2596         tmp = RREG32(MC_ARB_RAMCFG);
2597         if (tmp & CHANSIZE_OVERRIDE) {
2598                 chansize = 16;
2599         } else if (tmp & CHANSIZE_MASK) {
2600                 chansize = 64;
2601         } else {
2602                 chansize = 32;
2603         }
2604         tmp = RREG32(MC_SHARED_CHMAP);
2605         switch ((tmp & NOOFCHAN_MASK) >> NOOFCHAN_SHIFT) {
2606         case 0:
2607         default:
2608                 numchan = 1;
2609                 break;
2610         case 1:
2611                 numchan = 2;
2612                 break;
2613         case 2:
2614                 numchan = 4;
2615                 break;
2616         case 3:
2617                 numchan = 8;
2618                 break;
2619         case 4:
2620                 numchan = 3;
2621                 break;
2622         case 5:
2623                 numchan = 6;
2624                 break;
2625         case 6:
2626                 numchan = 10;
2627                 break;
2628         case 7:
2629                 numchan = 12;
2630                 break;
2631         case 8:
2632                 numchan = 16;
2633                 break;
2634         }
2635         rdev->mc.vram_width = numchan * chansize;
2636         /* Could aper size report 0 ? */
2637         rdev->mc.aper_base = pci_resource_start(rdev->pdev, 0);
2638         rdev->mc.aper_size = pci_resource_len(rdev->pdev, 0);
2639         /* size in MB on si */
2640         rdev->mc.mc_vram_size = RREG32(CONFIG_MEMSIZE) * 1024 * 1024;
2641         rdev->mc.real_vram_size = RREG32(CONFIG_MEMSIZE) * 1024 * 1024;
2642         rdev->mc.visible_vram_size = rdev->mc.aper_size;
2643         si_vram_gtt_location(rdev, &rdev->mc);
2644         radeon_update_bandwidth_info(rdev);
2645
2646         return 0;
2647 }
2648
2649 /*
2650  * GART
2651  */
2652 void si_pcie_gart_tlb_flush(struct radeon_device *rdev)
2653 {
2654         /* flush hdp cache */
2655         WREG32(HDP_MEM_COHERENCY_FLUSH_CNTL, 0x1);
2656
2657         /* bits 0-15 are the VM contexts0-15 */
2658         WREG32(VM_INVALIDATE_REQUEST, 1);
2659 }
2660
2661 static int si_pcie_gart_enable(struct radeon_device *rdev)
2662 {
2663         int r, i;
2664
2665         if (rdev->gart.robj == NULL) {
2666                 dev_err(rdev->dev, "No VRAM object for PCIE GART.\n");
2667                 return -EINVAL;
2668         }
2669         r = radeon_gart_table_vram_pin(rdev);
2670         if (r)
2671                 return r;
2672         radeon_gart_restore(rdev);
2673         /* Setup TLB control */
2674         WREG32(MC_VM_MX_L1_TLB_CNTL,
2675                (0xA << 7) |
2676                ENABLE_L1_TLB |
2677                SYSTEM_ACCESS_MODE_NOT_IN_SYS |
2678                ENABLE_ADVANCED_DRIVER_MODEL |
2679                SYSTEM_APERTURE_UNMAPPED_ACCESS_PASS_THRU);
2680         /* Setup L2 cache */
2681         WREG32(VM_L2_CNTL, ENABLE_L2_CACHE |
2682                ENABLE_L2_PTE_CACHE_LRU_UPDATE_BY_WRITE |
2683                ENABLE_L2_PDE0_CACHE_LRU_UPDATE_BY_WRITE |
2684                EFFECTIVE_L2_QUEUE_SIZE(7) |
2685                CONTEXT1_IDENTITY_ACCESS_MODE(1));
2686         WREG32(VM_L2_CNTL2, INVALIDATE_ALL_L1_TLBS | INVALIDATE_L2_CACHE);
2687         WREG32(VM_L2_CNTL3, L2_CACHE_BIGK_ASSOCIATIVITY |
2688                L2_CACHE_BIGK_FRAGMENT_SIZE(0));
2689         /* setup context0 */
2690         WREG32(VM_CONTEXT0_PAGE_TABLE_START_ADDR, rdev->mc.gtt_start >> 12);
2691         WREG32(VM_CONTEXT0_PAGE_TABLE_END_ADDR, rdev->mc.gtt_end >> 12);
2692         WREG32(VM_CONTEXT0_PAGE_TABLE_BASE_ADDR, rdev->gart.table_addr >> 12);
2693         WREG32(VM_CONTEXT0_PROTECTION_FAULT_DEFAULT_ADDR,
2694                         (u32)(rdev->dummy_page.addr >> 12));
2695         WREG32(VM_CONTEXT0_CNTL2, 0);
2696         WREG32(VM_CONTEXT0_CNTL, (ENABLE_CONTEXT | PAGE_TABLE_DEPTH(0) |
2697                                   RANGE_PROTECTION_FAULT_ENABLE_DEFAULT));
2698
2699         WREG32(0x15D4, 0);
2700         WREG32(0x15D8, 0);
2701         WREG32(0x15DC, 0);
2702
2703         /* empty context1-15 */
2704         /* set vm size, must be a multiple of 4 */
2705         WREG32(VM_CONTEXT1_PAGE_TABLE_START_ADDR, 0);
2706         WREG32(VM_CONTEXT1_PAGE_TABLE_END_ADDR, rdev->vm_manager.max_pfn);
2707         /* Assign the pt base to something valid for now; the pts used for
2708          * the VMs are determined by the application and setup and assigned
2709          * on the fly in the vm part of radeon_gart.c
2710          */
2711         for (i = 1; i < 16; i++) {
2712                 if (i < 8)
2713                         WREG32(VM_CONTEXT0_PAGE_TABLE_BASE_ADDR + (i << 2),
2714                                rdev->gart.table_addr >> 12);
2715                 else
2716                         WREG32(VM_CONTEXT8_PAGE_TABLE_BASE_ADDR + ((i - 8) << 2),
2717                                rdev->gart.table_addr >> 12);
2718         }
2719
2720         /* enable context1-15 */
2721         WREG32(VM_CONTEXT1_PROTECTION_FAULT_DEFAULT_ADDR,
2722                (u32)(rdev->dummy_page.addr >> 12));
2723         WREG32(VM_CONTEXT1_CNTL2, 4);
2724         WREG32(VM_CONTEXT1_CNTL, ENABLE_CONTEXT | PAGE_TABLE_DEPTH(1) |
2725                                 RANGE_PROTECTION_FAULT_ENABLE_INTERRUPT |
2726                                 RANGE_PROTECTION_FAULT_ENABLE_DEFAULT |
2727                                 DUMMY_PAGE_PROTECTION_FAULT_ENABLE_INTERRUPT |
2728                                 DUMMY_PAGE_PROTECTION_FAULT_ENABLE_DEFAULT |
2729                                 PDE0_PROTECTION_FAULT_ENABLE_INTERRUPT |
2730                                 PDE0_PROTECTION_FAULT_ENABLE_DEFAULT |
2731                                 VALID_PROTECTION_FAULT_ENABLE_INTERRUPT |
2732                                 VALID_PROTECTION_FAULT_ENABLE_DEFAULT |
2733                                 READ_PROTECTION_FAULT_ENABLE_INTERRUPT |
2734                                 READ_PROTECTION_FAULT_ENABLE_DEFAULT |
2735                                 WRITE_PROTECTION_FAULT_ENABLE_INTERRUPT |
2736                                 WRITE_PROTECTION_FAULT_ENABLE_DEFAULT);
2737
2738         si_pcie_gart_tlb_flush(rdev);
2739         DRM_INFO("PCIE GART of %uM enabled (table at 0x%016llX).\n",
2740                  (unsigned)(rdev->mc.gtt_size >> 20),
2741                  (unsigned long long)rdev->gart.table_addr);
2742         rdev->gart.ready = true;
2743         return 0;
2744 }
2745
2746 static void si_pcie_gart_disable(struct radeon_device *rdev)
2747 {
2748         /* Disable all tables */
2749         WREG32(VM_CONTEXT0_CNTL, 0);
2750         WREG32(VM_CONTEXT1_CNTL, 0);
2751         /* Setup TLB control */
2752         WREG32(MC_VM_MX_L1_TLB_CNTL, SYSTEM_ACCESS_MODE_NOT_IN_SYS |
2753                SYSTEM_APERTURE_UNMAPPED_ACCESS_PASS_THRU);
2754         /* Setup L2 cache */
2755         WREG32(VM_L2_CNTL, ENABLE_L2_PTE_CACHE_LRU_UPDATE_BY_WRITE |
2756                ENABLE_L2_PDE0_CACHE_LRU_UPDATE_BY_WRITE |
2757                EFFECTIVE_L2_QUEUE_SIZE(7) |
2758                CONTEXT1_IDENTITY_ACCESS_MODE(1));
2759         WREG32(VM_L2_CNTL2, 0);
2760         WREG32(VM_L2_CNTL3, L2_CACHE_BIGK_ASSOCIATIVITY |
2761                L2_CACHE_BIGK_FRAGMENT_SIZE(0));
2762         radeon_gart_table_vram_unpin(rdev);
2763 }
2764
2765 static void si_pcie_gart_fini(struct radeon_device *rdev)
2766 {
2767         si_pcie_gart_disable(rdev);
2768         radeon_gart_table_vram_free(rdev);
2769         radeon_gart_fini(rdev);
2770 }
2771
2772 /* vm parser */
2773 static bool si_vm_reg_valid(u32 reg)
2774 {
2775         /* context regs are fine */
2776         if (reg >= 0x28000)
2777                 return true;
2778
2779         /* check config regs */
2780         switch (reg) {
2781         case GRBM_GFX_INDEX:
2782         case CP_STRMOUT_CNTL:
2783         case VGT_VTX_VECT_EJECT_REG:
2784         case VGT_CACHE_INVALIDATION:
2785         case VGT_ESGS_RING_SIZE:
2786         case VGT_GSVS_RING_SIZE:
2787         case VGT_GS_VERTEX_REUSE:
2788         case VGT_PRIMITIVE_TYPE:
2789         case VGT_INDEX_TYPE:
2790         case VGT_NUM_INDICES:
2791         case VGT_NUM_INSTANCES:
2792         case VGT_TF_RING_SIZE:
2793         case VGT_HS_OFFCHIP_PARAM:
2794         case VGT_TF_MEMORY_BASE:
2795         case PA_CL_ENHANCE:
2796         case PA_SU_LINE_STIPPLE_VALUE:
2797         case PA_SC_LINE_STIPPLE_STATE:
2798         case PA_SC_ENHANCE:
2799         case SQC_CACHES:
2800         case SPI_STATIC_THREAD_MGMT_1:
2801         case SPI_STATIC_THREAD_MGMT_2:
2802         case SPI_STATIC_THREAD_MGMT_3:
2803         case SPI_PS_MAX_WAVE_ID:
2804         case SPI_CONFIG_CNTL:
2805         case SPI_CONFIG_CNTL_1:
2806         case TA_CNTL_AUX:
2807                 return true;
2808         default:
2809                 DRM_ERROR("Invalid register 0x%x in CS\n", reg);
2810                 return false;
2811         }
2812 }
2813
2814 static int si_vm_packet3_ce_check(struct radeon_device *rdev,
2815                                   u32 *ib, struct radeon_cs_packet *pkt)
2816 {
2817         switch (pkt->opcode) {
2818         case PACKET3_NOP:
2819         case PACKET3_SET_BASE:
2820         case PACKET3_SET_CE_DE_COUNTERS:
2821         case PACKET3_LOAD_CONST_RAM:
2822         case PACKET3_WRITE_CONST_RAM:
2823         case PACKET3_WRITE_CONST_RAM_OFFSET:
2824         case PACKET3_DUMP_CONST_RAM:
2825         case PACKET3_INCREMENT_CE_COUNTER:
2826         case PACKET3_WAIT_ON_DE_COUNTER:
2827         case PACKET3_CE_WRITE:
2828                 break;
2829         default:
2830                 DRM_ERROR("Invalid CE packet3: 0x%x\n", pkt->opcode);
2831                 return -EINVAL;
2832         }
2833         return 0;
2834 }
2835
2836 static int si_vm_packet3_gfx_check(struct radeon_device *rdev,
2837                                    u32 *ib, struct radeon_cs_packet *pkt)
2838 {
2839         u32 idx = pkt->idx + 1;
2840         u32 idx_value = ib[idx];
2841         u32 start_reg, end_reg, reg, i;
2842         u32 command, info;
2843
2844         switch (pkt->opcode) {
2845         case PACKET3_NOP:
2846         case PACKET3_SET_BASE:
2847         case PACKET3_CLEAR_STATE:
2848         case PACKET3_INDEX_BUFFER_SIZE:
2849         case PACKET3_DISPATCH_DIRECT:
2850         case PACKET3_DISPATCH_INDIRECT:
2851         case PACKET3_ALLOC_GDS:
2852         case PACKET3_WRITE_GDS_RAM:
2853         case PACKET3_ATOMIC_GDS:
2854         case PACKET3_ATOMIC:
2855         case PACKET3_OCCLUSION_QUERY:
2856         case PACKET3_SET_PREDICATION:
2857         case PACKET3_COND_EXEC:
2858         case PACKET3_PRED_EXEC:
2859         case PACKET3_DRAW_INDIRECT:
2860         case PACKET3_DRAW_INDEX_INDIRECT:
2861         case PACKET3_INDEX_BASE:
2862         case PACKET3_DRAW_INDEX_2:
2863         case PACKET3_CONTEXT_CONTROL:
2864         case PACKET3_INDEX_TYPE:
2865         case PACKET3_DRAW_INDIRECT_MULTI:
2866         case PACKET3_DRAW_INDEX_AUTO:
2867         case PACKET3_DRAW_INDEX_IMMD:
2868         case PACKET3_NUM_INSTANCES:
2869         case PACKET3_DRAW_INDEX_MULTI_AUTO:
2870         case PACKET3_STRMOUT_BUFFER_UPDATE:
2871         case PACKET3_DRAW_INDEX_OFFSET_2:
2872         case PACKET3_DRAW_INDEX_MULTI_ELEMENT:
2873         case PACKET3_DRAW_INDEX_INDIRECT_MULTI:
2874         case PACKET3_MPEG_INDEX:
2875         case PACKET3_WAIT_REG_MEM:
2876         case PACKET3_MEM_WRITE:
2877         case PACKET3_PFP_SYNC_ME:
2878         case PACKET3_SURFACE_SYNC:
2879         case PACKET3_EVENT_WRITE:
2880         case PACKET3_EVENT_WRITE_EOP:
2881         case PACKET3_EVENT_WRITE_EOS:
2882         case PACKET3_SET_CONTEXT_REG:
2883         case PACKET3_SET_CONTEXT_REG_INDIRECT:
2884         case PACKET3_SET_SH_REG:
2885         case PACKET3_SET_SH_REG_OFFSET:
2886         case PACKET3_INCREMENT_DE_COUNTER:
2887         case PACKET3_WAIT_ON_CE_COUNTER:
2888         case PACKET3_WAIT_ON_AVAIL_BUFFER:
2889         case PACKET3_ME_WRITE:
2890                 break;
2891         case PACKET3_COPY_DATA:
2892                 if ((idx_value & 0xf00) == 0) {
2893                         reg = ib[idx + 3] * 4;
2894                         if (!si_vm_reg_valid(reg))
2895                                 return -EINVAL;
2896                 }
2897                 break;
2898         case PACKET3_WRITE_DATA:
2899                 if ((idx_value & 0xf00) == 0) {
2900                         start_reg = ib[idx + 1] * 4;
2901                         if (idx_value & 0x10000) {
2902                                 if (!si_vm_reg_valid(start_reg))
2903                                         return -EINVAL;
2904                         } else {
2905                                 for (i = 0; i < (pkt->count - 2); i++) {
2906                                         reg = start_reg + (4 * i);
2907                                         if (!si_vm_reg_valid(reg))
2908                                                 return -EINVAL;
2909                                 }
2910                         }
2911                 }
2912                 break;
2913         case PACKET3_COND_WRITE:
2914                 if (idx_value & 0x100) {
2915                         reg = ib[idx + 5] * 4;
2916                         if (!si_vm_reg_valid(reg))
2917                                 return -EINVAL;
2918                 }
2919                 break;
2920         case PACKET3_COPY_DW:
2921                 if (idx_value & 0x2) {
2922                         reg = ib[idx + 3] * 4;
2923                         if (!si_vm_reg_valid(reg))
2924                                 return -EINVAL;
2925                 }
2926                 break;
2927         case PACKET3_SET_CONFIG_REG:
2928                 start_reg = (idx_value << 2) + PACKET3_SET_CONFIG_REG_START;
2929                 end_reg = 4 * pkt->count + start_reg - 4;
2930                 if ((start_reg < PACKET3_SET_CONFIG_REG_START) ||
2931                     (start_reg >= PACKET3_SET_CONFIG_REG_END) ||
2932                     (end_reg >= PACKET3_SET_CONFIG_REG_END)) {
2933                         DRM_ERROR("bad PACKET3_SET_CONFIG_REG\n");
2934                         return -EINVAL;
2935                 }
2936                 for (i = 0; i < pkt->count; i++) {
2937                         reg = start_reg + (4 * i);
2938                         if (!si_vm_reg_valid(reg))
2939                                 return -EINVAL;
2940                 }
2941                 break;
2942         case PACKET3_CP_DMA:
2943                 command = ib[idx + 4];
2944                 info = ib[idx + 1];
2945                 if (command & PACKET3_CP_DMA_CMD_SAS) {
2946                         /* src address space is register */
2947                         if (((info & 0x60000000) >> 29) == 0) {
2948                                 start_reg = idx_value << 2;
2949                                 if (command & PACKET3_CP_DMA_CMD_SAIC) {
2950                                         reg = start_reg;
2951                                         if (!si_vm_reg_valid(reg)) {
2952                                                 DRM_ERROR("CP DMA Bad SRC register\n");
2953                                                 return -EINVAL;
2954                                         }
2955                                 } else {
2956                                         for (i = 0; i < (command & 0x1fffff); i++) {
2957                                                 reg = start_reg + (4 * i);
2958                                                 if (!si_vm_reg_valid(reg)) {
2959                                                         DRM_ERROR("CP DMA Bad SRC register\n");
2960                                                         return -EINVAL;
2961                                                 }
2962                                         }
2963                                 }
2964                         }
2965                 }
2966                 if (command & PACKET3_CP_DMA_CMD_DAS) {
2967                         /* dst address space is register */
2968                         if (((info & 0x00300000) >> 20) == 0) {
2969                                 start_reg = ib[idx + 2];
2970                                 if (command & PACKET3_CP_DMA_CMD_DAIC) {
2971                                         reg = start_reg;
2972                                         if (!si_vm_reg_valid(reg)) {
2973                                                 DRM_ERROR("CP DMA Bad DST register\n");
2974                                                 return -EINVAL;
2975                                         }
2976                                 } else {
2977                                         for (i = 0; i < (command & 0x1fffff); i++) {
2978                                                 reg = start_reg + (4 * i);
2979                                                 if (!si_vm_reg_valid(reg)) {
2980                                                         DRM_ERROR("CP DMA Bad DST register\n");
2981                                                         return -EINVAL;
2982                                                 }
2983                                         }
2984                                 }
2985                         }
2986                 }
2987                 break;
2988         default:
2989                 DRM_ERROR("Invalid GFX packet3: 0x%x\n", pkt->opcode);
2990                 return -EINVAL;
2991         }
2992         return 0;
2993 }
2994
2995 static int si_vm_packet3_compute_check(struct radeon_device *rdev,
2996                                        u32 *ib, struct radeon_cs_packet *pkt)
2997 {
2998         u32 idx = pkt->idx + 1;
2999         u32 idx_value = ib[idx];
3000         u32 start_reg, reg, i;
3001
3002         switch (pkt->opcode) {
3003         case PACKET3_NOP:
3004         case PACKET3_SET_BASE:
3005         case PACKET3_CLEAR_STATE:
3006         case PACKET3_DISPATCH_DIRECT:
3007         case PACKET3_DISPATCH_INDIRECT:
3008         case PACKET3_ALLOC_GDS:
3009         case PACKET3_WRITE_GDS_RAM:
3010         case PACKET3_ATOMIC_GDS:
3011         case PACKET3_ATOMIC:
3012         case PACKET3_OCCLUSION_QUERY:
3013         case PACKET3_SET_PREDICATION:
3014         case PACKET3_COND_EXEC:
3015         case PACKET3_PRED_EXEC:
3016         case PACKET3_CONTEXT_CONTROL:
3017         case PACKET3_STRMOUT_BUFFER_UPDATE:
3018         case PACKET3_WAIT_REG_MEM:
3019         case PACKET3_MEM_WRITE:
3020         case PACKET3_PFP_SYNC_ME:
3021         case PACKET3_SURFACE_SYNC:
3022         case PACKET3_EVENT_WRITE:
3023         case PACKET3_EVENT_WRITE_EOP:
3024         case PACKET3_EVENT_WRITE_EOS:
3025         case PACKET3_SET_CONTEXT_REG:
3026         case PACKET3_SET_CONTEXT_REG_INDIRECT:
3027         case PACKET3_SET_SH_REG:
3028         case PACKET3_SET_SH_REG_OFFSET:
3029         case PACKET3_INCREMENT_DE_COUNTER:
3030         case PACKET3_WAIT_ON_CE_COUNTER:
3031         case PACKET3_WAIT_ON_AVAIL_BUFFER:
3032         case PACKET3_ME_WRITE:
3033                 break;
3034         case PACKET3_COPY_DATA:
3035                 if ((idx_value & 0xf00) == 0) {
3036                         reg = ib[idx + 3] * 4;
3037                         if (!si_vm_reg_valid(reg))
3038                                 return -EINVAL;
3039                 }
3040                 break;
3041         case PACKET3_WRITE_DATA:
3042                 if ((idx_value & 0xf00) == 0) {
3043                         start_reg = ib[idx + 1] * 4;
3044                         if (idx_value & 0x10000) {
3045                                 if (!si_vm_reg_valid(start_reg))
3046                                         return -EINVAL;
3047                         } else {
3048                                 for (i = 0; i < (pkt->count - 2); i++) {
3049                                         reg = start_reg + (4 * i);
3050                                         if (!si_vm_reg_valid(reg))
3051                                                 return -EINVAL;
3052                                 }
3053                         }
3054                 }
3055                 break;
3056         case PACKET3_COND_WRITE:
3057                 if (idx_value & 0x100) {
3058                         reg = ib[idx + 5] * 4;
3059                         if (!si_vm_reg_valid(reg))
3060                                 return -EINVAL;
3061                 }
3062                 break;
3063         case PACKET3_COPY_DW:
3064                 if (idx_value & 0x2) {
3065                         reg = ib[idx + 3] * 4;
3066                         if (!si_vm_reg_valid(reg))
3067                                 return -EINVAL;
3068                 }
3069                 break;
3070         default:
3071                 DRM_ERROR("Invalid Compute packet3: 0x%x\n", pkt->opcode);
3072                 return -EINVAL;
3073         }
3074         return 0;
3075 }
3076
3077 int si_ib_parse(struct radeon_device *rdev, struct radeon_ib *ib)
3078 {
3079         int ret = 0;
3080         u32 idx = 0;
3081         struct radeon_cs_packet pkt;
3082
3083         do {
3084                 pkt.idx = idx;
3085                 pkt.type = RADEON_CP_PACKET_GET_TYPE(ib->ptr[idx]);
3086                 pkt.count = RADEON_CP_PACKET_GET_COUNT(ib->ptr[idx]);
3087                 pkt.one_reg_wr = 0;
3088                 switch (pkt.type) {
3089                 case RADEON_PACKET_TYPE0:
3090                         dev_err(rdev->dev, "Packet0 not allowed!\n");
3091                         ret = -EINVAL;
3092                         break;
3093                 case RADEON_PACKET_TYPE2:
3094                         idx += 1;
3095                         break;
3096                 case RADEON_PACKET_TYPE3:
3097                         pkt.opcode = RADEON_CP_PACKET3_GET_OPCODE(ib->ptr[idx]);
3098                         if (ib->is_const_ib)
3099                                 ret = si_vm_packet3_ce_check(rdev, ib->ptr, &pkt);
3100                         else {
3101                                 switch (ib->ring) {
3102                                 case RADEON_RING_TYPE_GFX_INDEX:
3103                                         ret = si_vm_packet3_gfx_check(rdev, ib->ptr, &pkt);
3104                                         break;
3105                                 case CAYMAN_RING_TYPE_CP1_INDEX:
3106                                 case CAYMAN_RING_TYPE_CP2_INDEX:
3107                                         ret = si_vm_packet3_compute_check(rdev, ib->ptr, &pkt);
3108                                         break;
3109                                 default:
3110                                         dev_err(rdev->dev, "Non-PM4 ring %d !\n", ib->ring);
3111                                         ret = -EINVAL;
3112                                         break;
3113                                 }
3114                         }
3115                         idx += pkt.count + 2;
3116                         break;
3117                 default:
3118                         dev_err(rdev->dev, "Unknown packet type %d !\n", pkt.type);
3119                         ret = -EINVAL;
3120                         break;
3121                 }
3122                 if (ret)
3123                         break;
3124         } while (idx < ib->length_dw);
3125
3126         return ret;
3127 }
3128
3129 /*
3130  * vm
3131  */
3132 int si_vm_init(struct radeon_device *rdev)
3133 {
3134         /* number of VMs */
3135         rdev->vm_manager.nvm = 16;
3136         /* base offset of vram pages */
3137         rdev->vm_manager.vram_base_offset = 0;
3138
3139         return 0;
3140 }
3141
3142 void si_vm_fini(struct radeon_device *rdev)
3143 {
3144 }
3145
3146 /**
3147  * si_vm_set_page - update the page tables using the CP
3148  *
3149  * @rdev: radeon_device pointer
3150  * @ib: indirect buffer to fill with commands
3151  * @pe: addr of the page entry
3152  * @addr: dst addr to write into pe
3153  * @count: number of page entries to update
3154  * @incr: increase next addr by incr bytes
3155  * @flags: access flags
3156  *
3157  * Update the page tables using the CP (SI).
3158  */
3159 void si_vm_set_page(struct radeon_device *rdev,
3160                     struct radeon_ib *ib,
3161                     uint64_t pe,
3162                     uint64_t addr, unsigned count,
3163                     uint32_t incr, uint32_t flags)
3164 {
3165         uint32_t r600_flags = cayman_vm_page_flags(rdev, flags);
3166         uint64_t value;
3167         unsigned ndw;
3168
3169         if (rdev->asic->vm.pt_ring_index == RADEON_RING_TYPE_GFX_INDEX) {
3170                 while (count) {
3171                         ndw = 2 + count * 2;
3172                         if (ndw > 0x3FFE)
3173                                 ndw = 0x3FFE;
3174
3175                         ib->ptr[ib->length_dw++] = PACKET3(PACKET3_WRITE_DATA, ndw);
3176                         ib->ptr[ib->length_dw++] = (WRITE_DATA_ENGINE_SEL(0) |
3177                                         WRITE_DATA_DST_SEL(1));
3178                         ib->ptr[ib->length_dw++] = pe;
3179                         ib->ptr[ib->length_dw++] = upper_32_bits(pe);
3180                         for (; ndw > 2; ndw -= 2, --count, pe += 8) {
3181                                 if (flags & RADEON_VM_PAGE_SYSTEM) {
3182                                         value = radeon_vm_map_gart(rdev, addr);
3183                                         value &= 0xFFFFFFFFFFFFF000ULL;
3184                                 } else if (flags & RADEON_VM_PAGE_VALID) {
3185                                         value = addr;
3186                                 } else {
3187                                         value = 0;
3188                                 }
3189                                 addr += incr;
3190                                 value |= r600_flags;
3191                                 ib->ptr[ib->length_dw++] = value;
3192                                 ib->ptr[ib->length_dw++] = upper_32_bits(value);
3193                         }
3194                 }
3195         } else {
3196                 /* DMA */
3197                 if (flags & RADEON_VM_PAGE_SYSTEM) {
3198                         while (count) {
3199                                 ndw = count * 2;
3200                                 if (ndw > 0xFFFFE)
3201                                         ndw = 0xFFFFE;
3202
3203                                 /* for non-physically contiguous pages (system) */
3204                                 ib->ptr[ib->length_dw++] = DMA_PACKET(DMA_PACKET_WRITE, 0, 0, 0, ndw);
3205                                 ib->ptr[ib->length_dw++] = pe;
3206                                 ib->ptr[ib->length_dw++] = upper_32_bits(pe) & 0xff;
3207                                 for (; ndw > 0; ndw -= 2, --count, pe += 8) {
3208                                         if (flags & RADEON_VM_PAGE_SYSTEM) {
3209                                                 value = radeon_vm_map_gart(rdev, addr);
3210                                                 value &= 0xFFFFFFFFFFFFF000ULL;
3211                                         } else if (flags & RADEON_VM_PAGE_VALID) {
3212                                                 value = addr;
3213                                         } else {
3214                                                 value = 0;
3215                                         }
3216                                         addr += incr;
3217                                         value |= r600_flags;
3218                                         ib->ptr[ib->length_dw++] = value;
3219                                         ib->ptr[ib->length_dw++] = upper_32_bits(value);
3220                                 }
3221                         }
3222                 } else {
3223                         while (count) {
3224                                 ndw = count * 2;
3225                                 if (ndw > 0xFFFFE)
3226                                         ndw = 0xFFFFE;
3227
3228                                 if (flags & RADEON_VM_PAGE_VALID)
3229                                         value = addr;
3230                                 else
3231                                         value = 0;
3232                                 /* for physically contiguous pages (vram) */
3233                                 ib->ptr[ib->length_dw++] = DMA_PTE_PDE_PACKET(ndw);
3234                                 ib->ptr[ib->length_dw++] = pe; /* dst addr */
3235                                 ib->ptr[ib->length_dw++] = upper_32_bits(pe) & 0xff;
3236                                 ib->ptr[ib->length_dw++] = r600_flags; /* mask */
3237                                 ib->ptr[ib->length_dw++] = 0;
3238                                 ib->ptr[ib->length_dw++] = value; /* value */
3239                                 ib->ptr[ib->length_dw++] = upper_32_bits(value);
3240                                 ib->ptr[ib->length_dw++] = incr; /* increment size */
3241                                 ib->ptr[ib->length_dw++] = 0;
3242                                 pe += ndw * 4;
3243                                 addr += (ndw / 2) * incr;
3244                                 count -= ndw / 2;
3245                         }
3246                 }
3247                 while (ib->length_dw & 0x7)
3248                         ib->ptr[ib->length_dw++] = DMA_PACKET(DMA_PACKET_NOP, 0, 0, 0, 0);
3249         }
3250 }
3251
3252 void si_vm_flush(struct radeon_device *rdev, int ridx, struct radeon_vm *vm)
3253 {
3254         struct radeon_ring *ring = &rdev->ring[ridx];
3255
3256         if (vm == NULL)
3257                 return;
3258
3259         /* write new base address */
3260         radeon_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
3261         radeon_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
3262                                  WRITE_DATA_DST_SEL(0)));
3263
3264         if (vm->id < 8) {
3265                 radeon_ring_write(ring,
3266                                   (VM_CONTEXT0_PAGE_TABLE_BASE_ADDR + (vm->id << 2)) >> 2);
3267         } else {
3268                 radeon_ring_write(ring,
3269                                   (VM_CONTEXT8_PAGE_TABLE_BASE_ADDR + ((vm->id - 8) << 2)) >> 2);
3270         }
3271         radeon_ring_write(ring, 0);
3272         radeon_ring_write(ring, vm->pd_gpu_addr >> 12);
3273
3274         /* flush hdp cache */
3275         radeon_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
3276         radeon_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
3277                                  WRITE_DATA_DST_SEL(0)));
3278         radeon_ring_write(ring, HDP_MEM_COHERENCY_FLUSH_CNTL >> 2);
3279         radeon_ring_write(ring, 0);
3280         radeon_ring_write(ring, 0x1);
3281
3282         /* bits 0-15 are the VM contexts0-15 */
3283         radeon_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
3284         radeon_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
3285                                  WRITE_DATA_DST_SEL(0)));
3286         radeon_ring_write(ring, VM_INVALIDATE_REQUEST >> 2);
3287         radeon_ring_write(ring, 0);
3288         radeon_ring_write(ring, 1 << vm->id);
3289
3290         /* sync PFP to ME, otherwise we might get invalid PFP reads */
3291         radeon_ring_write(ring, PACKET3(PACKET3_PFP_SYNC_ME, 0));
3292         radeon_ring_write(ring, 0x0);
3293 }
3294
3295 void si_dma_vm_flush(struct radeon_device *rdev, int ridx, struct radeon_vm *vm)
3296 {
3297         struct radeon_ring *ring = &rdev->ring[ridx];
3298
3299         if (vm == NULL)
3300                 return;
3301
3302         radeon_ring_write(ring, DMA_PACKET(DMA_PACKET_SRBM_WRITE, 0, 0, 0, 0));
3303         if (vm->id < 8) {
3304                 radeon_ring_write(ring, (0xf << 16) | ((VM_CONTEXT0_PAGE_TABLE_BASE_ADDR + (vm->id << 2)) >> 2));
3305         } else {
3306                 radeon_ring_write(ring, (0xf << 16) | ((VM_CONTEXT8_PAGE_TABLE_BASE_ADDR + ((vm->id - 8) << 2)) >> 2));
3307         }
3308         radeon_ring_write(ring, vm->pd_gpu_addr >> 12);
3309
3310         /* flush hdp cache */
3311         radeon_ring_write(ring, DMA_PACKET(DMA_PACKET_SRBM_WRITE, 0, 0, 0, 0));
3312         radeon_ring_write(ring, (0xf << 16) | (HDP_MEM_COHERENCY_FLUSH_CNTL >> 2));
3313         radeon_ring_write(ring, 1);
3314
3315         /* bits 0-7 are the VM contexts0-7 */
3316         radeon_ring_write(ring, DMA_PACKET(DMA_PACKET_SRBM_WRITE, 0, 0, 0, 0));
3317         radeon_ring_write(ring, (0xf << 16) | (VM_INVALIDATE_REQUEST >> 2));
3318         radeon_ring_write(ring, 1 << vm->id);
3319 }
3320
3321 /*
3322  * RLC
3323  */
3324 void si_rlc_fini(struct radeon_device *rdev)
3325 {
3326         int r;
3327
3328         /* save restore block */
3329         if (rdev->rlc.save_restore_obj) {
3330                 r = radeon_bo_reserve(rdev->rlc.save_restore_obj, false);
3331                 if (unlikely(r != 0))
3332                         dev_warn(rdev->dev, "(%d) reserve RLC sr bo failed\n", r);
3333                 radeon_bo_unpin(rdev->rlc.save_restore_obj);
3334                 radeon_bo_unreserve(rdev->rlc.save_restore_obj);
3335
3336                 radeon_bo_unref(&rdev->rlc.save_restore_obj);
3337                 rdev->rlc.save_restore_obj = NULL;
3338         }
3339
3340         /* clear state block */
3341         if (rdev->rlc.clear_state_obj) {
3342                 r = radeon_bo_reserve(rdev->rlc.clear_state_obj, false);
3343                 if (unlikely(r != 0))
3344                         dev_warn(rdev->dev, "(%d) reserve RLC c bo failed\n", r);
3345                 radeon_bo_unpin(rdev->rlc.clear_state_obj);
3346                 radeon_bo_unreserve(rdev->rlc.clear_state_obj);
3347
3348                 radeon_bo_unref(&rdev->rlc.clear_state_obj);
3349                 rdev->rlc.clear_state_obj = NULL;
3350         }
3351 }
3352
3353 int si_rlc_init(struct radeon_device *rdev)
3354 {
3355         int r;
3356
3357         /* save restore block */
3358         if (rdev->rlc.save_restore_obj == NULL) {
3359                 r = radeon_bo_create(rdev, RADEON_GPU_PAGE_SIZE, PAGE_SIZE, true,
3360                                      RADEON_GEM_DOMAIN_VRAM, NULL,
3361                                      &rdev->rlc.save_restore_obj);
3362                 if (r) {
3363                         dev_warn(rdev->dev, "(%d) create RLC sr bo failed\n", r);
3364                         return r;
3365                 }
3366         }
3367
3368         r = radeon_bo_reserve(rdev->rlc.save_restore_obj, false);
3369         if (unlikely(r != 0)) {
3370                 si_rlc_fini(rdev);
3371                 return r;
3372         }
3373         r = radeon_bo_pin(rdev->rlc.save_restore_obj, RADEON_GEM_DOMAIN_VRAM,
3374                           &rdev->rlc.save_restore_gpu_addr);
3375         radeon_bo_unreserve(rdev->rlc.save_restore_obj);
3376         if (r) {
3377                 dev_warn(rdev->dev, "(%d) pin RLC sr bo failed\n", r);
3378                 si_rlc_fini(rdev);
3379                 return r;
3380         }
3381
3382         /* clear state block */
3383         if (rdev->rlc.clear_state_obj == NULL) {
3384                 r = radeon_bo_create(rdev, RADEON_GPU_PAGE_SIZE, PAGE_SIZE, true,
3385                                      RADEON_GEM_DOMAIN_VRAM, NULL,
3386                                      &rdev->rlc.clear_state_obj);
3387                 if (r) {
3388                         dev_warn(rdev->dev, "(%d) create RLC c bo failed\n", r);
3389                         si_rlc_fini(rdev);
3390                         return r;
3391                 }
3392         }
3393         r = radeon_bo_reserve(rdev->rlc.clear_state_obj, false);
3394         if (unlikely(r != 0)) {
3395                 si_rlc_fini(rdev);
3396                 return r;
3397         }
3398         r = radeon_bo_pin(rdev->rlc.clear_state_obj, RADEON_GEM_DOMAIN_VRAM,
3399                           &rdev->rlc.clear_state_gpu_addr);
3400         radeon_bo_unreserve(rdev->rlc.clear_state_obj);
3401         if (r) {
3402                 dev_warn(rdev->dev, "(%d) pin RLC c bo failed\n", r);
3403                 si_rlc_fini(rdev);
3404                 return r;
3405         }
3406
3407         return 0;
3408 }
3409
3410 static void si_rlc_stop(struct radeon_device *rdev)
3411 {
3412         WREG32(RLC_CNTL, 0);
3413 }
3414
3415 static void si_rlc_start(struct radeon_device *rdev)
3416 {
3417         WREG32(RLC_CNTL, RLC_ENABLE);
3418 }
3419
3420 static int si_rlc_resume(struct radeon_device *rdev)
3421 {
3422         u32 i;
3423         const __be32 *fw_data;
3424
3425         if (!rdev->rlc_fw)
3426                 return -EINVAL;
3427
3428         si_rlc_stop(rdev);
3429
3430         WREG32(RLC_RL_BASE, 0);
3431         WREG32(RLC_RL_SIZE, 0);
3432         WREG32(RLC_LB_CNTL, 0);
3433         WREG32(RLC_LB_CNTR_MAX, 0xffffffff);
3434         WREG32(RLC_LB_CNTR_INIT, 0);
3435
3436         WREG32(RLC_SAVE_AND_RESTORE_BASE, rdev->rlc.save_restore_gpu_addr >> 8);
3437         WREG32(RLC_CLEAR_STATE_RESTORE_BASE, rdev->rlc.clear_state_gpu_addr >> 8);
3438
3439         WREG32(RLC_MC_CNTL, 0);
3440         WREG32(RLC_UCODE_CNTL, 0);
3441
3442         fw_data = (const __be32 *)rdev->rlc_fw->data;
3443         for (i = 0; i < SI_RLC_UCODE_SIZE; i++) {
3444                 WREG32(RLC_UCODE_ADDR, i);
3445                 WREG32(RLC_UCODE_DATA, be32_to_cpup(fw_data++));
3446         }
3447         WREG32(RLC_UCODE_ADDR, 0);
3448
3449         si_rlc_start(rdev);
3450
3451         return 0;
3452 }
3453
3454 static void si_enable_interrupts(struct radeon_device *rdev)
3455 {
3456         u32 ih_cntl = RREG32(IH_CNTL);
3457         u32 ih_rb_cntl = RREG32(IH_RB_CNTL);
3458
3459         ih_cntl |= ENABLE_INTR;
3460         ih_rb_cntl |= IH_RB_ENABLE;
3461         WREG32(IH_CNTL, ih_cntl);
3462         WREG32(IH_RB_CNTL, ih_rb_cntl);
3463         rdev->ih.enabled = true;
3464 }
3465
3466 static void si_disable_interrupts(struct radeon_device *rdev)
3467 {
3468         u32 ih_rb_cntl = RREG32(IH_RB_CNTL);
3469         u32 ih_cntl = RREG32(IH_CNTL);
3470
3471         ih_rb_cntl &= ~IH_RB_ENABLE;
3472         ih_cntl &= ~ENABLE_INTR;
3473         WREG32(IH_RB_CNTL, ih_rb_cntl);
3474         WREG32(IH_CNTL, ih_cntl);
3475         /* set rptr, wptr to 0 */
3476         WREG32(IH_RB_RPTR, 0);
3477         WREG32(IH_RB_WPTR, 0);
3478         rdev->ih.enabled = false;
3479         rdev->ih.rptr = 0;
3480 }
3481
3482 static void si_disable_interrupt_state(struct radeon_device *rdev)
3483 {
3484         u32 tmp;
3485
3486         WREG32(CP_INT_CNTL_RING0, CNTX_BUSY_INT_ENABLE | CNTX_EMPTY_INT_ENABLE);
3487         WREG32(CP_INT_CNTL_RING1, 0);
3488         WREG32(CP_INT_CNTL_RING2, 0);
3489         tmp = RREG32(DMA_CNTL + DMA0_REGISTER_OFFSET) & ~TRAP_ENABLE;
3490         WREG32(DMA_CNTL + DMA0_REGISTER_OFFSET, tmp);
3491         tmp = RREG32(DMA_CNTL + DMA1_REGISTER_OFFSET) & ~TRAP_ENABLE;
3492         WREG32(DMA_CNTL + DMA1_REGISTER_OFFSET, tmp);
3493         WREG32(GRBM_INT_CNTL, 0);
3494         WREG32(INT_MASK + EVERGREEN_CRTC0_REGISTER_OFFSET, 0);
3495         WREG32(INT_MASK + EVERGREEN_CRTC1_REGISTER_OFFSET, 0);
3496         if (rdev->num_crtc >= 4) {
3497                 WREG32(INT_MASK + EVERGREEN_CRTC2_REGISTER_OFFSET, 0);
3498                 WREG32(INT_MASK + EVERGREEN_CRTC3_REGISTER_OFFSET, 0);
3499         }
3500         if (rdev->num_crtc >= 6) {
3501                 WREG32(INT_MASK + EVERGREEN_CRTC4_REGISTER_OFFSET, 0);
3502                 WREG32(INT_MASK + EVERGREEN_CRTC5_REGISTER_OFFSET, 0);
3503         }
3504
3505         WREG32(GRPH_INT_CONTROL + EVERGREEN_CRTC0_REGISTER_OFFSET, 0);
3506         WREG32(GRPH_INT_CONTROL + EVERGREEN_CRTC1_REGISTER_OFFSET, 0);
3507         if (rdev->num_crtc >= 4) {
3508                 WREG32(GRPH_INT_CONTROL + EVERGREEN_CRTC2_REGISTER_OFFSET, 0);
3509                 WREG32(GRPH_INT_CONTROL + EVERGREEN_CRTC3_REGISTER_OFFSET, 0);
3510         }
3511         if (rdev->num_crtc >= 6) {
3512                 WREG32(GRPH_INT_CONTROL + EVERGREEN_CRTC4_REGISTER_OFFSET, 0);
3513                 WREG32(GRPH_INT_CONTROL + EVERGREEN_CRTC5_REGISTER_OFFSET, 0);
3514         }
3515
3516         WREG32(DACA_AUTODETECT_INT_CONTROL, 0);
3517
3518         tmp = RREG32(DC_HPD1_INT_CONTROL) & DC_HPDx_INT_POLARITY;
3519         WREG32(DC_HPD1_INT_CONTROL, tmp);
3520         tmp = RREG32(DC_HPD2_INT_CONTROL) & DC_HPDx_INT_POLARITY;
3521         WREG32(DC_HPD2_INT_CONTROL, tmp);
3522         tmp = RREG32(DC_HPD3_INT_CONTROL) & DC_HPDx_INT_POLARITY;
3523         WREG32(DC_HPD3_INT_CONTROL, tmp);
3524         tmp = RREG32(DC_HPD4_INT_CONTROL) & DC_HPDx_INT_POLARITY;
3525         WREG32(DC_HPD4_INT_CONTROL, tmp);
3526         tmp = RREG32(DC_HPD5_INT_CONTROL) & DC_HPDx_INT_POLARITY;
3527         WREG32(DC_HPD5_INT_CONTROL, tmp);
3528         tmp = RREG32(DC_HPD6_INT_CONTROL) & DC_HPDx_INT_POLARITY;
3529         WREG32(DC_HPD6_INT_CONTROL, tmp);
3530
3531 }
3532
3533 static int si_irq_init(struct radeon_device *rdev)
3534 {
3535         int ret = 0;
3536         int rb_bufsz;
3537         u32 interrupt_cntl, ih_cntl, ih_rb_cntl;
3538
3539         /* allocate ring */
3540         ret = r600_ih_ring_alloc(rdev);
3541         if (ret)
3542                 return ret;
3543
3544         /* disable irqs */
3545         si_disable_interrupts(rdev);
3546
3547         /* init rlc */
3548         ret = si_rlc_resume(rdev);
3549         if (ret) {
3550                 r600_ih_ring_fini(rdev);
3551                 return ret;
3552         }
3553
3554         /* setup interrupt control */
3555         /* set dummy read address to ring address */
3556         WREG32(INTERRUPT_CNTL2, rdev->ih.gpu_addr >> 8);
3557         interrupt_cntl = RREG32(INTERRUPT_CNTL);
3558         /* IH_DUMMY_RD_OVERRIDE=0 - dummy read disabled with msi, enabled without msi
3559          * IH_DUMMY_RD_OVERRIDE=1 - dummy read controlled by IH_DUMMY_RD_EN
3560          */
3561         interrupt_cntl &= ~IH_DUMMY_RD_OVERRIDE;
3562         /* IH_REQ_NONSNOOP_EN=1 if ring is in non-cacheable memory, e.g., vram */
3563         interrupt_cntl &= ~IH_REQ_NONSNOOP_EN;
3564         WREG32(INTERRUPT_CNTL, interrupt_cntl);
3565
3566         WREG32(IH_RB_BASE, rdev->ih.gpu_addr >> 8);
3567         rb_bufsz = drm_order(rdev->ih.ring_size / 4);
3568
3569         ih_rb_cntl = (IH_WPTR_OVERFLOW_ENABLE |
3570                       IH_WPTR_OVERFLOW_CLEAR |
3571                       (rb_bufsz << 1));
3572
3573         if (rdev->wb.enabled)
3574                 ih_rb_cntl |= IH_WPTR_WRITEBACK_ENABLE;
3575
3576         /* set the writeback address whether it's enabled or not */
3577         WREG32(IH_RB_WPTR_ADDR_LO, (rdev->wb.gpu_addr + R600_WB_IH_WPTR_OFFSET) & 0xFFFFFFFC);
3578         WREG32(IH_RB_WPTR_ADDR_HI, upper_32_bits(rdev->wb.gpu_addr + R600_WB_IH_WPTR_OFFSET) & 0xFF);
3579
3580         WREG32(IH_RB_CNTL, ih_rb_cntl);
3581
3582         /* set rptr, wptr to 0 */
3583         WREG32(IH_RB_RPTR, 0);
3584         WREG32(IH_RB_WPTR, 0);
3585
3586         /* Default settings for IH_CNTL (disabled at first) */
3587         ih_cntl = MC_WRREQ_CREDIT(0x10) | MC_WR_CLEAN_CNT(0x10) | MC_VMID(0);
3588         /* RPTR_REARM only works if msi's are enabled */
3589         if (rdev->msi_enabled)
3590                 ih_cntl |= RPTR_REARM;
3591         WREG32(IH_CNTL, ih_cntl);
3592
3593         /* force the active interrupt state to all disabled */
3594         si_disable_interrupt_state(rdev);
3595
3596         pci_set_master(rdev->pdev);
3597
3598         /* enable irqs */
3599         si_enable_interrupts(rdev);
3600
3601         return ret;
3602 }
3603
3604 int si_irq_set(struct radeon_device *rdev)
3605 {
3606         u32 cp_int_cntl = CNTX_BUSY_INT_ENABLE | CNTX_EMPTY_INT_ENABLE;
3607         u32 cp_int_cntl1 = 0, cp_int_cntl2 = 0;
3608         u32 crtc1 = 0, crtc2 = 0, crtc3 = 0, crtc4 = 0, crtc5 = 0, crtc6 = 0;
3609         u32 hpd1, hpd2, hpd3, hpd4, hpd5, hpd6;
3610         u32 grbm_int_cntl = 0;
3611         u32 grph1 = 0, grph2 = 0, grph3 = 0, grph4 = 0, grph5 = 0, grph6 = 0;
3612         u32 dma_cntl, dma_cntl1;
3613
3614         if (!rdev->irq.installed) {
3615                 WARN(1, "Can't enable IRQ/MSI because no handler is installed\n");
3616                 return -EINVAL;
3617         }
3618         /* don't enable anything if the ih is disabled */
3619         if (!rdev->ih.enabled) {
3620                 si_disable_interrupts(rdev);
3621                 /* force the active interrupt state to all disabled */
3622                 si_disable_interrupt_state(rdev);
3623                 return 0;
3624         }
3625
3626         hpd1 = RREG32(DC_HPD1_INT_CONTROL) & ~DC_HPDx_INT_EN;
3627         hpd2 = RREG32(DC_HPD2_INT_CONTROL) & ~DC_HPDx_INT_EN;
3628         hpd3 = RREG32(DC_HPD3_INT_CONTROL) & ~DC_HPDx_INT_EN;
3629         hpd4 = RREG32(DC_HPD4_INT_CONTROL) & ~DC_HPDx_INT_EN;
3630         hpd5 = RREG32(DC_HPD5_INT_CONTROL) & ~DC_HPDx_INT_EN;
3631         hpd6 = RREG32(DC_HPD6_INT_CONTROL) & ~DC_HPDx_INT_EN;
3632
3633         dma_cntl = RREG32(DMA_CNTL + DMA0_REGISTER_OFFSET) & ~TRAP_ENABLE;
3634         dma_cntl1 = RREG32(DMA_CNTL + DMA1_REGISTER_OFFSET) & ~TRAP_ENABLE;
3635
3636         /* enable CP interrupts on all rings */
3637         if (atomic_read(&rdev->irq.ring_int[RADEON_RING_TYPE_GFX_INDEX])) {
3638                 DRM_DEBUG("si_irq_set: sw int gfx\n");
3639                 cp_int_cntl |= TIME_STAMP_INT_ENABLE;
3640         }
3641         if (atomic_read(&rdev->irq.ring_int[CAYMAN_RING_TYPE_CP1_INDEX])) {
3642                 DRM_DEBUG("si_irq_set: sw int cp1\n");
3643                 cp_int_cntl1 |= TIME_STAMP_INT_ENABLE;
3644         }
3645         if (atomic_read(&rdev->irq.ring_int[CAYMAN_RING_TYPE_CP2_INDEX])) {
3646                 DRM_DEBUG("si_irq_set: sw int cp2\n");
3647                 cp_int_cntl2 |= TIME_STAMP_INT_ENABLE;
3648         }
3649         if (atomic_read(&rdev->irq.ring_int[R600_RING_TYPE_DMA_INDEX])) {
3650                 DRM_DEBUG("si_irq_set: sw int dma\n");
3651                 dma_cntl |= TRAP_ENABLE;
3652         }
3653
3654         if (atomic_read(&rdev->irq.ring_int[CAYMAN_RING_TYPE_DMA1_INDEX])) {
3655                 DRM_DEBUG("si_irq_set: sw int dma1\n");
3656                 dma_cntl1 |= TRAP_ENABLE;
3657         }
3658         if (rdev->irq.crtc_vblank_int[0] ||
3659             atomic_read(&rdev->irq.pflip[0])) {
3660                 DRM_DEBUG("si_irq_set: vblank 0\n");
3661                 crtc1 |= VBLANK_INT_MASK;
3662         }
3663         if (rdev->irq.crtc_vblank_int[1] ||
3664             atomic_read(&rdev->irq.pflip[1])) {
3665                 DRM_DEBUG("si_irq_set: vblank 1\n");
3666                 crtc2 |= VBLANK_INT_MASK;
3667         }
3668         if (rdev->irq.crtc_vblank_int[2] ||
3669             atomic_read(&rdev->irq.pflip[2])) {
3670                 DRM_DEBUG("si_irq_set: vblank 2\n");
3671                 crtc3 |= VBLANK_INT_MASK;
3672         }
3673         if (rdev->irq.crtc_vblank_int[3] ||
3674             atomic_read(&rdev->irq.pflip[3])) {
3675                 DRM_DEBUG("si_irq_set: vblank 3\n");
3676                 crtc4 |= VBLANK_INT_MASK;
3677         }
3678         if (rdev->irq.crtc_vblank_int[4] ||
3679             atomic_read(&rdev->irq.pflip[4])) {
3680                 DRM_DEBUG("si_irq_set: vblank 4\n");
3681                 crtc5 |= VBLANK_INT_MASK;
3682         }
3683         if (rdev->irq.crtc_vblank_int[5] ||
3684             atomic_read(&rdev->irq.pflip[5])) {
3685                 DRM_DEBUG("si_irq_set: vblank 5\n");
3686                 crtc6 |= VBLANK_INT_MASK;
3687         }
3688         if (rdev->irq.hpd[0]) {
3689                 DRM_DEBUG("si_irq_set: hpd 1\n");
3690                 hpd1 |= DC_HPDx_INT_EN;
3691         }
3692         if (rdev->irq.hpd[1]) {
3693                 DRM_DEBUG("si_irq_set: hpd 2\n");
3694                 hpd2 |= DC_HPDx_INT_EN;
3695         }
3696         if (rdev->irq.hpd[2]) {
3697                 DRM_DEBUG("si_irq_set: hpd 3\n");
3698                 hpd3 |= DC_HPDx_INT_EN;
3699         }
3700         if (rdev->irq.hpd[3]) {
3701                 DRM_DEBUG("si_irq_set: hpd 4\n");
3702                 hpd4 |= DC_HPDx_INT_EN;
3703         }
3704         if (rdev->irq.hpd[4]) {
3705                 DRM_DEBUG("si_irq_set: hpd 5\n");
3706                 hpd5 |= DC_HPDx_INT_EN;
3707         }
3708         if (rdev->irq.hpd[5]) {
3709                 DRM_DEBUG("si_irq_set: hpd 6\n");
3710                 hpd6 |= DC_HPDx_INT_EN;
3711         }
3712
3713         WREG32(CP_INT_CNTL_RING0, cp_int_cntl);
3714         WREG32(CP_INT_CNTL_RING1, cp_int_cntl1);
3715         WREG32(CP_INT_CNTL_RING2, cp_int_cntl2);
3716
3717         WREG32(DMA_CNTL + DMA0_REGISTER_OFFSET, dma_cntl);
3718         WREG32(DMA_CNTL + DMA1_REGISTER_OFFSET, dma_cntl1);
3719
3720         WREG32(GRBM_INT_CNTL, grbm_int_cntl);
3721
3722         WREG32(INT_MASK + EVERGREEN_CRTC0_REGISTER_OFFSET, crtc1);
3723         WREG32(INT_MASK + EVERGREEN_CRTC1_REGISTER_OFFSET, crtc2);
3724         if (rdev->num_crtc >= 4) {
3725                 WREG32(INT_MASK + EVERGREEN_CRTC2_REGISTER_OFFSET, crtc3);
3726                 WREG32(INT_MASK + EVERGREEN_CRTC3_REGISTER_OFFSET, crtc4);
3727         }
3728         if (rdev->num_crtc >= 6) {
3729                 WREG32(INT_MASK + EVERGREEN_CRTC4_REGISTER_OFFSET, crtc5);
3730                 WREG32(INT_MASK + EVERGREEN_CRTC5_REGISTER_OFFSET, crtc6);
3731         }
3732
3733         WREG32(GRPH_INT_CONTROL + EVERGREEN_CRTC0_REGISTER_OFFSET, grph1);
3734         WREG32(GRPH_INT_CONTROL + EVERGREEN_CRTC1_REGISTER_OFFSET, grph2);
3735         if (rdev->num_crtc >= 4) {
3736                 WREG32(GRPH_INT_CONTROL + EVERGREEN_CRTC2_REGISTER_OFFSET, grph3);
3737                 WREG32(GRPH_INT_CONTROL + EVERGREEN_CRTC3_REGISTER_OFFSET, grph4);
3738         }
3739         if (rdev->num_crtc >= 6) {
3740                 WREG32(GRPH_INT_CONTROL + EVERGREEN_CRTC4_REGISTER_OFFSET, grph5);
3741                 WREG32(GRPH_INT_CONTROL + EVERGREEN_CRTC5_REGISTER_OFFSET, grph6);
3742         }
3743
3744         WREG32(DC_HPD1_INT_CONTROL, hpd1);
3745         WREG32(DC_HPD2_INT_CONTROL, hpd2);
3746         WREG32(DC_HPD3_INT_CONTROL, hpd3);
3747         WREG32(DC_HPD4_INT_CONTROL, hpd4);
3748         WREG32(DC_HPD5_INT_CONTROL, hpd5);
3749         WREG32(DC_HPD6_INT_CONTROL, hpd6);
3750
3751         return 0;
3752 }
3753
3754 static inline void si_irq_ack(struct radeon_device *rdev)
3755 {
3756         u32 tmp;
3757
3758         rdev->irq.stat_regs.evergreen.disp_int = RREG32(DISP_INTERRUPT_STATUS);
3759         rdev->irq.stat_regs.evergreen.disp_int_cont = RREG32(DISP_INTERRUPT_STATUS_CONTINUE);
3760         rdev->irq.stat_regs.evergreen.disp_int_cont2 = RREG32(DISP_INTERRUPT_STATUS_CONTINUE2);
3761         rdev->irq.stat_regs.evergreen.disp_int_cont3 = RREG32(DISP_INTERRUPT_STATUS_CONTINUE3);
3762         rdev->irq.stat_regs.evergreen.disp_int_cont4 = RREG32(DISP_INTERRUPT_STATUS_CONTINUE4);
3763         rdev->irq.stat_regs.evergreen.disp_int_cont5 = RREG32(DISP_INTERRUPT_STATUS_CONTINUE5);
3764         rdev->irq.stat_regs.evergreen.d1grph_int = RREG32(GRPH_INT_STATUS + EVERGREEN_CRTC0_REGISTER_OFFSET);
3765         rdev->irq.stat_regs.evergreen.d2grph_int = RREG32(GRPH_INT_STATUS + EVERGREEN_CRTC1_REGISTER_OFFSET);
3766         if (rdev->num_crtc >= 4) {
3767                 rdev->irq.stat_regs.evergreen.d3grph_int = RREG32(GRPH_INT_STATUS + EVERGREEN_CRTC2_REGISTER_OFFSET);
3768                 rdev->irq.stat_regs.evergreen.d4grph_int = RREG32(GRPH_INT_STATUS + EVERGREEN_CRTC3_REGISTER_OFFSET);
3769         }
3770         if (rdev->num_crtc >= 6) {
3771                 rdev->irq.stat_regs.evergreen.d5grph_int = RREG32(GRPH_INT_STATUS + EVERGREEN_CRTC4_REGISTER_OFFSET);
3772                 rdev->irq.stat_regs.evergreen.d6grph_int = RREG32(GRPH_INT_STATUS + EVERGREEN_CRTC5_REGISTER_OFFSET);
3773         }
3774
3775         if (rdev->irq.stat_regs.evergreen.d1grph_int & GRPH_PFLIP_INT_OCCURRED)
3776                 WREG32(GRPH_INT_STATUS + EVERGREEN_CRTC0_REGISTER_OFFSET, GRPH_PFLIP_INT_CLEAR);
3777         if (rdev->irq.stat_regs.evergreen.d2grph_int & GRPH_PFLIP_INT_OCCURRED)
3778                 WREG32(GRPH_INT_STATUS + EVERGREEN_CRTC1_REGISTER_OFFSET, GRPH_PFLIP_INT_CLEAR);
3779         if (rdev->irq.stat_regs.evergreen.disp_int & LB_D1_VBLANK_INTERRUPT)
3780                 WREG32(VBLANK_STATUS + EVERGREEN_CRTC0_REGISTER_OFFSET, VBLANK_ACK);
3781         if (rdev->irq.stat_regs.evergreen.disp_int & LB_D1_VLINE_INTERRUPT)
3782                 WREG32(VLINE_STATUS + EVERGREEN_CRTC0_REGISTER_OFFSET, VLINE_ACK);
3783         if (rdev->irq.stat_regs.evergreen.disp_int_cont & LB_D2_VBLANK_INTERRUPT)
3784                 WREG32(VBLANK_STATUS + EVERGREEN_CRTC1_REGISTER_OFFSET, VBLANK_ACK);
3785         if (rdev->irq.stat_regs.evergreen.disp_int_cont & LB_D2_VLINE_INTERRUPT)
3786                 WREG32(VLINE_STATUS + EVERGREEN_CRTC1_REGISTER_OFFSET, VLINE_ACK);
3787
3788         if (rdev->num_crtc >= 4) {
3789                 if (rdev->irq.stat_regs.evergreen.d3grph_int & GRPH_PFLIP_INT_OCCURRED)
3790                         WREG32(GRPH_INT_STATUS + EVERGREEN_CRTC2_REGISTER_OFFSET, GRPH_PFLIP_INT_CLEAR);
3791                 if (rdev->irq.stat_regs.evergreen.d4grph_int & GRPH_PFLIP_INT_OCCURRED)
3792                         WREG32(GRPH_INT_STATUS + EVERGREEN_CRTC3_REGISTER_OFFSET, GRPH_PFLIP_INT_CLEAR);
3793                 if (rdev->irq.stat_regs.evergreen.disp_int_cont2 & LB_D3_VBLANK_INTERRUPT)
3794                         WREG32(VBLANK_STATUS + EVERGREEN_CRTC2_REGISTER_OFFSET, VBLANK_ACK);
3795                 if (rdev->irq.stat_regs.evergreen.disp_int_cont2 & LB_D3_VLINE_INTERRUPT)
3796                         WREG32(VLINE_STATUS + EVERGREEN_CRTC2_REGISTER_OFFSET, VLINE_ACK);
3797                 if (rdev->irq.stat_regs.evergreen.disp_int_cont3 & LB_D4_VBLANK_INTERRUPT)
3798                         WREG32(VBLANK_STATUS + EVERGREEN_CRTC3_REGISTER_OFFSET, VBLANK_ACK);
3799                 if (rdev->irq.stat_regs.evergreen.disp_int_cont3 & LB_D4_VLINE_INTERRUPT)
3800                         WREG32(VLINE_STATUS + EVERGREEN_CRTC3_REGISTER_OFFSET, VLINE_ACK);
3801         }
3802
3803         if (rdev->num_crtc >= 6) {
3804                 if (rdev->irq.stat_regs.evergreen.d5grph_int & GRPH_PFLIP_INT_OCCURRED)
3805                         WREG32(GRPH_INT_STATUS + EVERGREEN_CRTC4_REGISTER_OFFSET, GRPH_PFLIP_INT_CLEAR);
3806                 if (rdev->irq.stat_regs.evergreen.d6grph_int & GRPH_PFLIP_INT_OCCURRED)
3807                         WREG32(GRPH_INT_STATUS + EVERGREEN_CRTC5_REGISTER_OFFSET, GRPH_PFLIP_INT_CLEAR);
3808                 if (rdev->irq.stat_regs.evergreen.disp_int_cont4 & LB_D5_VBLANK_INTERRUPT)
3809                         WREG32(VBLANK_STATUS + EVERGREEN_CRTC4_REGISTER_OFFSET, VBLANK_ACK);
3810                 if (rdev->irq.stat_regs.evergreen.disp_int_cont4 & LB_D5_VLINE_INTERRUPT)
3811                         WREG32(VLINE_STATUS + EVERGREEN_CRTC4_REGISTER_OFFSET, VLINE_ACK);
3812                 if (rdev->irq.stat_regs.evergreen.disp_int_cont5 & LB_D6_VBLANK_INTERRUPT)
3813                         WREG32(VBLANK_STATUS + EVERGREEN_CRTC5_REGISTER_OFFSET, VBLANK_ACK);
3814                 if (rdev->irq.stat_regs.evergreen.disp_int_cont5 & LB_D6_VLINE_INTERRUPT)
3815                         WREG32(VLINE_STATUS + EVERGREEN_CRTC5_REGISTER_OFFSET, VLINE_ACK);
3816         }
3817
3818         if (rdev->irq.stat_regs.evergreen.disp_int & DC_HPD1_INTERRUPT) {
3819                 tmp = RREG32(DC_HPD1_INT_CONTROL);
3820                 tmp |= DC_HPDx_INT_ACK;
3821                 WREG32(DC_HPD1_INT_CONTROL, tmp);
3822         }
3823         if (rdev->irq.stat_regs.evergreen.disp_int_cont & DC_HPD2_INTERRUPT) {
3824                 tmp = RREG32(DC_HPD2_INT_CONTROL);
3825                 tmp |= DC_HPDx_INT_ACK;
3826                 WREG32(DC_HPD2_INT_CONTROL, tmp);
3827         }
3828         if (rdev->irq.stat_regs.evergreen.disp_int_cont2 & DC_HPD3_INTERRUPT) {
3829                 tmp = RREG32(DC_HPD3_INT_CONTROL);
3830                 tmp |= DC_HPDx_INT_ACK;
3831                 WREG32(DC_HPD3_INT_CONTROL, tmp);
3832         }
3833         if (rdev->irq.stat_regs.evergreen.disp_int_cont3 & DC_HPD4_INTERRUPT) {
3834                 tmp = RREG32(DC_HPD4_INT_CONTROL);
3835                 tmp |= DC_HPDx_INT_ACK;
3836                 WREG32(DC_HPD4_INT_CONTROL, tmp);
3837         }
3838         if (rdev->irq.stat_regs.evergreen.disp_int_cont4 & DC_HPD5_INTERRUPT) {
3839                 tmp = RREG32(DC_HPD5_INT_CONTROL);
3840                 tmp |= DC_HPDx_INT_ACK;
3841                 WREG32(DC_HPD5_INT_CONTROL, tmp);
3842         }
3843         if (rdev->irq.stat_regs.evergreen.disp_int_cont5 & DC_HPD6_INTERRUPT) {
3844                 tmp = RREG32(DC_HPD5_INT_CONTROL);
3845                 tmp |= DC_HPDx_INT_ACK;
3846                 WREG32(DC_HPD6_INT_CONTROL, tmp);
3847         }
3848 }
3849
3850 static void si_irq_disable(struct radeon_device *rdev)
3851 {
3852         si_disable_interrupts(rdev);
3853         /* Wait and acknowledge irq */
3854         mdelay(1);
3855         si_irq_ack(rdev);
3856         si_disable_interrupt_state(rdev);
3857 }
3858
3859 static void si_irq_suspend(struct radeon_device *rdev)
3860 {
3861         si_irq_disable(rdev);
3862         si_rlc_stop(rdev);
3863 }
3864
3865 static void si_irq_fini(struct radeon_device *rdev)
3866 {
3867         si_irq_suspend(rdev);
3868         r600_ih_ring_fini(rdev);
3869 }
3870
3871 static inline u32 si_get_ih_wptr(struct radeon_device *rdev)
3872 {
3873         u32 wptr, tmp;
3874
3875         if (rdev->wb.enabled)
3876                 wptr = le32_to_cpu(rdev->wb.wb[R600_WB_IH_WPTR_OFFSET/4]);
3877         else
3878                 wptr = RREG32(IH_RB_WPTR);
3879
3880         if (wptr & RB_OVERFLOW) {
3881                 /* When a ring buffer overflow happen start parsing interrupt
3882                  * from the last not overwritten vector (wptr + 16). Hopefully
3883                  * this should allow us to catchup.
3884                  */
3885                 dev_warn(rdev->dev, "IH ring buffer overflow (0x%08X, %d, %d)\n",
3886                         wptr, rdev->ih.rptr, (wptr + 16) + rdev->ih.ptr_mask);
3887                 rdev->ih.rptr = (wptr + 16) & rdev->ih.ptr_mask;
3888                 tmp = RREG32(IH_RB_CNTL);
3889                 tmp |= IH_WPTR_OVERFLOW_CLEAR;
3890                 WREG32(IH_RB_CNTL, tmp);
3891         }
3892         return (wptr & rdev->ih.ptr_mask);
3893 }
3894
3895 /*        SI IV Ring
3896  * Each IV ring entry is 128 bits:
3897  * [7:0]    - interrupt source id
3898  * [31:8]   - reserved
3899  * [59:32]  - interrupt source data
3900  * [63:60]  - reserved
3901  * [71:64]  - RINGID
3902  * [79:72]  - VMID
3903  * [127:80] - reserved
3904  */
3905 int si_irq_process(struct radeon_device *rdev)
3906 {
3907         u32 wptr;
3908         u32 rptr;
3909         u32 src_id, src_data, ring_id;
3910         u32 ring_index;
3911         bool queue_hotplug = false;
3912
3913         if (!rdev->ih.enabled || rdev->shutdown)
3914                 return IRQ_NONE;
3915
3916         wptr = si_get_ih_wptr(rdev);
3917
3918 restart_ih:
3919         /* is somebody else already processing irqs? */
3920         if (atomic_xchg(&rdev->ih.lock, 1))
3921                 return IRQ_NONE;
3922
3923         rptr = rdev->ih.rptr;
3924         DRM_DEBUG("si_irq_process start: rptr %d, wptr %d\n", rptr, wptr);
3925
3926         /* Order reading of wptr vs. reading of IH ring data */
3927         rmb();
3928
3929         /* display interrupts */
3930         si_irq_ack(rdev);
3931
3932         while (rptr != wptr) {
3933                 /* wptr/rptr are in bytes! */
3934                 ring_index = rptr / 4;
3935                 src_id =  le32_to_cpu(rdev->ih.ring[ring_index]) & 0xff;
3936                 src_data = le32_to_cpu(rdev->ih.ring[ring_index + 1]) & 0xfffffff;
3937                 ring_id = le32_to_cpu(rdev->ih.ring[ring_index + 2]) & 0xff;
3938
3939                 switch (src_id) {
3940                 case 1: /* D1 vblank/vline */
3941                         switch (src_data) {
3942                         case 0: /* D1 vblank */
3943                                 if (rdev->irq.stat_regs.evergreen.disp_int & LB_D1_VBLANK_INTERRUPT) {
3944                                         if (rdev->irq.crtc_vblank_int[0]) {
3945                                                 drm_handle_vblank(rdev->ddev, 0);
3946                                                 rdev->pm.vblank_sync = true;
3947                                                 wake_up(&rdev->irq.vblank_queue);
3948                                         }
3949                                         if (atomic_read(&rdev->irq.pflip[0]))
3950                                                 radeon_crtc_handle_flip(rdev, 0);
3951                                         rdev->irq.stat_regs.evergreen.disp_int &= ~LB_D1_VBLANK_INTERRUPT;
3952                                         DRM_DEBUG("IH: D1 vblank\n");
3953                                 }
3954                                 break;
3955                         case 1: /* D1 vline */
3956                                 if (rdev->irq.stat_regs.evergreen.disp_int & LB_D1_VLINE_INTERRUPT) {
3957                                         rdev->irq.stat_regs.evergreen.disp_int &= ~LB_D1_VLINE_INTERRUPT;
3958                                         DRM_DEBUG("IH: D1 vline\n");
3959                                 }
3960                                 break;
3961                         default:
3962                                 DRM_DEBUG("Unhandled interrupt: %d %d\n", src_id, src_data);
3963                                 break;
3964                         }
3965                         break;
3966                 case 2: /* D2 vblank/vline */
3967                         switch (src_data) {
3968                         case 0: /* D2 vblank */
3969                                 if (rdev->irq.stat_regs.evergreen.disp_int_cont & LB_D2_VBLANK_INTERRUPT) {
3970                                         if (rdev->irq.crtc_vblank_int[1]) {
3971                                                 drm_handle_vblank(rdev->ddev, 1);
3972                                                 rdev->pm.vblank_sync = true;
3973                                                 wake_up(&rdev->irq.vblank_queue);
3974                                         }
3975                                         if (atomic_read(&rdev->irq.pflip[1]))
3976                                                 radeon_crtc_handle_flip(rdev, 1);
3977                                         rdev->irq.stat_regs.evergreen.disp_int_cont &= ~LB_D2_VBLANK_INTERRUPT;
3978                                         DRM_DEBUG("IH: D2 vblank\n");
3979                                 }
3980                                 break;
3981                         case 1: /* D2 vline */
3982                                 if (rdev->irq.stat_regs.evergreen.disp_int_cont & LB_D2_VLINE_INTERRUPT) {
3983                                         rdev->irq.stat_regs.evergreen.disp_int_cont &= ~LB_D2_VLINE_INTERRUPT;
3984                                         DRM_DEBUG("IH: D2 vline\n");
3985                                 }
3986                                 break;
3987                         default:
3988                                 DRM_DEBUG("Unhandled interrupt: %d %d\n", src_id, src_data);
3989                                 break;
3990                         }
3991                         break;
3992                 case 3: /* D3 vblank/vline */
3993                         switch (src_data) {
3994                         case 0: /* D3 vblank */
3995                                 if (rdev->irq.stat_regs.evergreen.disp_int_cont2 & LB_D3_VBLANK_INTERRUPT) {
3996                                         if (rdev->irq.crtc_vblank_int[2]) {
3997                                                 drm_handle_vblank(rdev->ddev, 2);
3998                                                 rdev->pm.vblank_sync = true;
3999                                                 wake_up(&rdev->irq.vblank_queue);
4000                                         }
4001                                         if (atomic_read(&rdev->irq.pflip[2]))
4002                                                 radeon_crtc_handle_flip(rdev, 2);
4003                                         rdev->irq.stat_regs.evergreen.disp_int_cont2 &= ~LB_D3_VBLANK_INTERRUPT;
4004                                         DRM_DEBUG("IH: D3 vblank\n");
4005                                 }
4006                                 break;
4007                         case 1: /* D3 vline */
4008                                 if (rdev->irq.stat_regs.evergreen.disp_int_cont2 & LB_D3_VLINE_INTERRUPT) {
4009                                         rdev->irq.stat_regs.evergreen.disp_int_cont2 &= ~LB_D3_VLINE_INTERRUPT;
4010                                         DRM_DEBUG("IH: D3 vline\n");
4011                                 }
4012                                 break;
4013                         default:
4014                                 DRM_DEBUG("Unhandled interrupt: %d %d\n", src_id, src_data);
4015                                 break;
4016                         }
4017                         break;
4018                 case 4: /* D4 vblank/vline */
4019                         switch (src_data) {
4020                         case 0: /* D4 vblank */
4021                                 if (rdev->irq.stat_regs.evergreen.disp_int_cont3 & LB_D4_VBLANK_INTERRUPT) {
4022                                         if (rdev->irq.crtc_vblank_int[3]) {
4023                                                 drm_handle_vblank(rdev->ddev, 3);
4024                                                 rdev->pm.vblank_sync = true;
4025                                                 wake_up(&rdev->irq.vblank_queue);
4026                                         }
4027                                         if (atomic_read(&rdev->irq.pflip[3]))
4028                                                 radeon_crtc_handle_flip(rdev, 3);
4029                                         rdev->irq.stat_regs.evergreen.disp_int_cont3 &= ~LB_D4_VBLANK_INTERRUPT;
4030                                         DRM_DEBUG("IH: D4 vblank\n");
4031                                 }
4032                                 break;
4033                         case 1: /* D4 vline */
4034                                 if (rdev->irq.stat_regs.evergreen.disp_int_cont3 & LB_D4_VLINE_INTERRUPT) {
4035                                         rdev->irq.stat_regs.evergreen.disp_int_cont3 &= ~LB_D4_VLINE_INTERRUPT;
4036                                         DRM_DEBUG("IH: D4 vline\n");
4037                                 }
4038                                 break;
4039                         default:
4040                                 DRM_DEBUG("Unhandled interrupt: %d %d\n", src_id, src_data);
4041                                 break;
4042                         }
4043                         break;
4044                 case 5: /* D5 vblank/vline */
4045                         switch (src_data) {
4046                         case 0: /* D5 vblank */
4047                                 if (rdev->irq.stat_regs.evergreen.disp_int_cont4 & LB_D5_VBLANK_INTERRUPT) {
4048                                         if (rdev->irq.crtc_vblank_int[4]) {
4049                                                 drm_handle_vblank(rdev->ddev, 4);
4050                                                 rdev->pm.vblank_sync = true;
4051                                                 wake_up(&rdev->irq.vblank_queue);
4052                                         }
4053                                         if (atomic_read(&rdev->irq.pflip[4]))
4054                                                 radeon_crtc_handle_flip(rdev, 4);
4055                                         rdev->irq.stat_regs.evergreen.disp_int_cont4 &= ~LB_D5_VBLANK_INTERRUPT;
4056                                         DRM_DEBUG("IH: D5 vblank\n");
4057                                 }
4058                                 break;
4059                         case 1: /* D5 vline */
4060                                 if (rdev->irq.stat_regs.evergreen.disp_int_cont4 & LB_D5_VLINE_INTERRUPT) {
4061                                         rdev->irq.stat_regs.evergreen.disp_int_cont4 &= ~LB_D5_VLINE_INTERRUPT;
4062                                         DRM_DEBUG("IH: D5 vline\n");
4063                                 }
4064                                 break;
4065                         default:
4066                                 DRM_DEBUG("Unhandled interrupt: %d %d\n", src_id, src_data);
4067                                 break;
4068                         }
4069                         break;
4070                 case 6: /* D6 vblank/vline */
4071                         switch (src_data) {
4072                         case 0: /* D6 vblank */
4073                                 if (rdev->irq.stat_regs.evergreen.disp_int_cont5 & LB_D6_VBLANK_INTERRUPT) {
4074                                         if (rdev->irq.crtc_vblank_int[5]) {
4075                                                 drm_handle_vblank(rdev->ddev, 5);
4076                                                 rdev->pm.vblank_sync = true;
4077                                                 wake_up(&rdev->irq.vblank_queue);
4078                                         }
4079                                         if (atomic_read(&rdev->irq.pflip[5]))
4080                                                 radeon_crtc_handle_flip(rdev, 5);
4081                                         rdev->irq.stat_regs.evergreen.disp_int_cont5 &= ~LB_D6_VBLANK_INTERRUPT;
4082                                         DRM_DEBUG("IH: D6 vblank\n");
4083                                 }
4084                                 break;
4085                         case 1: /* D6 vline */
4086                                 if (rdev->irq.stat_regs.evergreen.disp_int_cont5 & LB_D6_VLINE_INTERRUPT) {
4087                                         rdev->irq.stat_regs.evergreen.disp_int_cont5 &= ~LB_D6_VLINE_INTERRUPT;
4088                                         DRM_DEBUG("IH: D6 vline\n");
4089                                 }
4090                                 break;
4091                         default:
4092                                 DRM_DEBUG("Unhandled interrupt: %d %d\n", src_id, src_data);
4093                                 break;
4094                         }
4095                         break;
4096                 case 42: /* HPD hotplug */
4097                         switch (src_data) {
4098                         case 0:
4099                                 if (rdev->irq.stat_regs.evergreen.disp_int & DC_HPD1_INTERRUPT) {
4100                                         rdev->irq.stat_regs.evergreen.disp_int &= ~DC_HPD1_INTERRUPT;
4101                                         queue_hotplug = true;
4102                                         DRM_DEBUG("IH: HPD1\n");
4103                                 }
4104                                 break;
4105                         case 1:
4106                                 if (rdev->irq.stat_regs.evergreen.disp_int_cont & DC_HPD2_INTERRUPT) {
4107                                         rdev->irq.stat_regs.evergreen.disp_int_cont &= ~DC_HPD2_INTERRUPT;
4108                                         queue_hotplug = true;
4109                                         DRM_DEBUG("IH: HPD2\n");
4110                                 }
4111                                 break;
4112                         case 2:
4113                                 if (rdev->irq.stat_regs.evergreen.disp_int_cont2 & DC_HPD3_INTERRUPT) {
4114                                         rdev->irq.stat_regs.evergreen.disp_int_cont2 &= ~DC_HPD3_INTERRUPT;
4115                                         queue_hotplug = true;
4116                                         DRM_DEBUG("IH: HPD3\n");
4117                                 }
4118                                 break;
4119                         case 3:
4120                                 if (rdev->irq.stat_regs.evergreen.disp_int_cont3 & DC_HPD4_INTERRUPT) {
4121                                         rdev->irq.stat_regs.evergreen.disp_int_cont3 &= ~DC_HPD4_INTERRUPT;
4122                                         queue_hotplug = true;
4123                                         DRM_DEBUG("IH: HPD4\n");
4124                                 }
4125                                 break;
4126                         case 4:
4127                                 if (rdev->irq.stat_regs.evergreen.disp_int_cont4 & DC_HPD5_INTERRUPT) {
4128                                         rdev->irq.stat_regs.evergreen.disp_int_cont4 &= ~DC_HPD5_INTERRUPT;
4129                                         queue_hotplug = true;
4130                                         DRM_DEBUG("IH: HPD5\n");
4131                                 }
4132                                 break;
4133                         case 5:
4134                                 if (rdev->irq.stat_regs.evergreen.disp_int_cont5 & DC_HPD6_INTERRUPT) {
4135                                         rdev->irq.stat_regs.evergreen.disp_int_cont5 &= ~DC_HPD6_INTERRUPT;
4136                                         queue_hotplug = true;
4137                                         DRM_DEBUG("IH: HPD6\n");
4138                                 }
4139                                 break;
4140                         default:
4141                                 DRM_DEBUG("Unhandled interrupt: %d %d\n", src_id, src_data);
4142                                 break;
4143                         }
4144                         break;
4145                 case 146:
4146                 case 147:
4147                         dev_err(rdev->dev, "GPU fault detected: %d 0x%08x\n", src_id, src_data);
4148                         dev_err(rdev->dev, "  VM_CONTEXT1_PROTECTION_FAULT_ADDR   0x%08X\n",
4149                                 RREG32(VM_CONTEXT1_PROTECTION_FAULT_ADDR));
4150                         dev_err(rdev->dev, "  VM_CONTEXT1_PROTECTION_FAULT_STATUS 0x%08X\n",
4151                                 RREG32(VM_CONTEXT1_PROTECTION_FAULT_STATUS));
4152                         /* reset addr and status */
4153                         WREG32_P(VM_CONTEXT1_CNTL2, 1, ~1);
4154                         break;
4155                 case 176: /* RINGID0 CP_INT */
4156                         radeon_fence_process(rdev, RADEON_RING_TYPE_GFX_INDEX);
4157                         break;
4158                 case 177: /* RINGID1 CP_INT */
4159                         radeon_fence_process(rdev, CAYMAN_RING_TYPE_CP1_INDEX);
4160                         break;
4161                 case 178: /* RINGID2 CP_INT */
4162                         radeon_fence_process(rdev, CAYMAN_RING_TYPE_CP2_INDEX);
4163                         break;
4164                 case 181: /* CP EOP event */
4165                         DRM_DEBUG("IH: CP EOP\n");
4166                         switch (ring_id) {
4167                         case 0:
4168                                 radeon_fence_process(rdev, RADEON_RING_TYPE_GFX_INDEX);
4169                                 break;
4170                         case 1:
4171                                 radeon_fence_process(rdev, CAYMAN_RING_TYPE_CP1_INDEX);
4172                                 break;
4173                         case 2:
4174                                 radeon_fence_process(rdev, CAYMAN_RING_TYPE_CP2_INDEX);
4175                                 break;
4176                         }
4177                         break;
4178                 case 224: /* DMA trap event */
4179                         DRM_DEBUG("IH: DMA trap\n");
4180                         radeon_fence_process(rdev, R600_RING_TYPE_DMA_INDEX);
4181                         break;
4182                 case 233: /* GUI IDLE */
4183                         DRM_DEBUG("IH: GUI idle\n");
4184                         break;
4185                 case 244: /* DMA trap event */
4186                         DRM_DEBUG("IH: DMA1 trap\n");
4187                         radeon_fence_process(rdev, CAYMAN_RING_TYPE_DMA1_INDEX);
4188                         break;
4189                 default:
4190                         DRM_DEBUG("Unhandled interrupt: %d %d\n", src_id, src_data);
4191                         break;
4192                 }
4193
4194                 /* wptr/rptr are in bytes! */
4195                 rptr += 16;
4196                 rptr &= rdev->ih.ptr_mask;
4197         }
4198         if (queue_hotplug)
4199                 schedule_work(&rdev->hotplug_work);
4200         rdev->ih.rptr = rptr;
4201         WREG32(IH_RB_RPTR, rdev->ih.rptr);
4202         atomic_set(&rdev->ih.lock, 0);
4203
4204         /* make sure wptr hasn't changed while processing */
4205         wptr = si_get_ih_wptr(rdev);
4206         if (wptr != rptr)
4207                 goto restart_ih;
4208
4209         return IRQ_HANDLED;
4210 }
4211
4212 /**
4213  * si_copy_dma - copy pages using the DMA engine
4214  *
4215  * @rdev: radeon_device pointer
4216  * @src_offset: src GPU address
4217  * @dst_offset: dst GPU address
4218  * @num_gpu_pages: number of GPU pages to xfer
4219  * @fence: radeon fence object
4220  *
4221  * Copy GPU paging using the DMA engine (SI).
4222  * Used by the radeon ttm implementation to move pages if
4223  * registered as the asic copy callback.
4224  */
4225 int si_copy_dma(struct radeon_device *rdev,
4226                 uint64_t src_offset, uint64_t dst_offset,
4227                 unsigned num_gpu_pages,
4228                 struct radeon_fence **fence)
4229 {
4230         struct radeon_semaphore *sem = NULL;
4231         int ring_index = rdev->asic->copy.dma_ring_index;
4232         struct radeon_ring *ring = &rdev->ring[ring_index];
4233         u32 size_in_bytes, cur_size_in_bytes;
4234         int i, num_loops;
4235         int r = 0;
4236
4237         r = radeon_semaphore_create(rdev, &sem);
4238         if (r) {
4239                 DRM_ERROR("radeon: moving bo (%d).\n", r);
4240                 return r;
4241         }
4242
4243         size_in_bytes = (num_gpu_pages << RADEON_GPU_PAGE_SHIFT);
4244         num_loops = DIV_ROUND_UP(size_in_bytes, 0xfffff);
4245         r = radeon_ring_lock(rdev, ring, num_loops * 5 + 11);
4246         if (r) {
4247                 DRM_ERROR("radeon: moving bo (%d).\n", r);
4248                 radeon_semaphore_free(rdev, &sem, NULL);
4249                 return r;
4250         }
4251
4252         if (radeon_fence_need_sync(*fence, ring->idx)) {
4253                 radeon_semaphore_sync_rings(rdev, sem, (*fence)->ring,
4254                                             ring->idx);
4255                 radeon_fence_note_sync(*fence, ring->idx);
4256         } else {
4257                 radeon_semaphore_free(rdev, &sem, NULL);
4258         }
4259
4260         for (i = 0; i < num_loops; i++) {
4261                 cur_size_in_bytes = size_in_bytes;
4262                 if (cur_size_in_bytes > 0xFFFFF)
4263                         cur_size_in_bytes = 0xFFFFF;
4264                 size_in_bytes -= cur_size_in_bytes;
4265                 radeon_ring_write(ring, DMA_PACKET(DMA_PACKET_COPY, 1, 0, 0, cur_size_in_bytes));
4266                 radeon_ring_write(ring, dst_offset & 0xffffffff);
4267                 radeon_ring_write(ring, src_offset & 0xffffffff);
4268                 radeon_ring_write(ring, upper_32_bits(dst_offset) & 0xff);
4269                 radeon_ring_write(ring, upper_32_bits(src_offset) & 0xff);
4270                 src_offset += cur_size_in_bytes;
4271                 dst_offset += cur_size_in_bytes;
4272         }
4273
4274         r = radeon_fence_emit(rdev, fence, ring->idx);
4275         if (r) {
4276                 radeon_ring_unlock_undo(rdev, ring);
4277                 return r;
4278         }
4279
4280         radeon_ring_unlock_commit(rdev, ring);
4281         radeon_semaphore_free(rdev, &sem, *fence);
4282
4283         return r;
4284 }
4285
4286 /*
4287  * startup/shutdown callbacks
4288  */
4289 static int si_startup(struct radeon_device *rdev)
4290 {
4291         struct radeon_ring *ring;
4292         int r;
4293
4294         if (!rdev->me_fw || !rdev->pfp_fw || !rdev->ce_fw ||
4295             !rdev->rlc_fw || !rdev->mc_fw) {
4296                 r = si_init_microcode(rdev);
4297                 if (r) {
4298                         DRM_ERROR("Failed to load firmware!\n");
4299                         return r;
4300                 }
4301         }
4302
4303         r = si_mc_load_microcode(rdev);
4304         if (r) {
4305                 DRM_ERROR("Failed to load MC firmware!\n");
4306                 return r;
4307         }
4308
4309         r = r600_vram_scratch_init(rdev);
4310         if (r)
4311                 return r;
4312
4313         si_mc_program(rdev);
4314         r = si_pcie_gart_enable(rdev);
4315         if (r)
4316                 return r;
4317         si_gpu_init(rdev);
4318
4319 #if 0
4320         r = evergreen_blit_init(rdev);
4321         if (r) {
4322                 r600_blit_fini(rdev);
4323                 rdev->asic->copy = NULL;
4324                 dev_warn(rdev->dev, "failed blitter (%d) falling back to memcpy\n", r);
4325         }
4326 #endif
4327         /* allocate rlc buffers */
4328         r = si_rlc_init(rdev);
4329         if (r) {
4330                 DRM_ERROR("Failed to init rlc BOs!\n");
4331                 return r;
4332         }
4333
4334         /* allocate wb buffer */
4335         r = radeon_wb_init(rdev);
4336         if (r)
4337                 return r;
4338
4339         r = radeon_fence_driver_start_ring(rdev, RADEON_RING_TYPE_GFX_INDEX);
4340         if (r) {
4341                 dev_err(rdev->dev, "failed initializing CP fences (%d).\n", r);
4342                 return r;
4343         }
4344
4345         r = radeon_fence_driver_start_ring(rdev, CAYMAN_RING_TYPE_CP1_INDEX);
4346         if (r) {
4347                 dev_err(rdev->dev, "failed initializing CP fences (%d).\n", r);
4348                 return r;
4349         }
4350
4351         r = radeon_fence_driver_start_ring(rdev, CAYMAN_RING_TYPE_CP2_INDEX);
4352         if (r) {
4353                 dev_err(rdev->dev, "failed initializing CP fences (%d).\n", r);
4354                 return r;
4355         }
4356
4357         r = radeon_fence_driver_start_ring(rdev, R600_RING_TYPE_DMA_INDEX);
4358         if (r) {
4359                 dev_err(rdev->dev, "failed initializing DMA fences (%d).\n", r);
4360                 return r;
4361         }
4362
4363         r = radeon_fence_driver_start_ring(rdev, CAYMAN_RING_TYPE_DMA1_INDEX);
4364         if (r) {
4365                 dev_err(rdev->dev, "failed initializing DMA fences (%d).\n", r);
4366                 return r;
4367         }
4368
4369         /* Enable IRQ */
4370         r = si_irq_init(rdev);
4371         if (r) {
4372                 DRM_ERROR("radeon: IH init failed (%d).\n", r);
4373                 radeon_irq_kms_fini(rdev);
4374                 return r;
4375         }
4376         si_irq_set(rdev);
4377
4378         ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
4379         r = radeon_ring_init(rdev, ring, ring->ring_size, RADEON_WB_CP_RPTR_OFFSET,
4380                              CP_RB0_RPTR, CP_RB0_WPTR,
4381                              0, 0xfffff, RADEON_CP_PACKET2);
4382         if (r)
4383                 return r;
4384
4385         ring = &rdev->ring[CAYMAN_RING_TYPE_CP1_INDEX];
4386         r = radeon_ring_init(rdev, ring, ring->ring_size, RADEON_WB_CP1_RPTR_OFFSET,
4387                              CP_RB1_RPTR, CP_RB1_WPTR,
4388                              0, 0xfffff, RADEON_CP_PACKET2);
4389         if (r)
4390                 return r;
4391
4392         ring = &rdev->ring[CAYMAN_RING_TYPE_CP2_INDEX];
4393         r = radeon_ring_init(rdev, ring, ring->ring_size, RADEON_WB_CP2_RPTR_OFFSET,
4394                              CP_RB2_RPTR, CP_RB2_WPTR,
4395                              0, 0xfffff, RADEON_CP_PACKET2);
4396         if (r)
4397                 return r;
4398
4399         ring = &rdev->ring[R600_RING_TYPE_DMA_INDEX];
4400         r = radeon_ring_init(rdev, ring, ring->ring_size, R600_WB_DMA_RPTR_OFFSET,
4401                              DMA_RB_RPTR + DMA0_REGISTER_OFFSET,
4402                              DMA_RB_WPTR + DMA0_REGISTER_OFFSET,
4403                              2, 0x3fffc, DMA_PACKET(DMA_PACKET_NOP, 0, 0, 0, 0));
4404         if (r)
4405                 return r;
4406
4407         ring = &rdev->ring[CAYMAN_RING_TYPE_DMA1_INDEX];
4408         r = radeon_ring_init(rdev, ring, ring->ring_size, CAYMAN_WB_DMA1_RPTR_OFFSET,
4409                              DMA_RB_RPTR + DMA1_REGISTER_OFFSET,
4410                              DMA_RB_WPTR + DMA1_REGISTER_OFFSET,
4411                              2, 0x3fffc, DMA_PACKET(DMA_PACKET_NOP, 0, 0, 0, 0));
4412         if (r)
4413                 return r;
4414
4415         r = si_cp_load_microcode(rdev);
4416         if (r)
4417                 return r;
4418         r = si_cp_resume(rdev);
4419         if (r)
4420                 return r;
4421
4422         r = cayman_dma_resume(rdev);
4423         if (r)
4424                 return r;
4425
4426         r = radeon_ib_pool_init(rdev);
4427         if (r) {
4428                 dev_err(rdev->dev, "IB initialization failed (%d).\n", r);
4429                 return r;
4430         }
4431
4432         r = radeon_vm_manager_init(rdev);
4433         if (r) {
4434                 dev_err(rdev->dev, "vm manager initialization failed (%d).\n", r);
4435                 return r;
4436         }
4437
4438         return 0;
4439 }
4440
4441 int si_resume(struct radeon_device *rdev)
4442 {
4443         int r;
4444
4445         /* Do not reset GPU before posting, on rv770 hw unlike on r500 hw,
4446          * posting will perform necessary task to bring back GPU into good
4447          * shape.
4448          */
4449         /* post card */
4450         atom_asic_init(rdev->mode_info.atom_context);
4451
4452         rdev->accel_working = true;
4453         r = si_startup(rdev);
4454         if (r) {
4455                 DRM_ERROR("si startup failed on resume\n");
4456                 rdev->accel_working = false;
4457                 return r;
4458         }
4459
4460         return r;
4461
4462 }
4463
4464 int si_suspend(struct radeon_device *rdev)
4465 {
4466         si_cp_enable(rdev, false);
4467         cayman_dma_stop(rdev);
4468         si_irq_suspend(rdev);
4469         radeon_wb_disable(rdev);
4470         si_pcie_gart_disable(rdev);
4471         return 0;
4472 }
4473
4474 /* Plan is to move initialization in that function and use
4475  * helper function so that radeon_device_init pretty much
4476  * do nothing more than calling asic specific function. This
4477  * should also allow to remove a bunch of callback function
4478  * like vram_info.
4479  */
4480 int si_init(struct radeon_device *rdev)
4481 {
4482         struct radeon_ring *ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
4483         int r;
4484
4485         /* Read BIOS */
4486         if (!radeon_get_bios(rdev)) {
4487                 if (ASIC_IS_AVIVO(rdev))
4488                         return -EINVAL;
4489         }
4490         /* Must be an ATOMBIOS */
4491         if (!rdev->is_atom_bios) {
4492                 dev_err(rdev->dev, "Expecting atombios for cayman GPU\n");
4493                 return -EINVAL;
4494         }
4495         r = radeon_atombios_init(rdev);
4496         if (r)
4497                 return r;
4498
4499         /* Post card if necessary */
4500         if (!radeon_card_posted(rdev)) {
4501                 if (!rdev->bios) {
4502                         dev_err(rdev->dev, "Card not posted and no BIOS - ignoring\n");
4503                         return -EINVAL;
4504                 }
4505                 DRM_INFO("GPU not posted. posting now...\n");
4506                 atom_asic_init(rdev->mode_info.atom_context);
4507         }
4508         /* Initialize scratch registers */
4509         si_scratch_init(rdev);
4510         /* Initialize surface registers */
4511         radeon_surface_init(rdev);
4512         /* Initialize clocks */
4513         radeon_get_clock_info(rdev->ddev);
4514
4515         /* Fence driver */
4516         r = radeon_fence_driver_init(rdev);
4517         if (r)
4518                 return r;
4519
4520         /* initialize memory controller */
4521         r = si_mc_init(rdev);
4522         if (r)
4523                 return r;
4524         /* Memory manager */
4525         r = radeon_bo_init(rdev);
4526         if (r)
4527                 return r;
4528
4529         r = radeon_irq_kms_init(rdev);
4530         if (r)
4531                 return r;
4532
4533         ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
4534         ring->ring_obj = NULL;
4535         r600_ring_init(rdev, ring, 1024 * 1024);
4536
4537         ring = &rdev->ring[CAYMAN_RING_TYPE_CP1_INDEX];
4538         ring->ring_obj = NULL;
4539         r600_ring_init(rdev, ring, 1024 * 1024);
4540
4541         ring = &rdev->ring[CAYMAN_RING_TYPE_CP2_INDEX];
4542         ring->ring_obj = NULL;
4543         r600_ring_init(rdev, ring, 1024 * 1024);
4544
4545         ring = &rdev->ring[R600_RING_TYPE_DMA_INDEX];
4546         ring->ring_obj = NULL;
4547         r600_ring_init(rdev, ring, 64 * 1024);
4548
4549         ring = &rdev->ring[CAYMAN_RING_TYPE_DMA1_INDEX];
4550         ring->ring_obj = NULL;
4551         r600_ring_init(rdev, ring, 64 * 1024);
4552
4553         rdev->ih.ring_obj = NULL;
4554         r600_ih_ring_init(rdev, 64 * 1024);
4555
4556         r = r600_pcie_gart_init(rdev);
4557         if (r)
4558                 return r;
4559
4560         rdev->accel_working = true;
4561         r = si_startup(rdev);
4562         if (r) {
4563                 dev_err(rdev->dev, "disabling GPU acceleration\n");
4564                 si_cp_fini(rdev);
4565                 cayman_dma_fini(rdev);
4566                 si_irq_fini(rdev);
4567                 si_rlc_fini(rdev);
4568                 radeon_wb_fini(rdev);
4569                 radeon_ib_pool_fini(rdev);
4570                 radeon_vm_manager_fini(rdev);
4571                 radeon_irq_kms_fini(rdev);
4572                 si_pcie_gart_fini(rdev);
4573                 rdev->accel_working = false;
4574         }
4575
4576         /* Don't start up if the MC ucode is missing.
4577          * The default clocks and voltages before the MC ucode
4578          * is loaded are not suffient for advanced operations.
4579          */
4580         if (!rdev->mc_fw) {
4581                 DRM_ERROR("radeon: MC ucode required for NI+.\n");
4582                 return -EINVAL;
4583         }
4584
4585         return 0;
4586 }
4587
4588 void si_fini(struct radeon_device *rdev)
4589 {
4590 #if 0
4591         r600_blit_fini(rdev);
4592 #endif
4593         si_cp_fini(rdev);
4594         cayman_dma_fini(rdev);
4595         si_irq_fini(rdev);
4596         si_rlc_fini(rdev);
4597         radeon_wb_fini(rdev);
4598         radeon_vm_manager_fini(rdev);
4599         radeon_ib_pool_fini(rdev);
4600         radeon_irq_kms_fini(rdev);
4601         si_pcie_gart_fini(rdev);
4602         r600_vram_scratch_fini(rdev);
4603         radeon_gem_fini(rdev);
4604         radeon_fence_driver_fini(rdev);
4605         radeon_bo_fini(rdev);
4606         radeon_atombios_fini(rdev);
4607         kfree(rdev->bios);
4608         rdev->bios = NULL;
4609 }
4610
4611 /**
4612  * si_get_gpu_clock_counter - return GPU clock counter snapshot
4613  *
4614  * @rdev: radeon_device pointer
4615  *
4616  * Fetches a GPU clock counter snapshot (SI).
4617  * Returns the 64 bit clock counter snapshot.
4618  */
4619 uint64_t si_get_gpu_clock_counter(struct radeon_device *rdev)
4620 {
4621         uint64_t clock;
4622
4623         mutex_lock(&rdev->gpu_clock_mutex);
4624         WREG32(RLC_CAPTURE_GPU_CLOCK_COUNT, 1);
4625         clock = (uint64_t)RREG32(RLC_GPU_CLOCK_COUNT_LSB) |
4626                 ((uint64_t)RREG32(RLC_GPU_CLOCK_COUNT_MSB) << 32ULL);
4627         mutex_unlock(&rdev->gpu_clock_mutex);
4628         return clock;
4629 }