drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c

   1 /*
   2  * Copyright 2014 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  */
  23
  24 #include <linux/firmware.h>
  25 #include <linux/module.h>
  26 #include <linux/pci.h>
  27
  28 #include <drm/drm_cache.h>
  29 #include "amdgpu.h"
  30 #include "gmc_v8_0.h"
  31 #include "amdgpu_ucode.h"
  32 #include "amdgpu_amdkfd.h"
  33 #include "amdgpu_gem.h"
  34
  35 #include "gmc/gmc_8_1_d.h"
  36 #include "gmc/gmc_8_1_sh_mask.h"
  37
  38 #include "bif/bif_5_0_d.h"
  39 #include "bif/bif_5_0_sh_mask.h"
  40
  41 #include "oss/oss_3_0_d.h"
  42 #include "oss/oss_3_0_sh_mask.h"
  43
  44 #include "dce/dce_10_0_d.h"
  45 #include "dce/dce_10_0_sh_mask.h"
  46
  47 #include "vid.h"
  48 #include "vi.h"
  49
  50 #include "amdgpu_atombios.h"
  51
  52 #include "ivsrcid/ivsrcid_vislands30.h"
  53
  54 static void gmc_v8_0_set_gmc_funcs(struct amdgpu_device *adev);
  55 static void gmc_v8_0_set_irq_funcs(struct amdgpu_device *adev);
  56 static int gmc_v8_0_wait_for_idle(void *handle);
  57
  58 MODULE_FIRMWARE("amdgpu/tonga_mc.bin");
  59 MODULE_FIRMWARE("amdgpu/polaris11_mc.bin");
  60 MODULE_FIRMWARE("amdgpu/polaris10_mc.bin");
  61 MODULE_FIRMWARE("amdgpu/polaris12_mc.bin");
  62 MODULE_FIRMWARE("amdgpu/polaris11_k_mc.bin");
  63 MODULE_FIRMWARE("amdgpu/polaris10_k_mc.bin");
  64 MODULE_FIRMWARE("amdgpu/polaris12_k_mc.bin");
  65
  66 static const u32 golden_settings_tonga_a11[] =
  67 {
  68         mmMC_ARB_WTM_GRPWT_RD, 0x00000003, 0x00000000,
  69         mmMC_HUB_RDREQ_DMIF_LIMIT, 0x0000007f, 0x00000028,
  70         mmMC_HUB_WDP_UMC, 0x00007fb6, 0x00000991,
  71         mmVM_PRT_APERTURE0_LOW_ADDR, 0x0fffffff, 0x0fffffff,
  72         mmVM_PRT_APERTURE1_LOW_ADDR, 0x0fffffff, 0x0fffffff,
  73         mmVM_PRT_APERTURE2_LOW_ADDR, 0x0fffffff, 0x0fffffff,
  74         mmVM_PRT_APERTURE3_LOW_ADDR, 0x0fffffff, 0x0fffffff,
  75 };
  76
  77 static const u32 tonga_mgcg_cgcg_init[] =
  78 {
  79         mmMC_MEM_POWER_LS, 0xffffffff, 0x00000104
  80 };
  81
  82 static const u32 golden_settings_fiji_a10[] =
  83 {
  84         mmVM_PRT_APERTURE0_LOW_ADDR, 0x0fffffff, 0x0fffffff,
  85         mmVM_PRT_APERTURE1_LOW_ADDR, 0x0fffffff, 0x0fffffff,
  86         mmVM_PRT_APERTURE2_LOW_ADDR, 0x0fffffff, 0x0fffffff,
  87         mmVM_PRT_APERTURE3_LOW_ADDR, 0x0fffffff, 0x0fffffff,
  88 };
  89
  90 static const u32 fiji_mgcg_cgcg_init[] =
  91 {
  92         mmMC_MEM_POWER_LS, 0xffffffff, 0x00000104
  93 };
  94
  95 static const u32 golden_settings_polaris11_a11[] =
  96 {
  97         mmVM_PRT_APERTURE0_LOW_ADDR, 0x0fffffff, 0x0fffffff,
  98         mmVM_PRT_APERTURE1_LOW_ADDR, 0x0fffffff, 0x0fffffff,
  99         mmVM_PRT_APERTURE2_LOW_ADDR, 0x0fffffff, 0x0fffffff,
 100         mmVM_PRT_APERTURE3_LOW_ADDR, 0x0fffffff, 0x0fffffff
 101 };
 102
 103 static const u32 golden_settings_polaris10_a11[] =
 104 {
 105         mmMC_ARB_WTM_GRPWT_RD, 0x00000003, 0x00000000,
 106         mmVM_PRT_APERTURE0_LOW_ADDR, 0x0fffffff, 0x0fffffff,
 107         mmVM_PRT_APERTURE1_LOW_ADDR, 0x0fffffff, 0x0fffffff,
 108         mmVM_PRT_APERTURE2_LOW_ADDR, 0x0fffffff, 0x0fffffff,
 109         mmVM_PRT_APERTURE3_LOW_ADDR, 0x0fffffff, 0x0fffffff
 110 };
 111
 112 static const u32 cz_mgcg_cgcg_init[] =
 113 {
 114         mmMC_MEM_POWER_LS, 0xffffffff, 0x00000104
 115 };
 116
 117 static const u32 stoney_mgcg_cgcg_init[] =
 118 {
 119         mmATC_MISC_CG, 0xffffffff, 0x000c0200,
 120         mmMC_MEM_POWER_LS, 0xffffffff, 0x00000104
 121 };
 122
 123 static const u32 golden_settings_stoney_common[] =
 124 {
 125         mmMC_HUB_RDREQ_UVD, MC_HUB_RDREQ_UVD__PRESCALE_MASK, 0x00000004,
 126         mmMC_RD_GRP_OTH, MC_RD_GRP_OTH__UVD_MASK, 0x00600000
 127 };
 128
 129 static void gmc_v8_0_init_golden_registers(struct amdgpu_device *adev)
 130 {
 131         switch (adev->asic_type) {
 132         case CHIP_FIJI:
 133                 amdgpu_device_program_register_sequence(adev,
 134                                                         fiji_mgcg_cgcg_init,
 135                                                         ARRAY_SIZE(fiji_mgcg_cgcg_init));
 136                 amdgpu_device_program_register_sequence(adev,
 137                                                         golden_settings_fiji_a10,
 138                                                         ARRAY_SIZE(golden_settings_fiji_a10));
 139                 break;
 140         case CHIP_TONGA:
 141                 amdgpu_device_program_register_sequence(adev,
 142                                                         tonga_mgcg_cgcg_init,
 143                                                         ARRAY_SIZE(tonga_mgcg_cgcg_init));
 144                 amdgpu_device_program_register_sequence(adev,
 145                                                         golden_settings_tonga_a11,
 146                                                         ARRAY_SIZE(golden_settings_tonga_a11));
 147                 break;
 148         case CHIP_POLARIS11:
 149         case CHIP_POLARIS12:
 150         case CHIP_VEGAM:
 151                 amdgpu_device_program_register_sequence(adev,
 152                                                         golden_settings_polaris11_a11,
 153                                                         ARRAY_SIZE(golden_settings_polaris11_a11));
 154                 break;
 155         case CHIP_POLARIS10:
 156                 amdgpu_device_program_register_sequence(adev,
 157                                                         golden_settings_polaris10_a11,
 158                                                         ARRAY_SIZE(golden_settings_polaris10_a11));
 159                 break;
 160         case CHIP_CARRIZO:
 161                 amdgpu_device_program_register_sequence(adev,
 162                                                         cz_mgcg_cgcg_init,
 163                                                         ARRAY_SIZE(cz_mgcg_cgcg_init));
 164                 break;
 165         case CHIP_STONEY:
 166                 amdgpu_device_program_register_sequence(adev,
 167                                                         stoney_mgcg_cgcg_init,
 168                                                         ARRAY_SIZE(stoney_mgcg_cgcg_init));
 169                 amdgpu_device_program_register_sequence(adev,
 170                                                         golden_settings_stoney_common,
 171                                                         ARRAY_SIZE(golden_settings_stoney_common));
 172                 break;
 173         default:
 174                 break;
 175         }
 176 }
 177
 178 static void gmc_v8_0_mc_stop(struct amdgpu_device *adev)
 179 {
 180         u32 blackout;
 181
 182         gmc_v8_0_wait_for_idle(adev);
 183
 184         blackout = RREG32(mmMC_SHARED_BLACKOUT_CNTL);
 185         if (REG_GET_FIELD(blackout, MC_SHARED_BLACKOUT_CNTL, BLACKOUT_MODE) != 1) {
 186                 /* Block CPU access */
 187                 WREG32(mmBIF_FB_EN, 0);
 188                 /* blackout the MC */
 189                 blackout = REG_SET_FIELD(blackout,
 190                                          MC_SHARED_BLACKOUT_CNTL, BLACKOUT_MODE, 1);
 191                 WREG32(mmMC_SHARED_BLACKOUT_CNTL, blackout);
 192         }
 193         /* wait for the MC to settle */
 194         udelay(100);
 195 }
 196
 197 static void gmc_v8_0_mc_resume(struct amdgpu_device *adev)
 198 {
 199         u32 tmp;
 200
 201         /* unblackout the MC */
 202         tmp = RREG32(mmMC_SHARED_BLACKOUT_CNTL);
 203         tmp = REG_SET_FIELD(tmp, MC_SHARED_BLACKOUT_CNTL, BLACKOUT_MODE, 0);
 204         WREG32(mmMC_SHARED_BLACKOUT_CNTL, tmp);
 205         /* allow CPU access */
 206         tmp = REG_SET_FIELD(0, BIF_FB_EN, FB_READ_EN, 1);
 207         tmp = REG_SET_FIELD(tmp, BIF_FB_EN, FB_WRITE_EN, 1);
 208         WREG32(mmBIF_FB_EN, tmp);
 209 }
 210
 211 /**
 212  * gmc_v8_0_init_microcode - load ucode images from disk
 213  *
 214  * @adev: amdgpu_device pointer
 215  *
 216  * Use the firmware interface to load the ucode images into
 217  * the driver (not loaded into hw).
 218  * Returns 0 on success, error on failure.
 219  */
 220 static int gmc_v8_0_init_microcode(struct amdgpu_device *adev)
 221 {
 222         const char *chip_name;
 223         char fw_name[30];
 224         int err;
 225
 226         DRM_DEBUG("\n");
 227
 228         switch (adev->asic_type) {
 229         case CHIP_TONGA:
 230                 chip_name = "tonga";
 231                 break;
 232         case CHIP_POLARIS11:
 233                 if (((adev->pdev->device == 0x67ef) &&
 234                      ((adev->pdev->revision == 0xe0) ||
 235                       (adev->pdev->revision == 0xe5))) ||
 236                     ((adev->pdev->device == 0x67ff) &&
 237                      ((adev->pdev->revision == 0xcf) ||
 238                       (adev->pdev->revision == 0xef) ||
 239                       (adev->pdev->revision == 0xff))))
 240                         chip_name = "polaris11_k";
 241                 else if ((adev->pdev->device == 0x67ef) &&
 242                          (adev->pdev->revision == 0xe2))
 243                         chip_name = "polaris11_k";
 244                 else
 245                         chip_name = "polaris11";
 246                 break;
 247         case CHIP_POLARIS10:
 248                 if ((adev->pdev->device == 0x67df) &&
 249                     ((adev->pdev->revision == 0xe1) ||
 250                      (adev->pdev->revision == 0xf7)))
 251                         chip_name = "polaris10_k";
 252                 else
 253                         chip_name = "polaris10";
 254                 break;
 255         case CHIP_POLARIS12:
 256                 if (((adev->pdev->device == 0x6987) &&
 257                      ((adev->pdev->revision == 0xc0) ||
 258                       (adev->pdev->revision == 0xc3))) ||
 259                     ((adev->pdev->device == 0x6981) &&
 260                      ((adev->pdev->revision == 0x00) ||
 261                       (adev->pdev->revision == 0x01) ||
 262                       (adev->pdev->revision == 0x10))))
 263                         chip_name = "polaris12_k";
 264                 else
 265                         chip_name = "polaris12";
 266                 break;
 267         case CHIP_FIJI:
 268         case CHIP_CARRIZO:
 269         case CHIP_STONEY:
 270         case CHIP_VEGAM:
 271                 return 0;
 272         default: BUG();
 273         }
 274
 275         snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_mc.bin", chip_name);
 276         err = request_firmware(&adev->gmc.fw, fw_name, adev->dev);
 277         if (err)
 278                 goto out;
 279         err = amdgpu_ucode_validate(adev->gmc.fw);
 280
 281 out:
 282         if (err) {
 283                 pr_err("mc: Failed to load firmware \"%s\"\n", fw_name);
 284                 release_firmware(adev->gmc.fw);
 285                 adev->gmc.fw = NULL;
 286         }
 287         return err;
 288 }
 289
 290 /**
 291  * gmc_v8_0_tonga_mc_load_microcode - load tonga MC ucode into the hw
 292  *
 293  * @adev: amdgpu_device pointer
 294  *
 295  * Load the GDDR MC ucode into the hw (VI).
 296  * Returns 0 on success, error on failure.
 297  */
 298 static int gmc_v8_0_tonga_mc_load_microcode(struct amdgpu_device *adev)
 299 {
 300         const struct mc_firmware_header_v1_0 *hdr;
 301         const __le32 *fw_data = NULL;
 302         const __le32 *io_mc_regs = NULL;
 303         u32 running;
 304         int i, ucode_size, regs_size;
 305
 306         /* Skip MC ucode loading on SR-IOV capable boards.
 307          * vbios does this for us in asic_init in that case.
 308          * Skip MC ucode loading on VF, because hypervisor will do that
 309          * for this adaptor.
 310          */
 311         if (amdgpu_sriov_bios(adev))
 312                 return 0;
 313
 314         if (!adev->gmc.fw)
 315                 return -EINVAL;
 316
 317         hdr = (const struct mc_firmware_header_v1_0 *)adev->gmc.fw->data;
 318         amdgpu_ucode_print_mc_hdr(&hdr->header);
 319
 320         adev->gmc.fw_version = le32_to_cpu(hdr->header.ucode_version);
 321         regs_size = le32_to_cpu(hdr->io_debug_size_bytes) / (4 * 2);
 322         io_mc_regs = (const __le32 *)
 323                 (adev->gmc.fw->data + le32_to_cpu(hdr->io_debug_array_offset_bytes));
 324         ucode_size = le32_to_cpu(hdr->header.ucode_size_bytes) / 4;
 325         fw_data = (const __le32 *)
 326                 (adev->gmc.fw->data + le32_to_cpu(hdr->header.ucode_array_offset_bytes));
 327
 328         running = REG_GET_FIELD(RREG32(mmMC_SEQ_SUP_CNTL), MC_SEQ_SUP_CNTL, RUN);
 329
 330         if (running == 0) {
 331                 /* reset the engine and set to writable */
 332                 WREG32(mmMC_SEQ_SUP_CNTL, 0x00000008);
 333                 WREG32(mmMC_SEQ_SUP_CNTL, 0x00000010);
 334
 335                 /* load mc io regs */
 336                 for (i = 0; i < regs_size; i++) {
 337                         WREG32(mmMC_SEQ_IO_DEBUG_INDEX, le32_to_cpup(io_mc_regs++));
 338                         WREG32(mmMC_SEQ_IO_DEBUG_DATA, le32_to_cpup(io_mc_regs++));
 339                 }
 340                 /* load the MC ucode */
 341                 for (i = 0; i < ucode_size; i++)
 342                         WREG32(mmMC_SEQ_SUP_PGM, le32_to_cpup(fw_data++));
 343
 344                 /* put the engine back into the active state */
 345                 WREG32(mmMC_SEQ_SUP_CNTL, 0x00000008);
 346                 WREG32(mmMC_SEQ_SUP_CNTL, 0x00000004);
 347                 WREG32(mmMC_SEQ_SUP_CNTL, 0x00000001);
 348
 349                 /* wait for training to complete */
 350                 for (i = 0; i < adev->usec_timeout; i++) {
 351                         if (REG_GET_FIELD(RREG32(mmMC_SEQ_TRAIN_WAKEUP_CNTL),
 352                                           MC_SEQ_TRAIN_WAKEUP_CNTL, TRAIN_DONE_D0))
 353                                 break;
 354                         udelay(1);
 355                 }
 356                 for (i = 0; i < adev->usec_timeout; i++) {
 357                         if (REG_GET_FIELD(RREG32(mmMC_SEQ_TRAIN_WAKEUP_CNTL),
 358                                           MC_SEQ_TRAIN_WAKEUP_CNTL, TRAIN_DONE_D1))
 359                                 break;
 360                         udelay(1);
 361                 }
 362         }
 363
 364         return 0;
 365 }
 366
 367 static int gmc_v8_0_polaris_mc_load_microcode(struct amdgpu_device *adev)
 368 {
 369         const struct mc_firmware_header_v1_0 *hdr;
 370         const __le32 *fw_data = NULL;
 371         const __le32 *io_mc_regs = NULL;
 372         u32 data;
 373         int i, ucode_size, regs_size;
 374
 375         /* Skip MC ucode loading on SR-IOV capable boards.
 376          * vbios does this for us in asic_init in that case.
 377          * Skip MC ucode loading on VF, because hypervisor will do that
 378          * for this adaptor.
 379          */
 380         if (amdgpu_sriov_bios(adev))
 381                 return 0;
 382
 383         if (!adev->gmc.fw)
 384                 return -EINVAL;
 385
 386         hdr = (const struct mc_firmware_header_v1_0 *)adev->gmc.fw->data;
 387         amdgpu_ucode_print_mc_hdr(&hdr->header);
 388
 389         adev->gmc.fw_version = le32_to_cpu(hdr->header.ucode_version);
 390         regs_size = le32_to_cpu(hdr->io_debug_size_bytes) / (4 * 2);
 391         io_mc_regs = (const __le32 *)
 392                 (adev->gmc.fw->data + le32_to_cpu(hdr->io_debug_array_offset_bytes));
 393         ucode_size = le32_to_cpu(hdr->header.ucode_size_bytes) / 4;
 394         fw_data = (const __le32 *)
 395                 (adev->gmc.fw->data + le32_to_cpu(hdr->header.ucode_array_offset_bytes));
 396
 397         data = RREG32(mmMC_SEQ_MISC0);
 398         data &= ~(0x40);
 399         WREG32(mmMC_SEQ_MISC0, data);
 400
 401         /* load mc io regs */
 402         for (i = 0; i < regs_size; i++) {
 403                 WREG32(mmMC_SEQ_IO_DEBUG_INDEX, le32_to_cpup(io_mc_regs++));
 404                 WREG32(mmMC_SEQ_IO_DEBUG_DATA, le32_to_cpup(io_mc_regs++));
 405         }
 406
 407         WREG32(mmMC_SEQ_SUP_CNTL, 0x00000008);
 408         WREG32(mmMC_SEQ_SUP_CNTL, 0x00000010);
 409
 410         /* load the MC ucode */
 411         for (i = 0; i < ucode_size; i++)
 412                 WREG32(mmMC_SEQ_SUP_PGM, le32_to_cpup(fw_data++));
 413
 414         /* put the engine back into the active state */
 415         WREG32(mmMC_SEQ_SUP_CNTL, 0x00000008);
 416         WREG32(mmMC_SEQ_SUP_CNTL, 0x00000004);
 417         WREG32(mmMC_SEQ_SUP_CNTL, 0x00000001);
 418
 419         /* wait for training to complete */
 420         for (i = 0; i < adev->usec_timeout; i++) {
 421                 data = RREG32(mmMC_SEQ_MISC0);
 422                 if (data & 0x80)
 423                         break;
 424                 udelay(1);
 425         }
 426
 427         return 0;
 428 }
 429
 430 static void gmc_v8_0_vram_gtt_location(struct amdgpu_device *adev,
 431                                        struct amdgpu_gmc *mc)
 432 {
 433         u64 base = 0;
 434
 435         if (!amdgpu_sriov_vf(adev))
 436                 base = RREG32(mmMC_VM_FB_LOCATION) & 0xFFFF;
 437         base <<= 24;
 438
 439         amdgpu_gmc_vram_location(adev, mc, base);
 440         amdgpu_gmc_gart_location(adev, mc);
 441 }
 442
 443 /**
 444  * gmc_v8_0_mc_program - program the GPU memory controller
 445  *
 446  * @adev: amdgpu_device pointer
 447  *
 448  * Set the location of vram, gart, and AGP in the GPU's
 449  * physical address space (VI).
 450  */
 451 static void gmc_v8_0_mc_program(struct amdgpu_device *adev)
 452 {
 453         u32 tmp;
 454         int i, j;
 455
 456         /* Initialize HDP */
 457         for (i = 0, j = 0; i < 32; i++, j += 0x6) {
 458                 WREG32((0xb05 + j), 0x00000000);
 459                 WREG32((0xb06 + j), 0x00000000);
 460                 WREG32((0xb07 + j), 0x00000000);
 461                 WREG32((0xb08 + j), 0x00000000);
 462                 WREG32((0xb09 + j), 0x00000000);
 463         }
 464         WREG32(mmHDP_REG_COHERENCY_FLUSH_CNTL, 0);
 465
 466         if (gmc_v8_0_wait_for_idle((void *)adev)) {
 467                 dev_warn(adev->dev, "Wait for MC idle timedout !\n");
 468         }
 469         if (adev->mode_info.num_crtc) {
 470                 /* Lockout access through VGA aperture*/
 471                 tmp = RREG32(mmVGA_HDP_CONTROL);
 472                 tmp = REG_SET_FIELD(tmp, VGA_HDP_CONTROL, VGA_MEMORY_DISABLE, 1);
 473                 WREG32(mmVGA_HDP_CONTROL, tmp);
 474
 475                 /* disable VGA render */
 476                 tmp = RREG32(mmVGA_RENDER_CONTROL);
 477                 tmp = REG_SET_FIELD(tmp, VGA_RENDER_CONTROL, VGA_VSTATUS_CNTL, 0);
 478                 WREG32(mmVGA_RENDER_CONTROL, tmp);
 479         }
 480         /* Update configuration */
 481         WREG32(mmMC_VM_SYSTEM_APERTURE_LOW_ADDR,
 482                adev->gmc.vram_start >> 12);
 483         WREG32(mmMC_VM_SYSTEM_APERTURE_HIGH_ADDR,
 484                adev->gmc.vram_end >> 12);
 485         WREG32(mmMC_VM_SYSTEM_APERTURE_DEFAULT_ADDR,
 486                adev->vram_scratch.gpu_addr >> 12);
 487
 488         if (amdgpu_sriov_vf(adev)) {
 489                 tmp = ((adev->gmc.vram_end >> 24) & 0xFFFF) << 16;
 490                 tmp |= ((adev->gmc.vram_start >> 24) & 0xFFFF);
 491                 WREG32(mmMC_VM_FB_LOCATION, tmp);
 492                 /* XXX double check these! */
 493                 WREG32(mmHDP_NONSURFACE_BASE, (adev->gmc.vram_start >> 8));
 494                 WREG32(mmHDP_NONSURFACE_INFO, (2 << 7) | (1 << 30));
 495                 WREG32(mmHDP_NONSURFACE_SIZE, 0x3FFFFFFF);
 496         }
 497
 498         WREG32(mmMC_VM_AGP_BASE, 0);
 499         WREG32(mmMC_VM_AGP_TOP, 0x0FFFFFFF);
 500         WREG32(mmMC_VM_AGP_BOT, 0x0FFFFFFF);
 501         if (gmc_v8_0_wait_for_idle((void *)adev)) {
 502                 dev_warn(adev->dev, "Wait for MC idle timedout !\n");
 503         }
 504
 505         WREG32(mmBIF_FB_EN, BIF_FB_EN__FB_READ_EN_MASK | BIF_FB_EN__FB_WRITE_EN_MASK);
 506
 507         tmp = RREG32(mmHDP_MISC_CNTL);
 508         tmp = REG_SET_FIELD(tmp, HDP_MISC_CNTL, FLUSH_INVALIDATE_CACHE, 0);
 509         WREG32(mmHDP_MISC_CNTL, tmp);
 510
 511         tmp = RREG32(mmHDP_HOST_PATH_CNTL);
 512         WREG32(mmHDP_HOST_PATH_CNTL, tmp);
 513 }
 514
 515 /**
 516  * gmc_v8_0_mc_init - initialize the memory controller driver params
 517  *
 518  * @adev: amdgpu_device pointer
 519  *
 520  * Look up the amount of vram, vram width, and decide how to place
 521  * vram and gart within the GPU's physical address space (VI).
 522  * Returns 0 for success.
 523  */
 524 static int gmc_v8_0_mc_init(struct amdgpu_device *adev)
 525 {
 526         int r;
 527
 528         adev->gmc.vram_width = amdgpu_atombios_get_vram_width(adev);
 529         if (!adev->gmc.vram_width) {
 530                 u32 tmp;
 531                 int chansize, numchan;
 532
 533                 /* Get VRAM informations */
 534                 tmp = RREG32(mmMC_ARB_RAMCFG);
 535                 if (REG_GET_FIELD(tmp, MC_ARB_RAMCFG, CHANSIZE)) {
 536                         chansize = 64;
 537                 } else {
 538                         chansize = 32;
 539                 }
 540                 tmp = RREG32(mmMC_SHARED_CHMAP);
 541                 switch (REG_GET_FIELD(tmp, MC_SHARED_CHMAP, NOOFCHAN)) {
 542                 case 0:
 543                 default:
 544                         numchan = 1;
 545                         break;
 546                 case 1:
 547                         numchan = 2;
 548                         break;
 549                 case 2:
 550                         numchan = 4;
 551                         break;
 552                 case 3:
 553                         numchan = 8;
 554                         break;
 555                 case 4:
 556                         numchan = 3;
 557                         break;
 558                 case 5:
 559                         numchan = 6;
 560                         break;
 561                 case 6:
 562                         numchan = 10;
 563                         break;
 564                 case 7:
 565                         numchan = 12;
 566                         break;
 567                 case 8:
 568                         numchan = 16;
 569                         break;
 570                 }
 571                 adev->gmc.vram_width = numchan * chansize;
 572         }
 573         /* size in MB on si */
 574         adev->gmc.mc_vram_size = RREG32(mmCONFIG_MEMSIZE) * 1024ULL * 1024ULL;
 575         adev->gmc.real_vram_size = RREG32(mmCONFIG_MEMSIZE) * 1024ULL * 1024ULL;
 576
 577         if (!(adev->flags & AMD_IS_APU)) {
 578                 r = amdgpu_device_resize_fb_bar(adev);
 579                 if (r)
 580                         return r;
 581         }
 582         adev->gmc.aper_base = pci_resource_start(adev->pdev, 0);
 583         adev->gmc.aper_size = pci_resource_len(adev->pdev, 0);
 584
 585 #ifdef CONFIG_X86_64
 586         if (adev->flags & AMD_IS_APU) {
 587                 adev->gmc.aper_base = ((u64)RREG32(mmMC_VM_FB_OFFSET)) << 22;
 588                 adev->gmc.aper_size = adev->gmc.real_vram_size;
 589         }
 590 #endif
 591
 592         /* In case the PCI BAR is larger than the actual amount of vram */
 593         adev->gmc.visible_vram_size = adev->gmc.aper_size;
 594         if (adev->gmc.visible_vram_size > adev->gmc.real_vram_size)
 595                 adev->gmc.visible_vram_size = adev->gmc.real_vram_size;
 596
 597         /* set the gart size */
 598         if (amdgpu_gart_size == -1) {
 599                 switch (adev->asic_type) {
 600                 case CHIP_POLARIS10: /* all engines support GPUVM */
 601                 case CHIP_POLARIS11: /* all engines support GPUVM */
 602                 case CHIP_POLARIS12: /* all engines support GPUVM */
 603                 case CHIP_VEGAM:     /* all engines support GPUVM */
 604                 default:
 605                         adev->gmc.gart_size = 256ULL << 20;
 606                         break;
 607                 case CHIP_TONGA:   /* UVD, VCE do not support GPUVM */
 608                 case CHIP_FIJI:    /* UVD, VCE do not support GPUVM */
 609                 case CHIP_CARRIZO: /* UVD, VCE do not support GPUVM, DCE SG support */
 610                 case CHIP_STONEY:  /* UVD does not support GPUVM, DCE SG support */
 611                         adev->gmc.gart_size = 1024ULL << 20;
 612                         break;
 613                 }
 614         } else {
 615                 adev->gmc.gart_size = (u64)amdgpu_gart_size << 20;
 616         }
 617
 618         gmc_v8_0_vram_gtt_location(adev, &adev->gmc);
 619
 620         return 0;
 621 }
 622
 623 /*
 624  * GART
 625  * VMID 0 is the physical GPU addresses as used by the kernel.
 626  * VMIDs 1-15 are used for userspace clients and are handled
 627  * by the amdgpu vm/hsa code.
 628  */
 629
 630 /**
 631  * gmc_v8_0_flush_gpu_tlb - gart tlb flush callback
 632  *
 633  * @adev: amdgpu_device pointer
 634  * @vmid: vm instance to flush
 635  *
 636  * Flush the TLB for the requested page table (VI).
 637  */
 638 static void gmc_v8_0_flush_gpu_tlb(struct amdgpu_device *adev,
 639                                 uint32_t vmid, uint32_t flush_type)
 640 {
 641         /* bits 0-15 are the VM contexts0-15 */
 642         WREG32(mmVM_INVALIDATE_REQUEST, 1 << vmid);
 643 }
 644
 645 static uint64_t gmc_v8_0_emit_flush_gpu_tlb(struct amdgpu_ring *ring,
 646                                             unsigned vmid, uint64_t pd_addr)
 647 {
 648         uint32_t reg;
 649
 650         if (vmid < 8)
 651                 reg = mmVM_CONTEXT0_PAGE_TABLE_BASE_ADDR + vmid;
 652         else
 653                 reg = mmVM_CONTEXT8_PAGE_TABLE_BASE_ADDR + vmid - 8;
 654         amdgpu_ring_emit_wreg(ring, reg, pd_addr >> 12);
 655
 656         /* bits 0-15 are the VM contexts0-15 */
 657         amdgpu_ring_emit_wreg(ring, mmVM_INVALIDATE_REQUEST, 1 << vmid);
 658
 659         return pd_addr;
 660 }
 661
 662 static void gmc_v8_0_emit_pasid_mapping(struct amdgpu_ring *ring, unsigned vmid,
 663                                         unsigned pasid)
 664 {
 665         amdgpu_ring_emit_wreg(ring, mmIH_VMID_0_LUT + vmid, pasid);
 666 }
 667
 668 /*
 669  * PTE format on VI:
 670  * 63:40 reserved
 671  * 39:12 4k physical page base address
 672  * 11:7 fragment
 673  * 6 write
 674  * 5 read
 675  * 4 exe
 676  * 3 reserved
 677  * 2 snooped
 678  * 1 system
 679  * 0 valid
 680  *
 681  * PDE format on VI:
 682  * 63:59 block fragment size
 683  * 58:40 reserved
 684  * 39:1 physical base address of PTE
 685  * bits 5:1 must be 0.
 686  * 0 valid
 687  */
 688
 689 static uint64_t gmc_v8_0_get_vm_pte_flags(struct amdgpu_device *adev,
 690                                           uint32_t flags)
 691 {
 692         uint64_t pte_flag = 0;
 693
 694         if (flags & AMDGPU_VM_PAGE_EXECUTABLE)
 695                 pte_flag |= AMDGPU_PTE_EXECUTABLE;
 696         if (flags & AMDGPU_VM_PAGE_READABLE)
 697                 pte_flag |= AMDGPU_PTE_READABLE;
 698         if (flags & AMDGPU_VM_PAGE_WRITEABLE)
 699                 pte_flag |= AMDGPU_PTE_WRITEABLE;
 700         if (flags & AMDGPU_VM_PAGE_PRT)
 701                 pte_flag |= AMDGPU_PTE_PRT;
 702
 703         return pte_flag;
 704 }
 705
 706 static void gmc_v8_0_get_vm_pde(struct amdgpu_device *adev, int level,
 707                                 uint64_t *addr, uint64_t *flags)
 708 {
 709         BUG_ON(*addr & 0xFFFFFF0000000FFFULL);
 710 }
 711
 712 /**
 713  * gmc_v8_0_set_fault_enable_default - update VM fault handling
 714  *
 715  * @adev: amdgpu_device pointer
 716  * @value: true redirects VM faults to the default page
 717  */
 718 static void gmc_v8_0_set_fault_enable_default(struct amdgpu_device *adev,
 719                                               bool value)
 720 {
 721         u32 tmp;
 722
 723         tmp = RREG32(mmVM_CONTEXT1_CNTL);
 724         tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
 725                             RANGE_PROTECTION_FAULT_ENABLE_DEFAULT, value);
 726         tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
 727                             DUMMY_PAGE_PROTECTION_FAULT_ENABLE_DEFAULT, value);
 728         tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
 729                             PDE0_PROTECTION_FAULT_ENABLE_DEFAULT, value);
 730         tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
 731                             VALID_PROTECTION_FAULT_ENABLE_DEFAULT, value);
 732         tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
 733                             READ_PROTECTION_FAULT_ENABLE_DEFAULT, value);
 734         tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
 735                             WRITE_PROTECTION_FAULT_ENABLE_DEFAULT, value);
 736         tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
 737                             EXECUTE_PROTECTION_FAULT_ENABLE_DEFAULT, value);
 738         WREG32(mmVM_CONTEXT1_CNTL, tmp);
 739 }
 740
 741 /**
 742  * gmc_v8_0_set_prt - set PRT VM fault
 743  *
 744  * @adev: amdgpu_device pointer
 745  * @enable: enable/disable VM fault handling for PRT
 746 */
 747 static void gmc_v8_0_set_prt(struct amdgpu_device *adev, bool enable)
 748 {
 749         u32 tmp;
 750
 751         if (enable && !adev->gmc.prt_warning) {
 752                 dev_warn(adev->dev, "Disabling VM faults because of PRT request!\n");
 753                 adev->gmc.prt_warning = true;
 754         }
 755
 756         tmp = RREG32(mmVM_PRT_CNTL);
 757         tmp = REG_SET_FIELD(tmp, VM_PRT_CNTL,
 758                             CB_DISABLE_READ_FAULT_ON_UNMAPPED_ACCESS, enable);
 759         tmp = REG_SET_FIELD(tmp, VM_PRT_CNTL,
 760                             CB_DISABLE_WRITE_FAULT_ON_UNMAPPED_ACCESS, enable);
 761         tmp = REG_SET_FIELD(tmp, VM_PRT_CNTL,
 762                             TC_DISABLE_READ_FAULT_ON_UNMAPPED_ACCESS, enable);
 763         tmp = REG_SET_FIELD(tmp, VM_PRT_CNTL,
 764                             TC_DISABLE_WRITE_FAULT_ON_UNMAPPED_ACCESS, enable);
 765         tmp = REG_SET_FIELD(tmp, VM_PRT_CNTL,
 766                             L2_CACHE_STORE_INVALID_ENTRIES, enable);
 767         tmp = REG_SET_FIELD(tmp, VM_PRT_CNTL,
 768                             L1_TLB_STORE_INVALID_ENTRIES, enable);
 769         tmp = REG_SET_FIELD(tmp, VM_PRT_CNTL,
 770                             MASK_PDE0_FAULT, enable);
 771         WREG32(mmVM_PRT_CNTL, tmp);
 772
 773         if (enable) {
 774                 uint32_t low = AMDGPU_VA_RESERVED_SIZE >> AMDGPU_GPU_PAGE_SHIFT;
 775                 uint32_t high = adev->vm_manager.max_pfn -
 776                         (AMDGPU_VA_RESERVED_SIZE >> AMDGPU_GPU_PAGE_SHIFT);
 777
 778                 WREG32(mmVM_PRT_APERTURE0_LOW_ADDR, low);
 779                 WREG32(mmVM_PRT_APERTURE1_LOW_ADDR, low);
 780                 WREG32(mmVM_PRT_APERTURE2_LOW_ADDR, low);
 781                 WREG32(mmVM_PRT_APERTURE3_LOW_ADDR, low);
 782                 WREG32(mmVM_PRT_APERTURE0_HIGH_ADDR, high);
 783                 WREG32(mmVM_PRT_APERTURE1_HIGH_ADDR, high);
 784                 WREG32(mmVM_PRT_APERTURE2_HIGH_ADDR, high);
 785                 WREG32(mmVM_PRT_APERTURE3_HIGH_ADDR, high);
 786         } else {
 787                 WREG32(mmVM_PRT_APERTURE0_LOW_ADDR, 0xfffffff);
 788                 WREG32(mmVM_PRT_APERTURE1_LOW_ADDR, 0xfffffff);
 789                 WREG32(mmVM_PRT_APERTURE2_LOW_ADDR, 0xfffffff);
 790                 WREG32(mmVM_PRT_APERTURE3_LOW_ADDR, 0xfffffff);
 791                 WREG32(mmVM_PRT_APERTURE0_HIGH_ADDR, 0x0);
 792                 WREG32(mmVM_PRT_APERTURE1_HIGH_ADDR, 0x0);
 793                 WREG32(mmVM_PRT_APERTURE2_HIGH_ADDR, 0x0);
 794                 WREG32(mmVM_PRT_APERTURE3_HIGH_ADDR, 0x0);
 795         }
 796 }
 797
 798 /**
 799  * gmc_v8_0_gart_enable - gart enable
 800  *
 801  * @adev: amdgpu_device pointer
 802  *
 803  * This sets up the TLBs, programs the page tables for VMID0,
 804  * sets up the hw for VMIDs 1-15 which are allocated on
 805  * demand, and sets up the global locations for the LDS, GDS,
 806  * and GPUVM for FSA64 clients (VI).
 807  * Returns 0 for success, errors for failure.
 808  */
 809 static int gmc_v8_0_gart_enable(struct amdgpu_device *adev)
 810 {
 811         uint64_t table_addr;
 812         int r, i;
 813         u32 tmp, field;
 814
 815         if (adev->gart.bo == NULL) {
 816                 dev_err(adev->dev, "No VRAM object for PCIE GART.\n");
 817                 return -EINVAL;
 818         }
 819         r = amdgpu_gart_table_vram_pin(adev);
 820         if (r)
 821                 return r;
 822
 823         table_addr = amdgpu_bo_gpu_offset(adev->gart.bo);
 824
 825         /* Setup TLB control */
 826         tmp = RREG32(mmMC_VM_MX_L1_TLB_CNTL);
 827         tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL, ENABLE_L1_TLB, 1);
 828         tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL, ENABLE_L1_FRAGMENT_PROCESSING, 1);
 829         tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL, SYSTEM_ACCESS_MODE, 3);
 830         tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL, ENABLE_ADVANCED_DRIVER_MODEL, 1);
 831         tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL, SYSTEM_APERTURE_UNMAPPED_ACCESS, 0);
 832         WREG32(mmMC_VM_MX_L1_TLB_CNTL, tmp);
 833         /* Setup L2 cache */
 834         tmp = RREG32(mmVM_L2_CNTL);
 835         tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, ENABLE_L2_CACHE, 1);
 836         tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, ENABLE_L2_FRAGMENT_PROCESSING, 1);
 837         tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, ENABLE_L2_PTE_CACHE_LRU_UPDATE_BY_WRITE, 1);
 838         tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, ENABLE_L2_PDE0_CACHE_LRU_UPDATE_BY_WRITE, 1);
 839         tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, EFFECTIVE_L2_QUEUE_SIZE, 7);
 840         tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, CONTEXT1_IDENTITY_ACCESS_MODE, 1);
 841         tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, ENABLE_DEFAULT_PAGE_OUT_TO_SYSTEM_MEMORY, 1);
 842         WREG32(mmVM_L2_CNTL, tmp);
 843         tmp = RREG32(mmVM_L2_CNTL2);
 844         tmp = REG_SET_FIELD(tmp, VM_L2_CNTL2, INVALIDATE_ALL_L1_TLBS, 1);
 845         tmp = REG_SET_FIELD(tmp, VM_L2_CNTL2, INVALIDATE_L2_CACHE, 1);
 846         WREG32(mmVM_L2_CNTL2, tmp);
 847
 848         field = adev->vm_manager.fragment_size;
 849         tmp = RREG32(mmVM_L2_CNTL3);
 850         tmp = REG_SET_FIELD(tmp, VM_L2_CNTL3, L2_CACHE_BIGK_ASSOCIATIVITY, 1);
 851         tmp = REG_SET_FIELD(tmp, VM_L2_CNTL3, BANK_SELECT, field);
 852         tmp = REG_SET_FIELD(tmp, VM_L2_CNTL3, L2_CACHE_BIGK_FRAGMENT_SIZE, field);
 853         WREG32(mmVM_L2_CNTL3, tmp);
 854         /* XXX: set to enable PTE/PDE in system memory */
 855         tmp = RREG32(mmVM_L2_CNTL4);
 856         tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_CONTEXT0_PDE_REQUEST_PHYSICAL, 0);
 857         tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_CONTEXT0_PDE_REQUEST_SHARED, 0);
 858         tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_CONTEXT0_PDE_REQUEST_SNOOP, 0);
 859         tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_CONTEXT0_PTE_REQUEST_PHYSICAL, 0);
 860         tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_CONTEXT0_PTE_REQUEST_SHARED, 0);
 861         tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_CONTEXT0_PTE_REQUEST_SNOOP, 0);
 862         tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_CONTEXT1_PDE_REQUEST_PHYSICAL, 0);
 863         tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_CONTEXT1_PDE_REQUEST_SHARED, 0);
 864         tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_CONTEXT1_PDE_REQUEST_SNOOP, 0);
 865         tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_CONTEXT1_PTE_REQUEST_PHYSICAL, 0);
 866         tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_CONTEXT1_PTE_REQUEST_SHARED, 0);
 867         tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_CONTEXT1_PTE_REQUEST_SNOOP, 0);
 868         WREG32(mmVM_L2_CNTL4, tmp);
 869         /* setup context0 */
 870         WREG32(mmVM_CONTEXT0_PAGE_TABLE_START_ADDR, adev->gmc.gart_start >> 12);
 871         WREG32(mmVM_CONTEXT0_PAGE_TABLE_END_ADDR, adev->gmc.gart_end >> 12);
 872         WREG32(mmVM_CONTEXT0_PAGE_TABLE_BASE_ADDR, table_addr >> 12);
 873         WREG32(mmVM_CONTEXT0_PROTECTION_FAULT_DEFAULT_ADDR,
 874                         (u32)(adev->dummy_page_addr >> 12));
 875         WREG32(mmVM_CONTEXT0_CNTL2, 0);
 876         tmp = RREG32(mmVM_CONTEXT0_CNTL);
 877         tmp = REG_SET_FIELD(tmp, VM_CONTEXT0_CNTL, ENABLE_CONTEXT, 1);
 878         tmp = REG_SET_FIELD(tmp, VM_CONTEXT0_CNTL, PAGE_TABLE_DEPTH, 0);
 879         tmp = REG_SET_FIELD(tmp, VM_CONTEXT0_CNTL, RANGE_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
 880         WREG32(mmVM_CONTEXT0_CNTL, tmp);
 881
 882         WREG32(mmVM_L2_CONTEXT1_IDENTITY_APERTURE_LOW_ADDR, 0);
 883         WREG32(mmVM_L2_CONTEXT1_IDENTITY_APERTURE_HIGH_ADDR, 0);
 884         WREG32(mmVM_L2_CONTEXT_IDENTITY_PHYSICAL_OFFSET, 0);
 885
 886         /* empty context1-15 */
 887         /* FIXME start with 4G, once using 2 level pt switch to full
 888          * vm size space
 889          */
 890         /* set vm size, must be a multiple of 4 */
 891         WREG32(mmVM_CONTEXT1_PAGE_TABLE_START_ADDR, 0);
 892         WREG32(mmVM_CONTEXT1_PAGE_TABLE_END_ADDR, adev->vm_manager.max_pfn - 1);
 893         for (i = 1; i < 16; i++) {
 894                 if (i < 8)
 895                         WREG32(mmVM_CONTEXT0_PAGE_TABLE_BASE_ADDR + i,
 896                                table_addr >> 12);
 897                 else
 898                         WREG32(mmVM_CONTEXT8_PAGE_TABLE_BASE_ADDR + i - 8,
 899                                table_addr >> 12);
 900         }
 901
 902         /* enable context1-15 */
 903         WREG32(mmVM_CONTEXT1_PROTECTION_FAULT_DEFAULT_ADDR,
 904                (u32)(adev->dummy_page_addr >> 12));
 905         WREG32(mmVM_CONTEXT1_CNTL2, 4);
 906         tmp = RREG32(mmVM_CONTEXT1_CNTL);
 907         tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, ENABLE_CONTEXT, 1);
 908         tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, PAGE_TABLE_DEPTH, 1);
 909         tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, RANGE_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
 910         tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, DUMMY_PAGE_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
 911         tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, PDE0_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
 912         tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, VALID_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
 913         tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, READ_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
 914         tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, WRITE_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
 915         tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, EXECUTE_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
 916         tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, PAGE_TABLE_BLOCK_SIZE,
 917                             adev->vm_manager.block_size - 9);
 918         WREG32(mmVM_CONTEXT1_CNTL, tmp);
 919         if (amdgpu_vm_fault_stop == AMDGPU_VM_FAULT_STOP_ALWAYS)
 920                 gmc_v8_0_set_fault_enable_default(adev, false);
 921         else
 922                 gmc_v8_0_set_fault_enable_default(adev, true);
 923
 924         gmc_v8_0_flush_gpu_tlb(adev, 0, 0);
 925         DRM_INFO("PCIE GART of %uM enabled (table at 0x%016llX).\n",
 926                  (unsigned)(adev->gmc.gart_size >> 20),
 927                  (unsigned long long)table_addr);
 928         adev->gart.ready = true;
 929         return 0;
 930 }
 931
 932 static int gmc_v8_0_gart_init(struct amdgpu_device *adev)
 933 {
 934         int r;
 935
 936         if (adev->gart.bo) {
 937                 WARN(1, "R600 PCIE GART already initialized\n");
 938                 return 0;
 939         }
 940         /* Initialize common gart structure */
 941         r = amdgpu_gart_init(adev);
 942         if (r)
 943                 return r;
 944         adev->gart.table_size = adev->gart.num_gpu_pages * 8;
 945         adev->gart.gart_pte_flags = AMDGPU_PTE_EXECUTABLE;
 946         return amdgpu_gart_table_vram_alloc(adev);
 947 }
 948
 949 /**
 950  * gmc_v8_0_gart_disable - gart disable
 951  *
 952  * @adev: amdgpu_device pointer
 953  *
 954  * This disables all VM page table (VI).
 955  */
 956 static void gmc_v8_0_gart_disable(struct amdgpu_device *adev)
 957 {
 958         u32 tmp;
 959
 960         /* Disable all tables */
 961         WREG32(mmVM_CONTEXT0_CNTL, 0);
 962         WREG32(mmVM_CONTEXT1_CNTL, 0);
 963         /* Setup TLB control */
 964         tmp = RREG32(mmMC_VM_MX_L1_TLB_CNTL);
 965         tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL, ENABLE_L1_TLB, 0);
 966         tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL, ENABLE_L1_FRAGMENT_PROCESSING, 0);
 967         tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL, ENABLE_ADVANCED_DRIVER_MODEL, 0);
 968         WREG32(mmMC_VM_MX_L1_TLB_CNTL, tmp);
 969         /* Setup L2 cache */
 970         tmp = RREG32(mmVM_L2_CNTL);
 971         tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, ENABLE_L2_CACHE, 0);
 972         WREG32(mmVM_L2_CNTL, tmp);
 973         WREG32(mmVM_L2_CNTL2, 0);
 974         amdgpu_gart_table_vram_unpin(adev);
 975 }
 976
 977 /**
 978  * gmc_v8_0_vm_decode_fault - print human readable fault info
 979  *
 980  * @adev: amdgpu_device pointer
 981  * @status: VM_CONTEXT1_PROTECTION_FAULT_STATUS register value
 982  * @addr: VM_CONTEXT1_PROTECTION_FAULT_ADDR register value
 983  *
 984  * Print human readable fault information (VI).
 985  */
 986 static void gmc_v8_0_vm_decode_fault(struct amdgpu_device *adev, u32 status,
 987                                      u32 addr, u32 mc_client, unsigned pasid)
 988 {
 989         u32 vmid = REG_GET_FIELD(status, VM_CONTEXT1_PROTECTION_FAULT_STATUS, VMID);
 990         u32 protections = REG_GET_FIELD(status, VM_CONTEXT1_PROTECTION_FAULT_STATUS,
 991                                         PROTECTIONS);
 992         char block[5] = { mc_client >> 24, (mc_client >> 16) & 0xff,
 993                 (mc_client >> 8) & 0xff, mc_client & 0xff, 0 };
 994         u32 mc_id;
 995
 996         mc_id = REG_GET_FIELD(status, VM_CONTEXT1_PROTECTION_FAULT_STATUS,
 997                               MEMORY_CLIENT_ID);
 998
 999         dev_err(adev->dev, "VM fault (0x%02x, vmid %d, pasid %d) at page %u, %s from '%s' (0x%08x) (%d)\n",
1000                protections, vmid, pasid, addr,
1001                REG_GET_FIELD(status, VM_CONTEXT1_PROTECTION_FAULT_STATUS,
1002                              MEMORY_CLIENT_RW) ?
1003                "write" : "read", block, mc_client, mc_id);
1004 }
1005
1006 static int gmc_v8_0_convert_vram_type(int mc_seq_vram_type)
1007 {
1008         switch (mc_seq_vram_type) {
1009         case MC_SEQ_MISC0__MT__GDDR1:
1010                 return AMDGPU_VRAM_TYPE_GDDR1;
1011         case MC_SEQ_MISC0__MT__DDR2:
1012                 return AMDGPU_VRAM_TYPE_DDR2;
1013         case MC_SEQ_MISC0__MT__GDDR3:
1014                 return AMDGPU_VRAM_TYPE_GDDR3;
1015         case MC_SEQ_MISC0__MT__GDDR4:
1016                 return AMDGPU_VRAM_TYPE_GDDR4;
1017         case MC_SEQ_MISC0__MT__GDDR5:
1018                 return AMDGPU_VRAM_TYPE_GDDR5;
1019         case MC_SEQ_MISC0__MT__HBM:
1020                 return AMDGPU_VRAM_TYPE_HBM;
1021         case MC_SEQ_MISC0__MT__DDR3:
1022                 return AMDGPU_VRAM_TYPE_DDR3;
1023         default:
1024                 return AMDGPU_VRAM_TYPE_UNKNOWN;
1025         }
1026 }
1027
1028 static int gmc_v8_0_early_init(void *handle)
1029 {
1030         struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1031
1032         gmc_v8_0_set_gmc_funcs(adev);
1033         gmc_v8_0_set_irq_funcs(adev);
1034
1035         adev->gmc.shared_aperture_start = 0x2000000000000000ULL;
1036         adev->gmc.shared_aperture_end =
1037                 adev->gmc.shared_aperture_start + (4ULL << 30) - 1;
1038         adev->gmc.private_aperture_start =
1039                 adev->gmc.shared_aperture_end + 1;
1040         adev->gmc.private_aperture_end =
1041                 adev->gmc.private_aperture_start + (4ULL << 30) - 1;
1042
1043         return 0;
1044 }
1045
1046 static int gmc_v8_0_late_init(void *handle)
1047 {
1048         struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1049
1050         amdgpu_bo_late_init(adev);
1051
1052         if (amdgpu_vm_fault_stop != AMDGPU_VM_FAULT_STOP_ALWAYS)
1053                 return amdgpu_irq_get(adev, &adev->gmc.vm_fault, 0);
1054         else
1055                 return 0;
1056 }
1057
1058 static unsigned gmc_v8_0_get_vbios_fb_size(struct amdgpu_device *adev)
1059 {
1060         u32 d1vga_control = RREG32(mmD1VGA_CONTROL);
1061         unsigned size;
1062
1063         if (REG_GET_FIELD(d1vga_control, D1VGA_CONTROL, D1VGA_MODE_ENABLE)) {
1064                 size = 9 * 1024 * 1024; /* reserve 8MB for vga emulator and 1 MB for FB */
1065         } else {
1066                 u32 viewport = RREG32(mmVIEWPORT_SIZE);
1067                 size = (REG_GET_FIELD(viewport, VIEWPORT_SIZE, VIEWPORT_HEIGHT) *
1068                         REG_GET_FIELD(viewport, VIEWPORT_SIZE, VIEWPORT_WIDTH) *
1069                         4);
1070         }
1071         /* return 0 if the pre-OS buffer uses up most of vram */
1072         if ((adev->gmc.real_vram_size - size) < (8 * 1024 * 1024))
1073                 return 0;
1074         return size;
1075 }
1076
1077 #define mmMC_SEQ_MISC0_FIJI 0xA71
1078
1079 static int gmc_v8_0_sw_init(void *handle)
1080 {
1081         int r;
1082         int dma_bits;
1083         struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1084
1085         if (adev->flags & AMD_IS_APU) {
1086                 adev->gmc.vram_type = AMDGPU_VRAM_TYPE_UNKNOWN;
1087         } else {
1088                 u32 tmp;
1089
1090                 if ((adev->asic_type == CHIP_FIJI) ||
1091                     (adev->asic_type == CHIP_VEGAM))
1092                         tmp = RREG32(mmMC_SEQ_MISC0_FIJI);
1093                 else
1094                         tmp = RREG32(mmMC_SEQ_MISC0);
1095                 tmp &= MC_SEQ_MISC0__MT__MASK;
1096                 adev->gmc.vram_type = gmc_v8_0_convert_vram_type(tmp);
1097         }
1098
1099         r = amdgpu_irq_add_id(adev, AMDGPU_IRQ_CLIENTID_LEGACY, VISLANDS30_IV_SRCID_GFX_PAGE_INV_FAULT, &adev->gmc.vm_fault);
1100         if (r)
1101                 return r;
1102
1103         r = amdgpu_irq_add_id(adev, AMDGPU_IRQ_CLIENTID_LEGACY, VISLANDS30_IV_SRCID_GFX_MEM_PROT_FAULT, &adev->gmc.vm_fault);
1104         if (r)
1105                 return r;
1106
1107         /* Adjust VM size here.
1108          * Currently set to 4GB ((1 << 20) 4k pages).
1109          * Max GPUVM size for cayman and SI is 40 bits.
1110          */
1111         amdgpu_vm_adjust_size(adev, 64, 9, 1, 40);
1112
1113         /* Set the internal MC address mask
1114          * This is the max address of the GPU's
1115          * internal address space.
1116          */
1117         adev->gmc.mc_mask = 0xffffffffffULL; /* 40 bit MC */
1118
1119         /* set DMA mask + need_dma32 flags.
1120          * PCIE - can handle 40-bits.
1121          * IGP - can handle 40-bits
1122          * PCI - dma32 for legacy pci gart, 40 bits on newer asics
1123          */
1124         adev->need_dma32 = false;
1125         dma_bits = adev->need_dma32 ? 32 : 40;
1126         r = pci_set_dma_mask(adev->pdev, DMA_BIT_MASK(dma_bits));
1127         if (r) {
1128                 adev->need_dma32 = true;
1129                 dma_bits = 32;
1130                 pr_warn("amdgpu: No suitable DMA available\n");
1131         }
1132         r = pci_set_consistent_dma_mask(adev->pdev, DMA_BIT_MASK(dma_bits));
1133         if (r) {
1134                 pci_set_consistent_dma_mask(adev->pdev, DMA_BIT_MASK(32));
1135                 pr_warn("amdgpu: No coherent DMA available\n");
1136         }
1137         adev->need_swiotlb = drm_need_swiotlb(dma_bits);
1138
1139         r = gmc_v8_0_init_microcode(adev);
1140         if (r) {
1141                 DRM_ERROR("Failed to load mc firmware!\n");
1142                 return r;
1143         }
1144
1145         r = gmc_v8_0_mc_init(adev);
1146         if (r)
1147                 return r;
1148
1149         adev->gmc.stolen_size = gmc_v8_0_get_vbios_fb_size(adev);
1150
1151         /* Memory manager */
1152         r = amdgpu_bo_init(adev);
1153         if (r)
1154                 return r;
1155
1156         r = gmc_v8_0_gart_init(adev);
1157         if (r)
1158                 return r;
1159
1160         /*
1161          * number of VMs
1162          * VMID 0 is reserved for System
1163          * amdgpu graphics/compute will use VMIDs 1-7
1164          * amdkfd will use VMIDs 8-15
1165          */
1166         adev->vm_manager.id_mgr[0].num_ids = AMDGPU_NUM_OF_VMIDS;
1167         amdgpu_vm_manager_init(adev);
1168
1169         /* base offset of vram pages */
1170         if (adev->flags & AMD_IS_APU) {
1171                 u64 tmp = RREG32(mmMC_VM_FB_OFFSET);
1172
1173                 tmp <<= 22;
1174                 adev->vm_manager.vram_base_offset = tmp;
1175         } else {
1176                 adev->vm_manager.vram_base_offset = 0;
1177         }
1178
1179         adev->gmc.vm_fault_info = kmalloc(sizeof(struct kfd_vm_fault_info),
1180                                         GFP_KERNEL);
1181         if (!adev->gmc.vm_fault_info)
1182                 return -ENOMEM;
1183         atomic_set(&adev->gmc.vm_fault_info_updated, 0);
1184
1185         return 0;
1186 }
1187
1188 static int gmc_v8_0_sw_fini(void *handle)
1189 {
1190         struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1191
1192         amdgpu_gem_force_release(adev);
1193         amdgpu_vm_manager_fini(adev);
1194         kfree(adev->gmc.vm_fault_info);
1195         amdgpu_gart_table_vram_free(adev);
1196         amdgpu_bo_fini(adev);
1197         amdgpu_gart_fini(adev);
1198         release_firmware(adev->gmc.fw);
1199         adev->gmc.fw = NULL;
1200
1201         return 0;
1202 }
1203
1204 static int gmc_v8_0_hw_init(void *handle)
1205 {
1206         int r;
1207         struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1208
1209         gmc_v8_0_init_golden_registers(adev);
1210
1211         gmc_v8_0_mc_program(adev);
1212
1213         if (adev->asic_type == CHIP_TONGA) {
1214                 r = gmc_v8_0_tonga_mc_load_microcode(adev);
1215                 if (r) {
1216                         DRM_ERROR("Failed to load MC firmware!\n");
1217                         return r;
1218                 }
1219         } else if (adev->asic_type == CHIP_POLARIS11 ||
1220                         adev->asic_type == CHIP_POLARIS10 ||
1221                         adev->asic_type == CHIP_POLARIS12) {
1222                 r = gmc_v8_0_polaris_mc_load_microcode(adev);
1223                 if (r) {
1224                         DRM_ERROR("Failed to load MC firmware!\n");
1225                         return r;
1226                 }
1227         }
1228
1229         r = gmc_v8_0_gart_enable(adev);
1230         if (r)
1231                 return r;
1232
1233         return r;
1234 }
1235
1236 static int gmc_v8_0_hw_fini(void *handle)
1237 {
1238         struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1239
1240         amdgpu_irq_put(adev, &adev->gmc.vm_fault, 0);
1241         gmc_v8_0_gart_disable(adev);
1242
1243         return 0;
1244 }
1245
1246 static int gmc_v8_0_suspend(void *handle)
1247 {
1248         struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1249
1250         gmc_v8_0_hw_fini(adev);
1251
1252         return 0;
1253 }
1254
1255 static int gmc_v8_0_resume(void *handle)
1256 {
1257         int r;
1258         struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1259
1260         r = gmc_v8_0_hw_init(adev);
1261         if (r)
1262                 return r;
1263
1264         amdgpu_vmid_reset_all(adev);
1265
1266         return 0;
1267 }
1268
1269 static bool gmc_v8_0_is_idle(void *handle)
1270 {
1271         struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1272         u32 tmp = RREG32(mmSRBM_STATUS);
1273
1274         if (tmp & (SRBM_STATUS__MCB_BUSY_MASK | SRBM_STATUS__MCB_NON_DISPLAY_BUSY_MASK |
1275                    SRBM_STATUS__MCC_BUSY_MASK | SRBM_STATUS__MCD_BUSY_MASK | SRBM_STATUS__VMC_BUSY_MASK))
1276                 return false;
1277
1278         return true;
1279 }
1280
1281 static int gmc_v8_0_wait_for_idle(void *handle)
1282 {
1283         unsigned i;
1284         u32 tmp;
1285         struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1286
1287         for (i = 0; i < adev->usec_timeout; i++) {
1288                 /* read MC_STATUS */
1289                 tmp = RREG32(mmSRBM_STATUS) & (SRBM_STATUS__MCB_BUSY_MASK |
1290                                                SRBM_STATUS__MCB_NON_DISPLAY_BUSY_MASK |
1291                                                SRBM_STATUS__MCC_BUSY_MASK |
1292                                                SRBM_STATUS__MCD_BUSY_MASK |
1293                                                SRBM_STATUS__VMC_BUSY_MASK |
1294                                                SRBM_STATUS__VMC1_BUSY_MASK);
1295                 if (!tmp)
1296                         return 0;
1297                 udelay(1);
1298         }
1299         return -ETIMEDOUT;
1300
1301 }
1302
1303 static bool gmc_v8_0_check_soft_reset(void *handle)
1304 {
1305         u32 srbm_soft_reset = 0;
1306         struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1307         u32 tmp = RREG32(mmSRBM_STATUS);
1308
1309         if (tmp & SRBM_STATUS__VMC_BUSY_MASK)
1310                 srbm_soft_reset = REG_SET_FIELD(srbm_soft_reset,
1311                                                 SRBM_SOFT_RESET, SOFT_RESET_VMC, 1);
1312
1313         if (tmp & (SRBM_STATUS__MCB_BUSY_MASK | SRBM_STATUS__MCB_NON_DISPLAY_BUSY_MASK |
1314                    SRBM_STATUS__MCC_BUSY_MASK | SRBM_STATUS__MCD_BUSY_MASK)) {
1315                 if (!(adev->flags & AMD_IS_APU))
1316                         srbm_soft_reset = REG_SET_FIELD(srbm_soft_reset,
1317                                                         SRBM_SOFT_RESET, SOFT_RESET_MC, 1);
1318         }
1319         if (srbm_soft_reset) {
1320                 adev->gmc.srbm_soft_reset = srbm_soft_reset;
1321                 return true;
1322         } else {
1323                 adev->gmc.srbm_soft_reset = 0;
1324                 return false;
1325         }
1326 }
1327
1328 static int gmc_v8_0_pre_soft_reset(void *handle)
1329 {
1330         struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1331
1332         if (!adev->gmc.srbm_soft_reset)
1333                 return 0;
1334
1335         gmc_v8_0_mc_stop(adev);
1336         if (gmc_v8_0_wait_for_idle(adev)) {
1337                 dev_warn(adev->dev, "Wait for GMC idle timed out !\n");
1338         }
1339
1340         return 0;
1341 }
1342
1343 static int gmc_v8_0_soft_reset(void *handle)
1344 {
1345         struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1346         u32 srbm_soft_reset;
1347
1348         if (!adev->gmc.srbm_soft_reset)
1349                 return 0;
1350         srbm_soft_reset = adev->gmc.srbm_soft_reset;
1351
1352         if (srbm_soft_reset) {
1353                 u32 tmp;
1354
1355                 tmp = RREG32(mmSRBM_SOFT_RESET);
1356                 tmp |= srbm_soft_reset;
1357                 dev_info(adev->dev, "SRBM_SOFT_RESET=0x%08X\n", tmp);
1358                 WREG32(mmSRBM_SOFT_RESET, tmp);
1359                 tmp = RREG32(mmSRBM_SOFT_RESET);
1360
1361                 udelay(50);
1362
1363                 tmp &= ~srbm_soft_reset;
1364                 WREG32(mmSRBM_SOFT_RESET, tmp);
1365                 tmp = RREG32(mmSRBM_SOFT_RESET);
1366
1367                 /* Wait a little for things to settle down */
1368                 udelay(50);
1369         }
1370
1371         return 0;
1372 }
1373
1374 static int gmc_v8_0_post_soft_reset(void *handle)
1375 {
1376         struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1377
1378         if (!adev->gmc.srbm_soft_reset)
1379                 return 0;
1380
1381         gmc_v8_0_mc_resume(adev);
1382         return 0;
1383 }
1384
1385 static int gmc_v8_0_vm_fault_interrupt_state(struct amdgpu_device *adev,
1386                                              struct amdgpu_irq_src *src,
1387                                              unsigned type,
1388                                              enum amdgpu_interrupt_state state)
1389 {
1390         u32 tmp;
1391         u32 bits = (VM_CONTEXT1_CNTL__RANGE_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
1392                     VM_CONTEXT1_CNTL__DUMMY_PAGE_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
1393                     VM_CONTEXT1_CNTL__PDE0_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
1394                     VM_CONTEXT1_CNTL__VALID_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
1395                     VM_CONTEXT1_CNTL__READ_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
1396                     VM_CONTEXT1_CNTL__WRITE_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
1397                     VM_CONTEXT1_CNTL__EXECUTE_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK);
1398
1399         switch (state) {
1400         case AMDGPU_IRQ_STATE_DISABLE:
1401                 /* system context */
1402                 tmp = RREG32(mmVM_CONTEXT0_CNTL);
1403                 tmp &= ~bits;
1404                 WREG32(mmVM_CONTEXT0_CNTL, tmp);
1405                 /* VMs */
1406                 tmp = RREG32(mmVM_CONTEXT1_CNTL);
1407                 tmp &= ~bits;
1408                 WREG32(mmVM_CONTEXT1_CNTL, tmp);
1409                 break;
1410         case AMDGPU_IRQ_STATE_ENABLE:
1411                 /* system context */
1412                 tmp = RREG32(mmVM_CONTEXT0_CNTL);
1413                 tmp |= bits;
1414                 WREG32(mmVM_CONTEXT0_CNTL, tmp);
1415                 /* VMs */
1416                 tmp = RREG32(mmVM_CONTEXT1_CNTL);
1417                 tmp |= bits;
1418                 WREG32(mmVM_CONTEXT1_CNTL, tmp);
1419                 break;
1420         default:
1421                 break;
1422         }
1423
1424         return 0;
1425 }
1426
1427 static int gmc_v8_0_process_interrupt(struct amdgpu_device *adev,
1428                                       struct amdgpu_irq_src *source,
1429                                       struct amdgpu_iv_entry *entry)
1430 {
1431         u32 addr, status, mc_client, vmid;
1432
1433         if (amdgpu_sriov_vf(adev)) {
1434                 dev_err(adev->dev, "GPU fault detected: %d 0x%08x\n",
1435                         entry->src_id, entry->src_data[0]);
1436                 dev_err(adev->dev, " Can't decode VM fault info here on SRIOV VF\n");
1437                 return 0;
1438         }
1439
1440         addr = RREG32(mmVM_CONTEXT1_PROTECTION_FAULT_ADDR);
1441         status = RREG32(mmVM_CONTEXT1_PROTECTION_FAULT_STATUS);
1442         mc_client = RREG32(mmVM_CONTEXT1_PROTECTION_FAULT_MCCLIENT);
1443         /* reset addr and status */
1444         WREG32_P(mmVM_CONTEXT1_CNTL2, 1, ~1);
1445
1446         if (!addr && !status)
1447                 return 0;
1448
1449         if (amdgpu_vm_fault_stop == AMDGPU_VM_FAULT_STOP_FIRST)
1450                 gmc_v8_0_set_fault_enable_default(adev, false);
1451
1452         if (printk_ratelimit()) {
1453                 struct amdgpu_task_info task_info;
1454
1455                 memset(&task_info, 0, sizeof(struct amdgpu_task_info));
1456                 amdgpu_vm_get_task_info(adev, entry->pasid, &task_info);
1457
1458                 dev_err(adev->dev, "GPU fault detected: %d 0x%08x for process %s pid %d thread %s pid %d\n",
1459                         entry->src_id, entry->src_data[0], task_info.process_name,
1460                         task_info.tgid, task_info.task_name, task_info.pid);
1461                 dev_err(adev->dev, "  VM_CONTEXT1_PROTECTION_FAULT_ADDR   0x%08X\n",
1462                         addr);
1463                 dev_err(adev->dev, "  VM_CONTEXT1_PROTECTION_FAULT_STATUS 0x%08X\n",
1464                         status);
1465                 gmc_v8_0_vm_decode_fault(adev, status, addr, mc_client,
1466                                          entry->pasid);
1467         }
1468
1469         vmid = REG_GET_FIELD(status, VM_CONTEXT1_PROTECTION_FAULT_STATUS,
1470                              VMID);
1471         if (amdgpu_amdkfd_is_kfd_vmid(adev, vmid)
1472                 && !atomic_read(&adev->gmc.vm_fault_info_updated)) {
1473                 struct kfd_vm_fault_info *info = adev->gmc.vm_fault_info;
1474                 u32 protections = REG_GET_FIELD(status,
1475                                         VM_CONTEXT1_PROTECTION_FAULT_STATUS,
1476                                         PROTECTIONS);
1477
1478                 info->vmid = vmid;
1479                 info->mc_id = REG_GET_FIELD(status,
1480                                             VM_CONTEXT1_PROTECTION_FAULT_STATUS,
1481                                             MEMORY_CLIENT_ID);
1482                 info->status = status;
1483                 info->page_addr = addr;
1484                 info->prot_valid = protections & 0x7 ? true : false;
1485                 info->prot_read = protections & 0x8 ? true : false;
1486                 info->prot_write = protections & 0x10 ? true : false;
1487                 info->prot_exec = protections & 0x20 ? true : false;
1488                 mb();
1489                 atomic_set(&adev->gmc.vm_fault_info_updated, 1);
1490         }
1491
1492         return 0;
1493 }
1494
1495 static void fiji_update_mc_medium_grain_clock_gating(struct amdgpu_device *adev,
1496                                                      bool enable)
1497 {
1498         uint32_t data;
1499
1500         if (enable && (adev->cg_flags & AMD_CG_SUPPORT_MC_MGCG)) {
1501                 data = RREG32(mmMC_HUB_MISC_HUB_CG);
1502                 data |= MC_HUB_MISC_HUB_CG__ENABLE_MASK;
1503                 WREG32(mmMC_HUB_MISC_HUB_CG, data);
1504
1505                 data = RREG32(mmMC_HUB_MISC_SIP_CG);
1506                 data |= MC_HUB_MISC_SIP_CG__ENABLE_MASK;
1507                 WREG32(mmMC_HUB_MISC_SIP_CG, data);
1508
1509                 data = RREG32(mmMC_HUB_MISC_VM_CG);
1510                 data |= MC_HUB_MISC_VM_CG__ENABLE_MASK;
1511                 WREG32(mmMC_HUB_MISC_VM_CG, data);
1512
1513                 data = RREG32(mmMC_XPB_CLK_GAT);
1514                 data |= MC_XPB_CLK_GAT__ENABLE_MASK;
1515                 WREG32(mmMC_XPB_CLK_GAT, data);
1516
1517                 data = RREG32(mmATC_MISC_CG);
1518                 data |= ATC_MISC_CG__ENABLE_MASK;
1519                 WREG32(mmATC_MISC_CG, data);
1520
1521                 data = RREG32(mmMC_CITF_MISC_WR_CG);
1522                 data |= MC_CITF_MISC_WR_CG__ENABLE_MASK;
1523                 WREG32(mmMC_CITF_MISC_WR_CG, data);
1524
1525                 data = RREG32(mmMC_CITF_MISC_RD_CG);
1526                 data |= MC_CITF_MISC_RD_CG__ENABLE_MASK;
1527                 WREG32(mmMC_CITF_MISC_RD_CG, data);
1528
1529                 data = RREG32(mmMC_CITF_MISC_VM_CG);
1530                 data |= MC_CITF_MISC_VM_CG__ENABLE_MASK;
1531                 WREG32(mmMC_CITF_MISC_VM_CG, data);
1532
1533                 data = RREG32(mmVM_L2_CG);
1534                 data |= VM_L2_CG__ENABLE_MASK;
1535                 WREG32(mmVM_L2_CG, data);
1536         } else {
1537                 data = RREG32(mmMC_HUB_MISC_HUB_CG);
1538                 data &= ~MC_HUB_MISC_HUB_CG__ENABLE_MASK;
1539                 WREG32(mmMC_HUB_MISC_HUB_CG, data);
1540
1541                 data = RREG32(mmMC_HUB_MISC_SIP_CG);
1542                 data &= ~MC_HUB_MISC_SIP_CG__ENABLE_MASK;
1543                 WREG32(mmMC_HUB_MISC_SIP_CG, data);
1544
1545                 data = RREG32(mmMC_HUB_MISC_VM_CG);
1546                 data &= ~MC_HUB_MISC_VM_CG__ENABLE_MASK;
1547                 WREG32(mmMC_HUB_MISC_VM_CG, data);
1548
1549                 data = RREG32(mmMC_XPB_CLK_GAT);
1550                 data &= ~MC_XPB_CLK_GAT__ENABLE_MASK;
1551                 WREG32(mmMC_XPB_CLK_GAT, data);
1552
1553                 data = RREG32(mmATC_MISC_CG);
1554                 data &= ~ATC_MISC_CG__ENABLE_MASK;
1555                 WREG32(mmATC_MISC_CG, data);
1556
1557                 data = RREG32(mmMC_CITF_MISC_WR_CG);
1558                 data &= ~MC_CITF_MISC_WR_CG__ENABLE_MASK;
1559                 WREG32(mmMC_CITF_MISC_WR_CG, data);
1560
1561                 data = RREG32(mmMC_CITF_MISC_RD_CG);
1562                 data &= ~MC_CITF_MISC_RD_CG__ENABLE_MASK;
1563                 WREG32(mmMC_CITF_MISC_RD_CG, data);
1564
1565                 data = RREG32(mmMC_CITF_MISC_VM_CG);
1566                 data &= ~MC_CITF_MISC_VM_CG__ENABLE_MASK;
1567                 WREG32(mmMC_CITF_MISC_VM_CG, data);
1568
1569                 data = RREG32(mmVM_L2_CG);
1570                 data &= ~VM_L2_CG__ENABLE_MASK;
1571                 WREG32(mmVM_L2_CG, data);
1572         }
1573 }
1574
1575 static void fiji_update_mc_light_sleep(struct amdgpu_device *adev,
1576                                        bool enable)
1577 {
1578         uint32_t data;
1579
1580         if (enable && (adev->cg_flags & AMD_CG_SUPPORT_MC_LS)) {
1581                 data = RREG32(mmMC_HUB_MISC_HUB_CG);
1582                 data |= MC_HUB_MISC_HUB_CG__MEM_LS_ENABLE_MASK;
1583                 WREG32(mmMC_HUB_MISC_HUB_CG, data);
1584
1585                 data = RREG32(mmMC_HUB_MISC_SIP_CG);
1586                 data |= MC_HUB_MISC_SIP_CG__MEM_LS_ENABLE_MASK;
1587                 WREG32(mmMC_HUB_MISC_SIP_CG, data);
1588
1589                 data = RREG32(mmMC_HUB_MISC_VM_CG);
1590                 data |= MC_HUB_MISC_VM_CG__MEM_LS_ENABLE_MASK;
1591                 WREG32(mmMC_HUB_MISC_VM_CG, data);
1592
1593                 data = RREG32(mmMC_XPB_CLK_GAT);
1594                 data |= MC_XPB_CLK_GAT__MEM_LS_ENABLE_MASK;
1595                 WREG32(mmMC_XPB_CLK_GAT, data);
1596
1597                 data = RREG32(mmATC_MISC_CG);
1598                 data |= ATC_MISC_CG__MEM_LS_ENABLE_MASK;
1599                 WREG32(mmATC_MISC_CG, data);
1600
1601                 data = RREG32(mmMC_CITF_MISC_WR_CG);
1602                 data |= MC_CITF_MISC_WR_CG__MEM_LS_ENABLE_MASK;
1603                 WREG32(mmMC_CITF_MISC_WR_CG, data);
1604
1605                 data = RREG32(mmMC_CITF_MISC_RD_CG);
1606                 data |= MC_CITF_MISC_RD_CG__MEM_LS_ENABLE_MASK;
1607                 WREG32(mmMC_CITF_MISC_RD_CG, data);
1608
1609                 data = RREG32(mmMC_CITF_MISC_VM_CG);
1610                 data |= MC_CITF_MISC_VM_CG__MEM_LS_ENABLE_MASK;
1611                 WREG32(mmMC_CITF_MISC_VM_CG, data);
1612
1613                 data = RREG32(mmVM_L2_CG);
1614                 data |= VM_L2_CG__MEM_LS_ENABLE_MASK;
1615                 WREG32(mmVM_L2_CG, data);
1616         } else {
1617                 data = RREG32(mmMC_HUB_MISC_HUB_CG);
1618                 data &= ~MC_HUB_MISC_HUB_CG__MEM_LS_ENABLE_MASK;
1619                 WREG32(mmMC_HUB_MISC_HUB_CG, data);
1620
1621                 data = RREG32(mmMC_HUB_MISC_SIP_CG);
1622                 data &= ~MC_HUB_MISC_SIP_CG__MEM_LS_ENABLE_MASK;
1623                 WREG32(mmMC_HUB_MISC_SIP_CG, data);
1624
1625                 data = RREG32(mmMC_HUB_MISC_VM_CG);
1626                 data &= ~MC_HUB_MISC_VM_CG__MEM_LS_ENABLE_MASK;
1627                 WREG32(mmMC_HUB_MISC_VM_CG, data);
1628
1629                 data = RREG32(mmMC_XPB_CLK_GAT);
1630                 data &= ~MC_XPB_CLK_GAT__MEM_LS_ENABLE_MASK;
1631                 WREG32(mmMC_XPB_CLK_GAT, data);
1632
1633                 data = RREG32(mmATC_MISC_CG);
1634                 data &= ~ATC_MISC_CG__MEM_LS_ENABLE_MASK;
1635                 WREG32(mmATC_MISC_CG, data);
1636
1637                 data = RREG32(mmMC_CITF_MISC_WR_CG);
1638                 data &= ~MC_CITF_MISC_WR_CG__MEM_LS_ENABLE_MASK;
1639                 WREG32(mmMC_CITF_MISC_WR_CG, data);
1640
1641                 data = RREG32(mmMC_CITF_MISC_RD_CG);
1642                 data &= ~MC_CITF_MISC_RD_CG__MEM_LS_ENABLE_MASK;
1643                 WREG32(mmMC_CITF_MISC_RD_CG, data);
1644
1645                 data = RREG32(mmMC_CITF_MISC_VM_CG);
1646                 data &= ~MC_CITF_MISC_VM_CG__MEM_LS_ENABLE_MASK;
1647                 WREG32(mmMC_CITF_MISC_VM_CG, data);
1648
1649                 data = RREG32(mmVM_L2_CG);
1650                 data &= ~VM_L2_CG__MEM_LS_ENABLE_MASK;
1651                 WREG32(mmVM_L2_CG, data);
1652         }
1653 }
1654
1655 static int gmc_v8_0_set_clockgating_state(void *handle,
1656                                           enum amd_clockgating_state state)
1657 {
1658         struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1659
1660         if (amdgpu_sriov_vf(adev))
1661                 return 0;
1662
1663         switch (adev->asic_type) {
1664         case CHIP_FIJI:
1665                 fiji_update_mc_medium_grain_clock_gating(adev,
1666                                 state == AMD_CG_STATE_GATE);
1667                 fiji_update_mc_light_sleep(adev,
1668                                 state == AMD_CG_STATE_GATE);
1669                 break;
1670         default:
1671                 break;
1672         }
1673         return 0;
1674 }
1675
1676 static int gmc_v8_0_set_powergating_state(void *handle,
1677                                           enum amd_powergating_state state)
1678 {
1679         return 0;
1680 }
1681
1682 static void gmc_v8_0_get_clockgating_state(void *handle, u32 *flags)
1683 {
1684         struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1685         int data;
1686
1687         if (amdgpu_sriov_vf(adev))
1688                 *flags = 0;
1689
1690         /* AMD_CG_SUPPORT_MC_MGCG */
1691         data = RREG32(mmMC_HUB_MISC_HUB_CG);
1692         if (data & MC_HUB_MISC_HUB_CG__ENABLE_MASK)
1693                 *flags |= AMD_CG_SUPPORT_MC_MGCG;
1694
1695         /* AMD_CG_SUPPORT_MC_LS */
1696         if (data & MC_HUB_MISC_HUB_CG__MEM_LS_ENABLE_MASK)
1697                 *flags |= AMD_CG_SUPPORT_MC_LS;
1698 }
1699
1700 static const struct amd_ip_funcs gmc_v8_0_ip_funcs = {
1701         .name = "gmc_v8_0",
1702         .early_init = gmc_v8_0_early_init,
1703         .late_init = gmc_v8_0_late_init,
1704         .sw_init = gmc_v8_0_sw_init,
1705         .sw_fini = gmc_v8_0_sw_fini,
1706         .hw_init = gmc_v8_0_hw_init,
1707         .hw_fini = gmc_v8_0_hw_fini,
1708         .suspend = gmc_v8_0_suspend,
1709         .resume = gmc_v8_0_resume,
1710         .is_idle = gmc_v8_0_is_idle,
1711         .wait_for_idle = gmc_v8_0_wait_for_idle,
1712         .check_soft_reset = gmc_v8_0_check_soft_reset,
1713         .pre_soft_reset = gmc_v8_0_pre_soft_reset,
1714         .soft_reset = gmc_v8_0_soft_reset,
1715         .post_soft_reset = gmc_v8_0_post_soft_reset,
1716         .set_clockgating_state = gmc_v8_0_set_clockgating_state,
1717         .set_powergating_state = gmc_v8_0_set_powergating_state,
1718         .get_clockgating_state = gmc_v8_0_get_clockgating_state,
1719 };
1720
1721 static const struct amdgpu_gmc_funcs gmc_v8_0_gmc_funcs = {
1722         .flush_gpu_tlb = gmc_v8_0_flush_gpu_tlb,
1723         .emit_flush_gpu_tlb = gmc_v8_0_emit_flush_gpu_tlb,
1724         .emit_pasid_mapping = gmc_v8_0_emit_pasid_mapping,
1725         .set_prt = gmc_v8_0_set_prt,
1726         .get_vm_pte_flags = gmc_v8_0_get_vm_pte_flags,
1727         .get_vm_pde = gmc_v8_0_get_vm_pde
1728 };
1729
1730 static const struct amdgpu_irq_src_funcs gmc_v8_0_irq_funcs = {
1731         .set = gmc_v8_0_vm_fault_interrupt_state,
1732         .process = gmc_v8_0_process_interrupt,
1733 };
1734
1735 static void gmc_v8_0_set_gmc_funcs(struct amdgpu_device *adev)
1736 {
1737         adev->gmc.gmc_funcs = &gmc_v8_0_gmc_funcs;
1738 }
1739
1740 static void gmc_v8_0_set_irq_funcs(struct amdgpu_device *adev)
1741 {
1742         adev->gmc.vm_fault.num_types = 1;
1743         adev->gmc.vm_fault.funcs = &gmc_v8_0_irq_funcs;
1744 }
1745
1746 const struct amdgpu_ip_block_version gmc_v8_0_ip_block =
1747 {
1748         .type = AMD_IP_BLOCK_TYPE_GMC,
1749         .major = 8,
1750         .minor = 0,
1751         .rev = 0,
1752         .funcs = &gmc_v8_0_ip_funcs,
1753 };
1754
1755 const struct amdgpu_ip_block_version gmc_v8_1_ip_block =
1756 {
1757         .type = AMD_IP_BLOCK_TYPE_GMC,
1758         .major = 8,
1759         .minor = 1,
1760         .rev = 0,
1761         .funcs = &gmc_v8_0_ip_funcs,
1762 };
1763
1764 const struct amdgpu_ip_block_version gmc_v8_5_ip_block =
1765 {
1766         .type = AMD_IP_BLOCK_TYPE_GMC,
1767         .major = 8,
1768         .minor = 5,
1769         .rev = 0,
1770         .funcs = &gmc_v8_0_ip_funcs,
1771 };