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/polaris12_32_mc.bin");
  63 MODULE_FIRMWARE("amdgpu/polaris11_k_mc.bin");
  64 MODULE_FIRMWARE("amdgpu/polaris10_k_mc.bin");
  65 MODULE_FIRMWARE("amdgpu/polaris12_k_mc.bin");
  66
  67 static const u32 golden_settings_tonga_a11[] =
  68 {
  69         mmMC_ARB_WTM_GRPWT_RD, 0x00000003, 0x00000000,
  70         mmMC_HUB_RDREQ_DMIF_LIMIT, 0x0000007f, 0x00000028,
  71         mmMC_HUB_WDP_UMC, 0x00007fb6, 0x00000991,
  72         mmVM_PRT_APERTURE0_LOW_ADDR, 0x0fffffff, 0x0fffffff,
  73         mmVM_PRT_APERTURE1_LOW_ADDR, 0x0fffffff, 0x0fffffff,
  74         mmVM_PRT_APERTURE2_LOW_ADDR, 0x0fffffff, 0x0fffffff,
  75         mmVM_PRT_APERTURE3_LOW_ADDR, 0x0fffffff, 0x0fffffff,
  76 };
  77
  78 static const u32 tonga_mgcg_cgcg_init[] =
  79 {
  80         mmMC_MEM_POWER_LS, 0xffffffff, 0x00000104
  81 };
  82
  83 static const u32 golden_settings_fiji_a10[] =
  84 {
  85         mmVM_PRT_APERTURE0_LOW_ADDR, 0x0fffffff, 0x0fffffff,
  86         mmVM_PRT_APERTURE1_LOW_ADDR, 0x0fffffff, 0x0fffffff,
  87         mmVM_PRT_APERTURE2_LOW_ADDR, 0x0fffffff, 0x0fffffff,
  88         mmVM_PRT_APERTURE3_LOW_ADDR, 0x0fffffff, 0x0fffffff,
  89 };
  90
  91 static const u32 fiji_mgcg_cgcg_init[] =
  92 {
  93         mmMC_MEM_POWER_LS, 0xffffffff, 0x00000104
  94 };
  95
  96 static const u32 golden_settings_polaris11_a11[] =
  97 {
  98         mmVM_PRT_APERTURE0_LOW_ADDR, 0x0fffffff, 0x0fffffff,
  99         mmVM_PRT_APERTURE1_LOW_ADDR, 0x0fffffff, 0x0fffffff,
 100         mmVM_PRT_APERTURE2_LOW_ADDR, 0x0fffffff, 0x0fffffff,
 101         mmVM_PRT_APERTURE3_LOW_ADDR, 0x0fffffff, 0x0fffffff
 102 };
 103
 104 static const u32 golden_settings_polaris10_a11[] =
 105 {
 106         mmMC_ARB_WTM_GRPWT_RD, 0x00000003, 0x00000000,
 107         mmVM_PRT_APERTURE0_LOW_ADDR, 0x0fffffff, 0x0fffffff,
 108         mmVM_PRT_APERTURE1_LOW_ADDR, 0x0fffffff, 0x0fffffff,
 109         mmVM_PRT_APERTURE2_LOW_ADDR, 0x0fffffff, 0x0fffffff,
 110         mmVM_PRT_APERTURE3_LOW_ADDR, 0x0fffffff, 0x0fffffff
 111 };
 112
 113 static const u32 cz_mgcg_cgcg_init[] =
 114 {
 115         mmMC_MEM_POWER_LS, 0xffffffff, 0x00000104
 116 };
 117
 118 static const u32 stoney_mgcg_cgcg_init[] =
 119 {
 120         mmATC_MISC_CG, 0xffffffff, 0x000c0200,
 121         mmMC_MEM_POWER_LS, 0xffffffff, 0x00000104
 122 };
 123
 124 static const u32 golden_settings_stoney_common[] =
 125 {
 126         mmMC_HUB_RDREQ_UVD, MC_HUB_RDREQ_UVD__PRESCALE_MASK, 0x00000004,
 127         mmMC_RD_GRP_OTH, MC_RD_GRP_OTH__UVD_MASK, 0x00600000
 128 };
 129
 130 static void gmc_v8_0_init_golden_registers(struct amdgpu_device *adev)
 131 {
 132         switch (adev->asic_type) {
 133         case CHIP_FIJI:
 134                 amdgpu_device_program_register_sequence(adev,
 135                                                         fiji_mgcg_cgcg_init,
 136                                                         ARRAY_SIZE(fiji_mgcg_cgcg_init));
 137                 amdgpu_device_program_register_sequence(adev,
 138                                                         golden_settings_fiji_a10,
 139                                                         ARRAY_SIZE(golden_settings_fiji_a10));
 140                 break;
 141         case CHIP_TONGA:
 142                 amdgpu_device_program_register_sequence(adev,
 143                                                         tonga_mgcg_cgcg_init,
 144                                                         ARRAY_SIZE(tonga_mgcg_cgcg_init));
 145                 amdgpu_device_program_register_sequence(adev,
 146                                                         golden_settings_tonga_a11,
 147                                                         ARRAY_SIZE(golden_settings_tonga_a11));
 148                 break;
 149         case CHIP_POLARIS11:
 150         case CHIP_POLARIS12:
 151         case CHIP_VEGAM:
 152                 amdgpu_device_program_register_sequence(adev,
 153                                                         golden_settings_polaris11_a11,
 154                                                         ARRAY_SIZE(golden_settings_polaris11_a11));
 155                 break;
 156         case CHIP_POLARIS10:
 157                 amdgpu_device_program_register_sequence(adev,
 158                                                         golden_settings_polaris10_a11,
 159                                                         ARRAY_SIZE(golden_settings_polaris10_a11));
 160                 break;
 161         case CHIP_CARRIZO:
 162                 amdgpu_device_program_register_sequence(adev,
 163                                                         cz_mgcg_cgcg_init,
 164                                                         ARRAY_SIZE(cz_mgcg_cgcg_init));
 165                 break;
 166         case CHIP_STONEY:
 167                 amdgpu_device_program_register_sequence(adev,
 168                                                         stoney_mgcg_cgcg_init,
 169                                                         ARRAY_SIZE(stoney_mgcg_cgcg_init));
 170                 amdgpu_device_program_register_sequence(adev,
 171                                                         golden_settings_stoney_common,
 172                                                         ARRAY_SIZE(golden_settings_stoney_common));
 173                 break;
 174         default:
 175                 break;
 176         }
 177 }
 178
 179 static void gmc_v8_0_mc_stop(struct amdgpu_device *adev)
 180 {
 181         u32 blackout;
 182
 183         gmc_v8_0_wait_for_idle(adev);
 184
 185         blackout = RREG32(mmMC_SHARED_BLACKOUT_CNTL);
 186         if (REG_GET_FIELD(blackout, MC_SHARED_BLACKOUT_CNTL, BLACKOUT_MODE) != 1) {
 187                 /* Block CPU access */
 188                 WREG32(mmBIF_FB_EN, 0);
 189                 /* blackout the MC */
 190                 blackout = REG_SET_FIELD(blackout,
 191                                          MC_SHARED_BLACKOUT_CNTL, BLACKOUT_MODE, 1);
 192                 WREG32(mmMC_SHARED_BLACKOUT_CNTL, blackout);
 193         }
 194         /* wait for the MC to settle */
 195         udelay(100);
 196 }
 197
 198 static void gmc_v8_0_mc_resume(struct amdgpu_device *adev)
 199 {
 200         u32 tmp;
 201
 202         /* unblackout the MC */
 203         tmp = RREG32(mmMC_SHARED_BLACKOUT_CNTL);
 204         tmp = REG_SET_FIELD(tmp, MC_SHARED_BLACKOUT_CNTL, BLACKOUT_MODE, 0);
 205         WREG32(mmMC_SHARED_BLACKOUT_CNTL, tmp);
 206         /* allow CPU access */
 207         tmp = REG_SET_FIELD(0, BIF_FB_EN, FB_READ_EN, 1);
 208         tmp = REG_SET_FIELD(tmp, BIF_FB_EN, FB_WRITE_EN, 1);
 209         WREG32(mmBIF_FB_EN, tmp);
 210 }
 211
 212 /**
 213  * gmc_v8_0_init_microcode - load ucode images from disk
 214  *
 215  * @adev: amdgpu_device pointer
 216  *
 217  * Use the firmware interface to load the ucode images into
 218  * the driver (not loaded into hw).
 219  * Returns 0 on success, error on failure.
 220  */
 221 static int gmc_v8_0_init_microcode(struct amdgpu_device *adev)
 222 {
 223         const char *chip_name;
 224         char fw_name[30];
 225         int err;
 226
 227         DRM_DEBUG("\n");
 228
 229         switch (adev->asic_type) {
 230         case CHIP_TONGA:
 231                 chip_name = "tonga";
 232                 break;
 233         case CHIP_POLARIS11:
 234                 if (ASICID_IS_P21(adev->pdev->device, adev->pdev->revision) ||
 235                     ASICID_IS_P31(adev->pdev->device, adev->pdev->revision))
 236                         chip_name = "polaris11_k";
 237                 else
 238                         chip_name = "polaris11";
 239                 break;
 240         case CHIP_POLARIS10:
 241                 if (ASICID_IS_P30(adev->pdev->device, adev->pdev->revision))
 242                         chip_name = "polaris10_k";
 243                 else
 244                         chip_name = "polaris10";
 245                 break;
 246         case CHIP_POLARIS12:
 247                 if (ASICID_IS_P23(adev->pdev->device, adev->pdev->revision)) {
 248                         chip_name = "polaris12_k";
 249                 } else {
 250                         WREG32(mmMC_SEQ_IO_DEBUG_INDEX, ixMC_IO_DEBUG_UP_159);
 251                         /* Polaris12 32bit ASIC needs a special MC firmware */
 252                         if (RREG32(mmMC_SEQ_IO_DEBUG_DATA) == 0x05b4dc40)
 253                                 chip_name = "polaris12_32";
 254                         else
 255                                 chip_name = "polaris12";
 256                 }
 257                 break;
 258         case CHIP_FIJI:
 259         case CHIP_CARRIZO:
 260         case CHIP_STONEY:
 261         case CHIP_VEGAM:
 262                 return 0;
 263         default: BUG();
 264         }
 265
 266         snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_mc.bin", chip_name);
 267         err = amdgpu_ucode_request(adev, &adev->gmc.fw, fw_name);
 268         if (err) {
 269                 pr_err("mc: Failed to load firmware \"%s\"\n", fw_name);
 270                 amdgpu_ucode_release(&adev->gmc.fw);
 271         }
 272         return err;
 273 }
 274
 275 /**
 276  * gmc_v8_0_tonga_mc_load_microcode - load tonga MC ucode into the hw
 277  *
 278  * @adev: amdgpu_device pointer
 279  *
 280  * Load the GDDR MC ucode into the hw (VI).
 281  * Returns 0 on success, error on failure.
 282  */
 283 static int gmc_v8_0_tonga_mc_load_microcode(struct amdgpu_device *adev)
 284 {
 285         const struct mc_firmware_header_v1_0 *hdr;
 286         const __le32 *fw_data = NULL;
 287         const __le32 *io_mc_regs = NULL;
 288         u32 running;
 289         int i, ucode_size, regs_size;
 290
 291         /* Skip MC ucode loading on SR-IOV capable boards.
 292          * vbios does this for us in asic_init in that case.
 293          * Skip MC ucode loading on VF, because hypervisor will do that
 294          * for this adaptor.
 295          */
 296         if (amdgpu_sriov_bios(adev))
 297                 return 0;
 298
 299         if (!adev->gmc.fw)
 300                 return -EINVAL;
 301
 302         hdr = (const struct mc_firmware_header_v1_0 *)adev->gmc.fw->data;
 303         amdgpu_ucode_print_mc_hdr(&hdr->header);
 304
 305         adev->gmc.fw_version = le32_to_cpu(hdr->header.ucode_version);
 306         regs_size = le32_to_cpu(hdr->io_debug_size_bytes) / (4 * 2);
 307         io_mc_regs = (const __le32 *)
 308                 (adev->gmc.fw->data + le32_to_cpu(hdr->io_debug_array_offset_bytes));
 309         ucode_size = le32_to_cpu(hdr->header.ucode_size_bytes) / 4;
 310         fw_data = (const __le32 *)
 311                 (adev->gmc.fw->data + le32_to_cpu(hdr->header.ucode_array_offset_bytes));
 312
 313         running = REG_GET_FIELD(RREG32(mmMC_SEQ_SUP_CNTL), MC_SEQ_SUP_CNTL, RUN);
 314
 315         if (running == 0) {
 316                 /* reset the engine and set to writable */
 317                 WREG32(mmMC_SEQ_SUP_CNTL, 0x00000008);
 318                 WREG32(mmMC_SEQ_SUP_CNTL, 0x00000010);
 319
 320                 /* load mc io regs */
 321                 for (i = 0; i < regs_size; i++) {
 322                         WREG32(mmMC_SEQ_IO_DEBUG_INDEX, le32_to_cpup(io_mc_regs++));
 323                         WREG32(mmMC_SEQ_IO_DEBUG_DATA, le32_to_cpup(io_mc_regs++));
 324                 }
 325                 /* load the MC ucode */
 326                 for (i = 0; i < ucode_size; i++)
 327                         WREG32(mmMC_SEQ_SUP_PGM, le32_to_cpup(fw_data++));
 328
 329                 /* put the engine back into the active state */
 330                 WREG32(mmMC_SEQ_SUP_CNTL, 0x00000008);
 331                 WREG32(mmMC_SEQ_SUP_CNTL, 0x00000004);
 332                 WREG32(mmMC_SEQ_SUP_CNTL, 0x00000001);
 333
 334                 /* wait for training to complete */
 335                 for (i = 0; i < adev->usec_timeout; i++) {
 336                         if (REG_GET_FIELD(RREG32(mmMC_SEQ_TRAIN_WAKEUP_CNTL),
 337                                           MC_SEQ_TRAIN_WAKEUP_CNTL, TRAIN_DONE_D0))
 338                                 break;
 339                         udelay(1);
 340                 }
 341                 for (i = 0; i < adev->usec_timeout; i++) {
 342                         if (REG_GET_FIELD(RREG32(mmMC_SEQ_TRAIN_WAKEUP_CNTL),
 343                                           MC_SEQ_TRAIN_WAKEUP_CNTL, TRAIN_DONE_D1))
 344                                 break;
 345                         udelay(1);
 346                 }
 347         }
 348
 349         return 0;
 350 }
 351
 352 static int gmc_v8_0_polaris_mc_load_microcode(struct amdgpu_device *adev)
 353 {
 354         const struct mc_firmware_header_v1_0 *hdr;
 355         const __le32 *fw_data = NULL;
 356         const __le32 *io_mc_regs = NULL;
 357         u32 data;
 358         int i, ucode_size, regs_size;
 359
 360         /* Skip MC ucode loading on SR-IOV capable boards.
 361          * vbios does this for us in asic_init in that case.
 362          * Skip MC ucode loading on VF, because hypervisor will do that
 363          * for this adaptor.
 364          */
 365         if (amdgpu_sriov_bios(adev))
 366                 return 0;
 367
 368         if (!adev->gmc.fw)
 369                 return -EINVAL;
 370
 371         hdr = (const struct mc_firmware_header_v1_0 *)adev->gmc.fw->data;
 372         amdgpu_ucode_print_mc_hdr(&hdr->header);
 373
 374         adev->gmc.fw_version = le32_to_cpu(hdr->header.ucode_version);
 375         regs_size = le32_to_cpu(hdr->io_debug_size_bytes) / (4 * 2);
 376         io_mc_regs = (const __le32 *)
 377                 (adev->gmc.fw->data + le32_to_cpu(hdr->io_debug_array_offset_bytes));
 378         ucode_size = le32_to_cpu(hdr->header.ucode_size_bytes) / 4;
 379         fw_data = (const __le32 *)
 380                 (adev->gmc.fw->data + le32_to_cpu(hdr->header.ucode_array_offset_bytes));
 381
 382         data = RREG32(mmMC_SEQ_MISC0);
 383         data &= ~(0x40);
 384         WREG32(mmMC_SEQ_MISC0, data);
 385
 386         /* load mc io regs */
 387         for (i = 0; i < regs_size; i++) {
 388                 WREG32(mmMC_SEQ_IO_DEBUG_INDEX, le32_to_cpup(io_mc_regs++));
 389                 WREG32(mmMC_SEQ_IO_DEBUG_DATA, le32_to_cpup(io_mc_regs++));
 390         }
 391
 392         WREG32(mmMC_SEQ_SUP_CNTL, 0x00000008);
 393         WREG32(mmMC_SEQ_SUP_CNTL, 0x00000010);
 394
 395         /* load the MC ucode */
 396         for (i = 0; i < ucode_size; i++)
 397                 WREG32(mmMC_SEQ_SUP_PGM, le32_to_cpup(fw_data++));
 398
 399         /* put the engine back into the active state */
 400         WREG32(mmMC_SEQ_SUP_CNTL, 0x00000008);
 401         WREG32(mmMC_SEQ_SUP_CNTL, 0x00000004);
 402         WREG32(mmMC_SEQ_SUP_CNTL, 0x00000001);
 403
 404         /* wait for training to complete */
 405         for (i = 0; i < adev->usec_timeout; i++) {
 406                 data = RREG32(mmMC_SEQ_MISC0);
 407                 if (data & 0x80)
 408                         break;
 409                 udelay(1);
 410         }
 411
 412         return 0;
 413 }
 414
 415 static void gmc_v8_0_vram_gtt_location(struct amdgpu_device *adev,
 416                                        struct amdgpu_gmc *mc)
 417 {
 418         u64 base = 0;
 419
 420         if (!amdgpu_sriov_vf(adev))
 421                 base = RREG32(mmMC_VM_FB_LOCATION) & 0xFFFF;
 422         base <<= 24;
 423
 424         amdgpu_gmc_vram_location(adev, mc, base);
 425         amdgpu_gmc_gart_location(adev, mc);
 426 }
 427
 428 /**
 429  * gmc_v8_0_mc_program - program the GPU memory controller
 430  *
 431  * @adev: amdgpu_device pointer
 432  *
 433  * Set the location of vram, gart, and AGP in the GPU's
 434  * physical address space (VI).
 435  */
 436 static void gmc_v8_0_mc_program(struct amdgpu_device *adev)
 437 {
 438         u32 tmp;
 439         int i, j;
 440
 441         /* Initialize HDP */
 442         for (i = 0, j = 0; i < 32; i++, j += 0x6) {
 443                 WREG32((0xb05 + j), 0x00000000);
 444                 WREG32((0xb06 + j), 0x00000000);
 445                 WREG32((0xb07 + j), 0x00000000);
 446                 WREG32((0xb08 + j), 0x00000000);
 447                 WREG32((0xb09 + j), 0x00000000);
 448         }
 449         WREG32(mmHDP_REG_COHERENCY_FLUSH_CNTL, 0);
 450
 451         if (gmc_v8_0_wait_for_idle((void *)adev)) {
 452                 dev_warn(adev->dev, "Wait for MC idle timedout !\n");
 453         }
 454         if (adev->mode_info.num_crtc) {
 455                 /* Lockout access through VGA aperture*/
 456                 tmp = RREG32(mmVGA_HDP_CONTROL);
 457                 tmp = REG_SET_FIELD(tmp, VGA_HDP_CONTROL, VGA_MEMORY_DISABLE, 1);
 458                 WREG32(mmVGA_HDP_CONTROL, tmp);
 459
 460                 /* disable VGA render */
 461                 tmp = RREG32(mmVGA_RENDER_CONTROL);
 462                 tmp = REG_SET_FIELD(tmp, VGA_RENDER_CONTROL, VGA_VSTATUS_CNTL, 0);
 463                 WREG32(mmVGA_RENDER_CONTROL, tmp);
 464         }
 465         /* Update configuration */
 466         WREG32(mmMC_VM_SYSTEM_APERTURE_LOW_ADDR,
 467                adev->gmc.vram_start >> 12);
 468         WREG32(mmMC_VM_SYSTEM_APERTURE_HIGH_ADDR,
 469                adev->gmc.vram_end >> 12);
 470         WREG32(mmMC_VM_SYSTEM_APERTURE_DEFAULT_ADDR,
 471                adev->mem_scratch.gpu_addr >> 12);
 472
 473         if (amdgpu_sriov_vf(adev)) {
 474                 tmp = ((adev->gmc.vram_end >> 24) & 0xFFFF) << 16;
 475                 tmp |= ((adev->gmc.vram_start >> 24) & 0xFFFF);
 476                 WREG32(mmMC_VM_FB_LOCATION, tmp);
 477                 /* XXX double check these! */
 478                 WREG32(mmHDP_NONSURFACE_BASE, (adev->gmc.vram_start >> 8));
 479                 WREG32(mmHDP_NONSURFACE_INFO, (2 << 7) | (1 << 30));
 480                 WREG32(mmHDP_NONSURFACE_SIZE, 0x3FFFFFFF);
 481         }
 482
 483         WREG32(mmMC_VM_AGP_BASE, 0);
 484         WREG32(mmMC_VM_AGP_TOP, 0x0FFFFFFF);
 485         WREG32(mmMC_VM_AGP_BOT, 0x0FFFFFFF);
 486         if (gmc_v8_0_wait_for_idle((void *)adev)) {
 487                 dev_warn(adev->dev, "Wait for MC idle timedout !\n");
 488         }
 489
 490         WREG32(mmBIF_FB_EN, BIF_FB_EN__FB_READ_EN_MASK | BIF_FB_EN__FB_WRITE_EN_MASK);
 491
 492         tmp = RREG32(mmHDP_MISC_CNTL);
 493         tmp = REG_SET_FIELD(tmp, HDP_MISC_CNTL, FLUSH_INVALIDATE_CACHE, 0);
 494         WREG32(mmHDP_MISC_CNTL, tmp);
 495
 496         tmp = RREG32(mmHDP_HOST_PATH_CNTL);
 497         WREG32(mmHDP_HOST_PATH_CNTL, tmp);
 498 }
 499
 500 /**
 501  * gmc_v8_0_mc_init - initialize the memory controller driver params
 502  *
 503  * @adev: amdgpu_device pointer
 504  *
 505  * Look up the amount of vram, vram width, and decide how to place
 506  * vram and gart within the GPU's physical address space (VI).
 507  * Returns 0 for success.
 508  */
 509 static int gmc_v8_0_mc_init(struct amdgpu_device *adev)
 510 {
 511         int r;
 512         u32 tmp;
 513
 514         adev->gmc.vram_width = amdgpu_atombios_get_vram_width(adev);
 515         if (!adev->gmc.vram_width) {
 516                 int chansize, numchan;
 517
 518                 /* Get VRAM informations */
 519                 tmp = RREG32(mmMC_ARB_RAMCFG);
 520                 if (REG_GET_FIELD(tmp, MC_ARB_RAMCFG, CHANSIZE)) {
 521                         chansize = 64;
 522                 } else {
 523                         chansize = 32;
 524                 }
 525                 tmp = RREG32(mmMC_SHARED_CHMAP);
 526                 switch (REG_GET_FIELD(tmp, MC_SHARED_CHMAP, NOOFCHAN)) {
 527                 case 0:
 528                 default:
 529                         numchan = 1;
 530                         break;
 531                 case 1:
 532                         numchan = 2;
 533                         break;
 534                 case 2:
 535                         numchan = 4;
 536                         break;
 537                 case 3:
 538                         numchan = 8;
 539                         break;
 540                 case 4:
 541                         numchan = 3;
 542                         break;
 543                 case 5:
 544                         numchan = 6;
 545                         break;
 546                 case 6:
 547                         numchan = 10;
 548                         break;
 549                 case 7:
 550                         numchan = 12;
 551                         break;
 552                 case 8:
 553                         numchan = 16;
 554                         break;
 555                 }
 556                 adev->gmc.vram_width = numchan * chansize;
 557         }
 558         /* size in MB on si */
 559         tmp = RREG32(mmCONFIG_MEMSIZE);
 560         /* some boards may have garbage in the upper 16 bits */
 561         if (tmp & 0xffff0000) {
 562                 DRM_INFO("Probable bad vram size: 0x%08x\n", tmp);
 563                 if (tmp & 0xffff)
 564                         tmp &= 0xffff;
 565         }
 566         adev->gmc.mc_vram_size = tmp * 1024ULL * 1024ULL;
 567         adev->gmc.real_vram_size = adev->gmc.mc_vram_size;
 568
 569         if (!(adev->flags & AMD_IS_APU)) {
 570                 r = amdgpu_device_resize_fb_bar(adev);
 571                 if (r)
 572                         return r;
 573         }
 574         adev->gmc.aper_base = pci_resource_start(adev->pdev, 0);
 575         adev->gmc.aper_size = pci_resource_len(adev->pdev, 0);
 576
 577 #ifdef CONFIG_X86_64
 578         if ((adev->flags & AMD_IS_APU) && !amdgpu_passthrough(adev)) {
 579                 adev->gmc.aper_base = ((u64)RREG32(mmMC_VM_FB_OFFSET)) << 22;
 580                 adev->gmc.aper_size = adev->gmc.real_vram_size;
 581         }
 582 #endif
 583
 584         adev->gmc.visible_vram_size = adev->gmc.aper_size;
 585
 586         /* set the gart size */
 587         if (amdgpu_gart_size == -1) {
 588                 switch (adev->asic_type) {
 589                 case CHIP_POLARIS10: /* all engines support GPUVM */
 590                 case CHIP_POLARIS11: /* all engines support GPUVM */
 591                 case CHIP_POLARIS12: /* all engines support GPUVM */
 592                 case CHIP_VEGAM:     /* all engines support GPUVM */
 593                 default:
 594                         adev->gmc.gart_size = 256ULL << 20;
 595                         break;
 596                 case CHIP_TONGA:   /* UVD, VCE do not support GPUVM */
 597                 case CHIP_FIJI:    /* UVD, VCE do not support GPUVM */
 598                 case CHIP_CARRIZO: /* UVD, VCE do not support GPUVM, DCE SG support */
 599                 case CHIP_STONEY:  /* UVD does not support GPUVM, DCE SG support */
 600                         adev->gmc.gart_size = 1024ULL << 20;
 601                         break;
 602                 }
 603         } else {
 604                 adev->gmc.gart_size = (u64)amdgpu_gart_size << 20;
 605         }
 606
 607         adev->gmc.gart_size += adev->pm.smu_prv_buffer_size;
 608         gmc_v8_0_vram_gtt_location(adev, &adev->gmc);
 609
 610         return 0;
 611 }
 612
 613 /**
 614  * gmc_v8_0_flush_gpu_tlb_pasid - tlb flush via pasid
 615  *
 616  * @adev: amdgpu_device pointer
 617  * @pasid: pasid to be flush
 618  * @flush_type: type of flush
 619  * @all_hub: flush all hubs
 620  * @inst: is used to select which instance of KIQ to use for the invalidation
 621  *
 622  * Flush the TLB for the requested pasid.
 623  */
 624 static int gmc_v8_0_flush_gpu_tlb_pasid(struct amdgpu_device *adev,
 625                                         uint16_t pasid, uint32_t flush_type,
 626                                         bool all_hub, uint32_t inst)
 627 {
 628         int vmid;
 629         unsigned int tmp;
 630
 631         if (amdgpu_in_reset(adev))
 632                 return -EIO;
 633
 634         for (vmid = 1; vmid < 16; vmid++) {
 635
 636                 tmp = RREG32(mmATC_VMID0_PASID_MAPPING + vmid);
 637                 if ((tmp & ATC_VMID0_PASID_MAPPING__VALID_MASK) &&
 638                         (tmp & ATC_VMID0_PASID_MAPPING__PASID_MASK) == pasid) {
 639                         WREG32(mmVM_INVALIDATE_REQUEST, 1 << vmid);
 640                         RREG32(mmVM_INVALIDATE_RESPONSE);
 641                         break;
 642                 }
 643         }
 644
 645         return 0;
 646
 647 }
 648
 649 /*
 650  * GART
 651  * VMID 0 is the physical GPU addresses as used by the kernel.
 652  * VMIDs 1-15 are used for userspace clients and are handled
 653  * by the amdgpu vm/hsa code.
 654  */
 655
 656 /**
 657  * gmc_v8_0_flush_gpu_tlb - gart tlb flush callback
 658  *
 659  * @adev: amdgpu_device pointer
 660  * @vmid: vm instance to flush
 661  * @vmhub: which hub to flush
 662  * @flush_type: type of flush
 663  *
 664  * Flush the TLB for the requested page table (VI).
 665  */
 666 static void gmc_v8_0_flush_gpu_tlb(struct amdgpu_device *adev, uint32_t vmid,
 667                                         uint32_t vmhub, uint32_t flush_type)
 668 {
 669         /* bits 0-15 are the VM contexts0-15 */
 670         WREG32(mmVM_INVALIDATE_REQUEST, 1 << vmid);
 671 }
 672
 673 static uint64_t gmc_v8_0_emit_flush_gpu_tlb(struct amdgpu_ring *ring,
 674                                             unsigned vmid, uint64_t pd_addr)
 675 {
 676         uint32_t reg;
 677
 678         if (vmid < 8)
 679                 reg = mmVM_CONTEXT0_PAGE_TABLE_BASE_ADDR + vmid;
 680         else
 681                 reg = mmVM_CONTEXT8_PAGE_TABLE_BASE_ADDR + vmid - 8;
 682         amdgpu_ring_emit_wreg(ring, reg, pd_addr >> 12);
 683
 684         /* bits 0-15 are the VM contexts0-15 */
 685         amdgpu_ring_emit_wreg(ring, mmVM_INVALIDATE_REQUEST, 1 << vmid);
 686
 687         return pd_addr;
 688 }
 689
 690 static void gmc_v8_0_emit_pasid_mapping(struct amdgpu_ring *ring, unsigned vmid,
 691                                         unsigned pasid)
 692 {
 693         amdgpu_ring_emit_wreg(ring, mmIH_VMID_0_LUT + vmid, pasid);
 694 }
 695
 696 /*
 697  * PTE format on VI:
 698  * 63:40 reserved
 699  * 39:12 4k physical page base address
 700  * 11:7 fragment
 701  * 6 write
 702  * 5 read
 703  * 4 exe
 704  * 3 reserved
 705  * 2 snooped
 706  * 1 system
 707  * 0 valid
 708  *
 709  * PDE format on VI:
 710  * 63:59 block fragment size
 711  * 58:40 reserved
 712  * 39:1 physical base address of PTE
 713  * bits 5:1 must be 0.
 714  * 0 valid
 715  */
 716
 717 static void gmc_v8_0_get_vm_pde(struct amdgpu_device *adev, int level,
 718                                 uint64_t *addr, uint64_t *flags)
 719 {
 720         BUG_ON(*addr & 0xFFFFFF0000000FFFULL);
 721 }
 722
 723 static void gmc_v8_0_get_vm_pte(struct amdgpu_device *adev,
 724                                 struct amdgpu_bo_va_mapping *mapping,
 725                                 uint64_t *flags)
 726 {
 727         *flags &= ~AMDGPU_PTE_EXECUTABLE;
 728         *flags |= mapping->flags & AMDGPU_PTE_EXECUTABLE;
 729         *flags &= ~AMDGPU_PTE_PRT;
 730 }
 731
 732 /**
 733  * gmc_v8_0_set_fault_enable_default - update VM fault handling
 734  *
 735  * @adev: amdgpu_device pointer
 736  * @value: true redirects VM faults to the default page
 737  */
 738 static void gmc_v8_0_set_fault_enable_default(struct amdgpu_device *adev,
 739                                               bool value)
 740 {
 741         u32 tmp;
 742
 743         tmp = RREG32(mmVM_CONTEXT1_CNTL);
 744         tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
 745                             RANGE_PROTECTION_FAULT_ENABLE_DEFAULT, value);
 746         tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
 747                             DUMMY_PAGE_PROTECTION_FAULT_ENABLE_DEFAULT, value);
 748         tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
 749                             PDE0_PROTECTION_FAULT_ENABLE_DEFAULT, value);
 750         tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
 751                             VALID_PROTECTION_FAULT_ENABLE_DEFAULT, value);
 752         tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
 753                             READ_PROTECTION_FAULT_ENABLE_DEFAULT, value);
 754         tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
 755                             WRITE_PROTECTION_FAULT_ENABLE_DEFAULT, value);
 756         tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
 757                             EXECUTE_PROTECTION_FAULT_ENABLE_DEFAULT, value);
 758         WREG32(mmVM_CONTEXT1_CNTL, tmp);
 759 }
 760
 761 /**
 762  * gmc_v8_0_set_prt - set PRT VM fault
 763  *
 764  * @adev: amdgpu_device pointer
 765  * @enable: enable/disable VM fault handling for PRT
 766 */
 767 static void gmc_v8_0_set_prt(struct amdgpu_device *adev, bool enable)
 768 {
 769         u32 tmp;
 770
 771         if (enable && !adev->gmc.prt_warning) {
 772                 dev_warn(adev->dev, "Disabling VM faults because of PRT request!\n");
 773                 adev->gmc.prt_warning = true;
 774         }
 775
 776         tmp = RREG32(mmVM_PRT_CNTL);
 777         tmp = REG_SET_FIELD(tmp, VM_PRT_CNTL,
 778                             CB_DISABLE_READ_FAULT_ON_UNMAPPED_ACCESS, enable);
 779         tmp = REG_SET_FIELD(tmp, VM_PRT_CNTL,
 780                             CB_DISABLE_WRITE_FAULT_ON_UNMAPPED_ACCESS, enable);
 781         tmp = REG_SET_FIELD(tmp, VM_PRT_CNTL,
 782                             TC_DISABLE_READ_FAULT_ON_UNMAPPED_ACCESS, enable);
 783         tmp = REG_SET_FIELD(tmp, VM_PRT_CNTL,
 784                             TC_DISABLE_WRITE_FAULT_ON_UNMAPPED_ACCESS, enable);
 785         tmp = REG_SET_FIELD(tmp, VM_PRT_CNTL,
 786                             L2_CACHE_STORE_INVALID_ENTRIES, enable);
 787         tmp = REG_SET_FIELD(tmp, VM_PRT_CNTL,
 788                             L1_TLB_STORE_INVALID_ENTRIES, enable);
 789         tmp = REG_SET_FIELD(tmp, VM_PRT_CNTL,
 790                             MASK_PDE0_FAULT, enable);
 791         WREG32(mmVM_PRT_CNTL, tmp);
 792
 793         if (enable) {
 794                 uint32_t low = AMDGPU_VA_RESERVED_SIZE >> AMDGPU_GPU_PAGE_SHIFT;
 795                 uint32_t high = adev->vm_manager.max_pfn -
 796                         (AMDGPU_VA_RESERVED_SIZE >> AMDGPU_GPU_PAGE_SHIFT);
 797
 798                 WREG32(mmVM_PRT_APERTURE0_LOW_ADDR, low);
 799                 WREG32(mmVM_PRT_APERTURE1_LOW_ADDR, low);
 800                 WREG32(mmVM_PRT_APERTURE2_LOW_ADDR, low);
 801                 WREG32(mmVM_PRT_APERTURE3_LOW_ADDR, low);
 802                 WREG32(mmVM_PRT_APERTURE0_HIGH_ADDR, high);
 803                 WREG32(mmVM_PRT_APERTURE1_HIGH_ADDR, high);
 804                 WREG32(mmVM_PRT_APERTURE2_HIGH_ADDR, high);
 805                 WREG32(mmVM_PRT_APERTURE3_HIGH_ADDR, high);
 806         } else {
 807                 WREG32(mmVM_PRT_APERTURE0_LOW_ADDR, 0xfffffff);
 808                 WREG32(mmVM_PRT_APERTURE1_LOW_ADDR, 0xfffffff);
 809                 WREG32(mmVM_PRT_APERTURE2_LOW_ADDR, 0xfffffff);
 810                 WREG32(mmVM_PRT_APERTURE3_LOW_ADDR, 0xfffffff);
 811                 WREG32(mmVM_PRT_APERTURE0_HIGH_ADDR, 0x0);
 812                 WREG32(mmVM_PRT_APERTURE1_HIGH_ADDR, 0x0);
 813                 WREG32(mmVM_PRT_APERTURE2_HIGH_ADDR, 0x0);
 814                 WREG32(mmVM_PRT_APERTURE3_HIGH_ADDR, 0x0);
 815         }
 816 }
 817
 818 /**
 819  * gmc_v8_0_gart_enable - gart enable
 820  *
 821  * @adev: amdgpu_device pointer
 822  *
 823  * This sets up the TLBs, programs the page tables for VMID0,
 824  * sets up the hw for VMIDs 1-15 which are allocated on
 825  * demand, and sets up the global locations for the LDS, GDS,
 826  * and GPUVM for FSA64 clients (VI).
 827  * Returns 0 for success, errors for failure.
 828  */
 829 static int gmc_v8_0_gart_enable(struct amdgpu_device *adev)
 830 {
 831         uint64_t table_addr;
 832         u32 tmp, field;
 833         int i;
 834
 835         if (adev->gart.bo == NULL) {
 836                 dev_err(adev->dev, "No VRAM object for PCIE GART.\n");
 837                 return -EINVAL;
 838         }
 839         amdgpu_gtt_mgr_recover(&adev->mman.gtt_mgr);
 840         table_addr = amdgpu_bo_gpu_offset(adev->gart.bo);
 841
 842         /* Setup TLB control */
 843         tmp = RREG32(mmMC_VM_MX_L1_TLB_CNTL);
 844         tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL, ENABLE_L1_TLB, 1);
 845         tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL, ENABLE_L1_FRAGMENT_PROCESSING, 1);
 846         tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL, SYSTEM_ACCESS_MODE, 3);
 847         tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL, ENABLE_ADVANCED_DRIVER_MODEL, 1);
 848         tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL, SYSTEM_APERTURE_UNMAPPED_ACCESS, 0);
 849         WREG32(mmMC_VM_MX_L1_TLB_CNTL, tmp);
 850         /* Setup L2 cache */
 851         tmp = RREG32(mmVM_L2_CNTL);
 852         tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, ENABLE_L2_CACHE, 1);
 853         tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, ENABLE_L2_FRAGMENT_PROCESSING, 1);
 854         tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, ENABLE_L2_PTE_CACHE_LRU_UPDATE_BY_WRITE, 1);
 855         tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, ENABLE_L2_PDE0_CACHE_LRU_UPDATE_BY_WRITE, 1);
 856         tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, EFFECTIVE_L2_QUEUE_SIZE, 7);
 857         tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, CONTEXT1_IDENTITY_ACCESS_MODE, 1);
 858         tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, ENABLE_DEFAULT_PAGE_OUT_TO_SYSTEM_MEMORY, 1);
 859         WREG32(mmVM_L2_CNTL, tmp);
 860         tmp = RREG32(mmVM_L2_CNTL2);
 861         tmp = REG_SET_FIELD(tmp, VM_L2_CNTL2, INVALIDATE_ALL_L1_TLBS, 1);
 862         tmp = REG_SET_FIELD(tmp, VM_L2_CNTL2, INVALIDATE_L2_CACHE, 1);
 863         WREG32(mmVM_L2_CNTL2, tmp);
 864
 865         field = adev->vm_manager.fragment_size;
 866         tmp = RREG32(mmVM_L2_CNTL3);
 867         tmp = REG_SET_FIELD(tmp, VM_L2_CNTL3, L2_CACHE_BIGK_ASSOCIATIVITY, 1);
 868         tmp = REG_SET_FIELD(tmp, VM_L2_CNTL3, BANK_SELECT, field);
 869         tmp = REG_SET_FIELD(tmp, VM_L2_CNTL3, L2_CACHE_BIGK_FRAGMENT_SIZE, field);
 870         WREG32(mmVM_L2_CNTL3, tmp);
 871         /* XXX: set to enable PTE/PDE in system memory */
 872         tmp = RREG32(mmVM_L2_CNTL4);
 873         tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_CONTEXT0_PDE_REQUEST_PHYSICAL, 0);
 874         tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_CONTEXT0_PDE_REQUEST_SHARED, 0);
 875         tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_CONTEXT0_PDE_REQUEST_SNOOP, 0);
 876         tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_CONTEXT0_PTE_REQUEST_PHYSICAL, 0);
 877         tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_CONTEXT0_PTE_REQUEST_SHARED, 0);
 878         tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_CONTEXT0_PTE_REQUEST_SNOOP, 0);
 879         tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_CONTEXT1_PDE_REQUEST_PHYSICAL, 0);
 880         tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_CONTEXT1_PDE_REQUEST_SHARED, 0);
 881         tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_CONTEXT1_PDE_REQUEST_SNOOP, 0);
 882         tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_CONTEXT1_PTE_REQUEST_PHYSICAL, 0);
 883         tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_CONTEXT1_PTE_REQUEST_SHARED, 0);
 884         tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_CONTEXT1_PTE_REQUEST_SNOOP, 0);
 885         WREG32(mmVM_L2_CNTL4, tmp);
 886         /* setup context0 */
 887         WREG32(mmVM_CONTEXT0_PAGE_TABLE_START_ADDR, adev->gmc.gart_start >> 12);
 888         WREG32(mmVM_CONTEXT0_PAGE_TABLE_END_ADDR, adev->gmc.gart_end >> 12);
 889         WREG32(mmVM_CONTEXT0_PAGE_TABLE_BASE_ADDR, table_addr >> 12);
 890         WREG32(mmVM_CONTEXT0_PROTECTION_FAULT_DEFAULT_ADDR,
 891                         (u32)(adev->dummy_page_addr >> 12));
 892         WREG32(mmVM_CONTEXT0_CNTL2, 0);
 893         tmp = RREG32(mmVM_CONTEXT0_CNTL);
 894         tmp = REG_SET_FIELD(tmp, VM_CONTEXT0_CNTL, ENABLE_CONTEXT, 1);
 895         tmp = REG_SET_FIELD(tmp, VM_CONTEXT0_CNTL, PAGE_TABLE_DEPTH, 0);
 896         tmp = REG_SET_FIELD(tmp, VM_CONTEXT0_CNTL, RANGE_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
 897         WREG32(mmVM_CONTEXT0_CNTL, tmp);
 898
 899         WREG32(mmVM_L2_CONTEXT1_IDENTITY_APERTURE_LOW_ADDR, 0);
 900         WREG32(mmVM_L2_CONTEXT1_IDENTITY_APERTURE_HIGH_ADDR, 0);
 901         WREG32(mmVM_L2_CONTEXT_IDENTITY_PHYSICAL_OFFSET, 0);
 902
 903         /* empty context1-15 */
 904         /* FIXME start with 4G, once using 2 level pt switch to full
 905          * vm size space
 906          */
 907         /* set vm size, must be a multiple of 4 */
 908         WREG32(mmVM_CONTEXT1_PAGE_TABLE_START_ADDR, 0);
 909         WREG32(mmVM_CONTEXT1_PAGE_TABLE_END_ADDR, adev->vm_manager.max_pfn - 1);
 910         for (i = 1; i < AMDGPU_NUM_VMID; i++) {
 911                 if (i < 8)
 912                         WREG32(mmVM_CONTEXT0_PAGE_TABLE_BASE_ADDR + i,
 913                                table_addr >> 12);
 914                 else
 915                         WREG32(mmVM_CONTEXT8_PAGE_TABLE_BASE_ADDR + i - 8,
 916                                table_addr >> 12);
 917         }
 918
 919         /* enable context1-15 */
 920         WREG32(mmVM_CONTEXT1_PROTECTION_FAULT_DEFAULT_ADDR,
 921                (u32)(adev->dummy_page_addr >> 12));
 922         WREG32(mmVM_CONTEXT1_CNTL2, 4);
 923         tmp = RREG32(mmVM_CONTEXT1_CNTL);
 924         tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, ENABLE_CONTEXT, 1);
 925         tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, PAGE_TABLE_DEPTH, 1);
 926         tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, RANGE_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
 927         tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, DUMMY_PAGE_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
 928         tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, PDE0_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
 929         tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, VALID_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
 930         tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, READ_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
 931         tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, WRITE_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
 932         tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, EXECUTE_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
 933         tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, PAGE_TABLE_BLOCK_SIZE,
 934                             adev->vm_manager.block_size - 9);
 935         WREG32(mmVM_CONTEXT1_CNTL, tmp);
 936         if (amdgpu_vm_fault_stop == AMDGPU_VM_FAULT_STOP_ALWAYS)
 937                 gmc_v8_0_set_fault_enable_default(adev, false);
 938         else
 939                 gmc_v8_0_set_fault_enable_default(adev, true);
 940
 941         gmc_v8_0_flush_gpu_tlb(adev, 0, 0, 0);
 942         DRM_INFO("PCIE GART of %uM enabled (table at 0x%016llX).\n",
 943                  (unsigned)(adev->gmc.gart_size >> 20),
 944                  (unsigned long long)table_addr);
 945         return 0;
 946 }
 947
 948 static int gmc_v8_0_gart_init(struct amdgpu_device *adev)
 949 {
 950         int r;
 951
 952         if (adev->gart.bo) {
 953                 WARN(1, "R600 PCIE GART already initialized\n");
 954                 return 0;
 955         }
 956         /* Initialize common gart structure */
 957         r = amdgpu_gart_init(adev);
 958         if (r)
 959                 return r;
 960         adev->gart.table_size = adev->gart.num_gpu_pages * 8;
 961         adev->gart.gart_pte_flags = AMDGPU_PTE_EXECUTABLE;
 962         return amdgpu_gart_table_vram_alloc(adev);
 963 }
 964
 965 /**
 966  * gmc_v8_0_gart_disable - gart disable
 967  *
 968  * @adev: amdgpu_device pointer
 969  *
 970  * This disables all VM page table (VI).
 971  */
 972 static void gmc_v8_0_gart_disable(struct amdgpu_device *adev)
 973 {
 974         u32 tmp;
 975
 976         /* Disable all tables */
 977         WREG32(mmVM_CONTEXT0_CNTL, 0);
 978         WREG32(mmVM_CONTEXT1_CNTL, 0);
 979         /* Setup TLB control */
 980         tmp = RREG32(mmMC_VM_MX_L1_TLB_CNTL);
 981         tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL, ENABLE_L1_TLB, 0);
 982         tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL, ENABLE_L1_FRAGMENT_PROCESSING, 0);
 983         tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL, ENABLE_ADVANCED_DRIVER_MODEL, 0);
 984         WREG32(mmMC_VM_MX_L1_TLB_CNTL, tmp);
 985         /* Setup L2 cache */
 986         tmp = RREG32(mmVM_L2_CNTL);
 987         tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, ENABLE_L2_CACHE, 0);
 988         WREG32(mmVM_L2_CNTL, tmp);
 989         WREG32(mmVM_L2_CNTL2, 0);
 990 }
 991
 992 /**
 993  * gmc_v8_0_vm_decode_fault - print human readable fault info
 994  *
 995  * @adev: amdgpu_device pointer
 996  * @status: VM_CONTEXT1_PROTECTION_FAULT_STATUS register value
 997  * @addr: VM_CONTEXT1_PROTECTION_FAULT_ADDR register value
 998  * @mc_client: VM_CONTEXT1_PROTECTION_FAULT_MCCLIENT register value
 999  * @pasid: debug logging only - no functional use
1000  *
1001  * Print human readable fault information (VI).
1002  */
1003 static void gmc_v8_0_vm_decode_fault(struct amdgpu_device *adev, u32 status,
1004                                      u32 addr, u32 mc_client, unsigned pasid)
1005 {
1006         u32 vmid = REG_GET_FIELD(status, VM_CONTEXT1_PROTECTION_FAULT_STATUS, VMID);
1007         u32 protections = REG_GET_FIELD(status, VM_CONTEXT1_PROTECTION_FAULT_STATUS,
1008                                         PROTECTIONS);
1009         char block[5] = { mc_client >> 24, (mc_client >> 16) & 0xff,
1010                 (mc_client >> 8) & 0xff, mc_client & 0xff, 0 };
1011         u32 mc_id;
1012
1013         mc_id = REG_GET_FIELD(status, VM_CONTEXT1_PROTECTION_FAULT_STATUS,
1014                               MEMORY_CLIENT_ID);
1015
1016         dev_err(adev->dev, "VM fault (0x%02x, vmid %d, pasid %d) at page %u, %s from '%s' (0x%08x) (%d)\n",
1017                protections, vmid, pasid, addr,
1018                REG_GET_FIELD(status, VM_CONTEXT1_PROTECTION_FAULT_STATUS,
1019                              MEMORY_CLIENT_RW) ?
1020                "write" : "read", block, mc_client, mc_id);
1021 }
1022
1023 static int gmc_v8_0_convert_vram_type(int mc_seq_vram_type)
1024 {
1025         switch (mc_seq_vram_type) {
1026         case MC_SEQ_MISC0__MT__GDDR1:
1027                 return AMDGPU_VRAM_TYPE_GDDR1;
1028         case MC_SEQ_MISC0__MT__DDR2:
1029                 return AMDGPU_VRAM_TYPE_DDR2;
1030         case MC_SEQ_MISC0__MT__GDDR3:
1031                 return AMDGPU_VRAM_TYPE_GDDR3;
1032         case MC_SEQ_MISC0__MT__GDDR4:
1033                 return AMDGPU_VRAM_TYPE_GDDR4;
1034         case MC_SEQ_MISC0__MT__GDDR5:
1035                 return AMDGPU_VRAM_TYPE_GDDR5;
1036         case MC_SEQ_MISC0__MT__HBM:
1037                 return AMDGPU_VRAM_TYPE_HBM;
1038         case MC_SEQ_MISC0__MT__DDR3:
1039                 return AMDGPU_VRAM_TYPE_DDR3;
1040         default:
1041                 return AMDGPU_VRAM_TYPE_UNKNOWN;
1042         }
1043 }
1044
1045 static int gmc_v8_0_early_init(void *handle)
1046 {
1047         struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1048
1049         gmc_v8_0_set_gmc_funcs(adev);
1050         gmc_v8_0_set_irq_funcs(adev);
1051
1052         adev->gmc.shared_aperture_start = 0x2000000000000000ULL;
1053         adev->gmc.shared_aperture_end =
1054                 adev->gmc.shared_aperture_start + (4ULL << 30) - 1;
1055         adev->gmc.private_aperture_start =
1056                 adev->gmc.shared_aperture_end + 1;
1057         adev->gmc.private_aperture_end =
1058                 adev->gmc.private_aperture_start + (4ULL << 30) - 1;
1059
1060         return 0;
1061 }
1062
1063 static int gmc_v8_0_late_init(void *handle)
1064 {
1065         struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1066
1067         if (amdgpu_vm_fault_stop != AMDGPU_VM_FAULT_STOP_ALWAYS)
1068                 return amdgpu_irq_get(adev, &adev->gmc.vm_fault, 0);
1069         else
1070                 return 0;
1071 }
1072
1073 static unsigned gmc_v8_0_get_vbios_fb_size(struct amdgpu_device *adev)
1074 {
1075         u32 d1vga_control = RREG32(mmD1VGA_CONTROL);
1076         unsigned size;
1077
1078         if (REG_GET_FIELD(d1vga_control, D1VGA_CONTROL, D1VGA_MODE_ENABLE)) {
1079                 size = AMDGPU_VBIOS_VGA_ALLOCATION;
1080         } else {
1081                 u32 viewport = RREG32(mmVIEWPORT_SIZE);
1082                 size = (REG_GET_FIELD(viewport, VIEWPORT_SIZE, VIEWPORT_HEIGHT) *
1083                         REG_GET_FIELD(viewport, VIEWPORT_SIZE, VIEWPORT_WIDTH) *
1084                         4);
1085         }
1086
1087         return size;
1088 }
1089
1090 #define mmMC_SEQ_MISC0_FIJI 0xA71
1091
1092 static int gmc_v8_0_sw_init(void *handle)
1093 {
1094         int r;
1095         struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1096
1097         set_bit(AMDGPU_GFXHUB(0), adev->vmhubs_mask);
1098
1099         if (adev->flags & AMD_IS_APU) {
1100                 adev->gmc.vram_type = AMDGPU_VRAM_TYPE_UNKNOWN;
1101         } else {
1102                 u32 tmp;
1103
1104                 if ((adev->asic_type == CHIP_FIJI) ||
1105                     (adev->asic_type == CHIP_VEGAM))
1106                         tmp = RREG32(mmMC_SEQ_MISC0_FIJI);
1107                 else
1108                         tmp = RREG32(mmMC_SEQ_MISC0);
1109                 tmp &= MC_SEQ_MISC0__MT__MASK;
1110                 adev->gmc.vram_type = gmc_v8_0_convert_vram_type(tmp);
1111         }
1112
1113         r = amdgpu_irq_add_id(adev, AMDGPU_IRQ_CLIENTID_LEGACY, VISLANDS30_IV_SRCID_GFX_PAGE_INV_FAULT, &adev->gmc.vm_fault);
1114         if (r)
1115                 return r;
1116
1117         r = amdgpu_irq_add_id(adev, AMDGPU_IRQ_CLIENTID_LEGACY, VISLANDS30_IV_SRCID_GFX_MEM_PROT_FAULT, &adev->gmc.vm_fault);
1118         if (r)
1119                 return r;
1120
1121         /* Adjust VM size here.
1122          * Currently set to 4GB ((1 << 20) 4k pages).
1123          * Max GPUVM size for cayman and SI is 40 bits.
1124          */
1125         amdgpu_vm_adjust_size(adev, 64, 9, 1, 40);
1126
1127         /* Set the internal MC address mask
1128          * This is the max address of the GPU's
1129          * internal address space.
1130          */
1131         adev->gmc.mc_mask = 0xffffffffffULL; /* 40 bit MC */
1132
1133         r = dma_set_mask_and_coherent(adev->dev, DMA_BIT_MASK(40));
1134         if (r) {
1135                 pr_warn("No suitable DMA available\n");
1136                 return r;
1137         }
1138         adev->need_swiotlb = drm_need_swiotlb(40);
1139
1140         r = gmc_v8_0_init_microcode(adev);
1141         if (r) {
1142                 DRM_ERROR("Failed to load mc firmware!\n");
1143                 return r;
1144         }
1145
1146         r = gmc_v8_0_mc_init(adev);
1147         if (r)
1148                 return r;
1149
1150         amdgpu_gmc_get_vbios_allocations(adev);
1151
1152         /* Memory manager */
1153         r = amdgpu_bo_init(adev);
1154         if (r)
1155                 return r;
1156
1157         r = gmc_v8_0_gart_init(adev);
1158         if (r)
1159                 return r;
1160
1161         /*
1162          * number of VMs
1163          * VMID 0 is reserved for System
1164          * amdgpu graphics/compute will use VMIDs 1-7
1165          * amdkfd will use VMIDs 8-15
1166          */
1167         adev->vm_manager.first_kfd_vmid = 8;
1168         amdgpu_vm_manager_init(adev);
1169
1170         /* base offset of vram pages */
1171         if (adev->flags & AMD_IS_APU) {
1172                 u64 tmp = RREG32(mmMC_VM_FB_OFFSET);
1173
1174                 tmp <<= 22;
1175                 adev->vm_manager.vram_base_offset = tmp;
1176         } else {
1177                 adev->vm_manager.vram_base_offset = 0;
1178         }
1179
1180         adev->gmc.vm_fault_info = kmalloc(sizeof(struct kfd_vm_fault_info),
1181                                         GFP_KERNEL);
1182         if (!adev->gmc.vm_fault_info)
1183                 return -ENOMEM;
1184         atomic_set(&adev->gmc.vm_fault_info_updated, 0);
1185
1186         return 0;
1187 }
1188
1189 static int gmc_v8_0_sw_fini(void *handle)
1190 {
1191         struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1192
1193         amdgpu_gem_force_release(adev);
1194         amdgpu_vm_manager_fini(adev);
1195         kfree(adev->gmc.vm_fault_info);
1196         amdgpu_gart_table_vram_free(adev);
1197         amdgpu_bo_fini(adev);
1198         amdgpu_ucode_release(&adev->gmc.fw);
1199
1200         return 0;
1201 }
1202
1203 static int gmc_v8_0_hw_init(void *handle)
1204 {
1205         int r;
1206         struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1207
1208         gmc_v8_0_init_golden_registers(adev);
1209
1210         gmc_v8_0_mc_program(adev);
1211
1212         if (adev->asic_type == CHIP_TONGA) {
1213                 r = gmc_v8_0_tonga_mc_load_microcode(adev);
1214                 if (r) {
1215                         DRM_ERROR("Failed to load MC firmware!\n");
1216                         return r;
1217                 }
1218         } else if (adev->asic_type == CHIP_POLARIS11 ||
1219                         adev->asic_type == CHIP_POLARIS10 ||
1220                         adev->asic_type == CHIP_POLARIS12) {
1221                 r = gmc_v8_0_polaris_mc_load_microcode(adev);
1222                 if (r) {
1223                         DRM_ERROR("Failed to load MC firmware!\n");
1224                         return r;
1225                 }
1226         }
1227
1228         r = gmc_v8_0_gart_enable(adev);
1229         if (r)
1230                 return r;
1231
1232         if (amdgpu_emu_mode == 1)
1233                 return amdgpu_gmc_vram_checking(adev);
1234         else
1235                 return r;
1236 }
1237
1238 static int gmc_v8_0_hw_fini(void *handle)
1239 {
1240         struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1241
1242         amdgpu_irq_put(adev, &adev->gmc.vm_fault, 0);
1243         gmc_v8_0_gart_disable(adev);
1244
1245         return 0;
1246 }
1247
1248 static int gmc_v8_0_suspend(void *handle)
1249 {
1250         struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1251
1252         gmc_v8_0_hw_fini(adev);
1253
1254         return 0;
1255 }
1256
1257 static int gmc_v8_0_resume(void *handle)
1258 {
1259         int r;
1260         struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1261
1262         r = gmc_v8_0_hw_init(adev);
1263         if (r)
1264                 return r;
1265
1266         amdgpu_vmid_reset_all(adev);
1267
1268         return 0;
1269 }
1270
1271 static bool gmc_v8_0_is_idle(void *handle)
1272 {
1273         struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1274         u32 tmp = RREG32(mmSRBM_STATUS);
1275
1276         if (tmp & (SRBM_STATUS__MCB_BUSY_MASK | SRBM_STATUS__MCB_NON_DISPLAY_BUSY_MASK |
1277                    SRBM_STATUS__MCC_BUSY_MASK | SRBM_STATUS__MCD_BUSY_MASK | SRBM_STATUS__VMC_BUSY_MASK))
1278                 return false;
1279
1280         return true;
1281 }
1282
1283 static int gmc_v8_0_wait_for_idle(void *handle)
1284 {
1285         unsigned i;
1286         u32 tmp;
1287         struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1288
1289         for (i = 0; i < adev->usec_timeout; i++) {
1290                 /* read MC_STATUS */
1291                 tmp = RREG32(mmSRBM_STATUS) & (SRBM_STATUS__MCB_BUSY_MASK |
1292                                                SRBM_STATUS__MCB_NON_DISPLAY_BUSY_MASK |
1293                                                SRBM_STATUS__MCC_BUSY_MASK |
1294                                                SRBM_STATUS__MCD_BUSY_MASK |
1295                                                SRBM_STATUS__VMC_BUSY_MASK |
1296                                                SRBM_STATUS__VMC1_BUSY_MASK);
1297                 if (!tmp)
1298                         return 0;
1299                 udelay(1);
1300         }
1301         return -ETIMEDOUT;
1302
1303 }
1304
1305 static bool gmc_v8_0_check_soft_reset(void *handle)
1306 {
1307         u32 srbm_soft_reset = 0;
1308         struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1309         u32 tmp = RREG32(mmSRBM_STATUS);
1310
1311         if (tmp & SRBM_STATUS__VMC_BUSY_MASK)
1312                 srbm_soft_reset = REG_SET_FIELD(srbm_soft_reset,
1313                                                 SRBM_SOFT_RESET, SOFT_RESET_VMC, 1);
1314
1315         if (tmp & (SRBM_STATUS__MCB_BUSY_MASK | SRBM_STATUS__MCB_NON_DISPLAY_BUSY_MASK |
1316                    SRBM_STATUS__MCC_BUSY_MASK | SRBM_STATUS__MCD_BUSY_MASK)) {
1317                 if (!(adev->flags & AMD_IS_APU))
1318                         srbm_soft_reset = REG_SET_FIELD(srbm_soft_reset,
1319                                                         SRBM_SOFT_RESET, SOFT_RESET_MC, 1);
1320         }
1321         if (srbm_soft_reset) {
1322                 adev->gmc.srbm_soft_reset = srbm_soft_reset;
1323                 return true;
1324         } else {
1325                 adev->gmc.srbm_soft_reset = 0;
1326                 return false;
1327         }
1328 }
1329
1330 static int gmc_v8_0_pre_soft_reset(void *handle)
1331 {
1332         struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1333
1334         if (!adev->gmc.srbm_soft_reset)
1335                 return 0;
1336
1337         gmc_v8_0_mc_stop(adev);
1338         if (gmc_v8_0_wait_for_idle(adev)) {
1339                 dev_warn(adev->dev, "Wait for GMC idle timed out !\n");
1340         }
1341
1342         return 0;
1343 }
1344
1345 static int gmc_v8_0_soft_reset(void *handle)
1346 {
1347         struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1348         u32 srbm_soft_reset;
1349
1350         if (!adev->gmc.srbm_soft_reset)
1351                 return 0;
1352         srbm_soft_reset = adev->gmc.srbm_soft_reset;
1353
1354         if (srbm_soft_reset) {
1355                 u32 tmp;
1356
1357                 tmp = RREG32(mmSRBM_SOFT_RESET);
1358                 tmp |= srbm_soft_reset;
1359                 dev_info(adev->dev, "SRBM_SOFT_RESET=0x%08X\n", tmp);
1360                 WREG32(mmSRBM_SOFT_RESET, tmp);
1361                 tmp = RREG32(mmSRBM_SOFT_RESET);
1362
1363                 udelay(50);
1364
1365                 tmp &= ~srbm_soft_reset;
1366                 WREG32(mmSRBM_SOFT_RESET, tmp);
1367                 tmp = RREG32(mmSRBM_SOFT_RESET);
1368
1369                 /* Wait a little for things to settle down */
1370                 udelay(50);
1371         }
1372
1373         return 0;
1374 }
1375
1376 static int gmc_v8_0_post_soft_reset(void *handle)
1377 {
1378         struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1379
1380         if (!adev->gmc.srbm_soft_reset)
1381                 return 0;
1382
1383         gmc_v8_0_mc_resume(adev);
1384         return 0;
1385 }
1386
1387 static int gmc_v8_0_vm_fault_interrupt_state(struct amdgpu_device *adev,
1388                                              struct amdgpu_irq_src *src,
1389                                              unsigned type,
1390                                              enum amdgpu_interrupt_state state)
1391 {
1392         u32 tmp;
1393         u32 bits = (VM_CONTEXT1_CNTL__RANGE_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
1394                     VM_CONTEXT1_CNTL__DUMMY_PAGE_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
1395                     VM_CONTEXT1_CNTL__PDE0_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
1396                     VM_CONTEXT1_CNTL__VALID_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
1397                     VM_CONTEXT1_CNTL__READ_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
1398                     VM_CONTEXT1_CNTL__WRITE_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
1399                     VM_CONTEXT1_CNTL__EXECUTE_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK);
1400
1401         switch (state) {
1402         case AMDGPU_IRQ_STATE_DISABLE:
1403                 /* system context */
1404                 tmp = RREG32(mmVM_CONTEXT0_CNTL);
1405                 tmp &= ~bits;
1406                 WREG32(mmVM_CONTEXT0_CNTL, tmp);
1407                 /* VMs */
1408                 tmp = RREG32(mmVM_CONTEXT1_CNTL);
1409                 tmp &= ~bits;
1410                 WREG32(mmVM_CONTEXT1_CNTL, tmp);
1411                 break;
1412         case AMDGPU_IRQ_STATE_ENABLE:
1413                 /* system context */
1414                 tmp = RREG32(mmVM_CONTEXT0_CNTL);
1415                 tmp |= bits;
1416                 WREG32(mmVM_CONTEXT0_CNTL, tmp);
1417                 /* VMs */
1418                 tmp = RREG32(mmVM_CONTEXT1_CNTL);
1419                 tmp |= bits;
1420                 WREG32(mmVM_CONTEXT1_CNTL, tmp);
1421                 break;
1422         default:
1423                 break;
1424         }
1425
1426         return 0;
1427 }
1428
1429 static int gmc_v8_0_process_interrupt(struct amdgpu_device *adev,
1430                                       struct amdgpu_irq_src *source,
1431                                       struct amdgpu_iv_entry *entry)
1432 {
1433         u32 addr, status, mc_client, vmid;
1434
1435         if (amdgpu_sriov_vf(adev)) {
1436                 dev_err(adev->dev, "GPU fault detected: %d 0x%08x\n",
1437                         entry->src_id, entry->src_data[0]);
1438                 dev_err(adev->dev, " Can't decode VM fault info here on SRIOV VF\n");
1439                 return 0;
1440         }
1441
1442         addr = RREG32(mmVM_CONTEXT1_PROTECTION_FAULT_ADDR);
1443         status = RREG32(mmVM_CONTEXT1_PROTECTION_FAULT_STATUS);
1444         mc_client = RREG32(mmVM_CONTEXT1_PROTECTION_FAULT_MCCLIENT);
1445         /* reset addr and status */
1446         WREG32_P(mmVM_CONTEXT1_CNTL2, 1, ~1);
1447
1448         if (!addr && !status)
1449                 return 0;
1450
1451         if (amdgpu_vm_fault_stop == AMDGPU_VM_FAULT_STOP_FIRST)
1452                 gmc_v8_0_set_fault_enable_default(adev, false);
1453
1454         if (printk_ratelimit()) {
1455                 struct amdgpu_task_info task_info;
1456
1457                 memset(&task_info, 0, sizeof(struct amdgpu_task_info));
1458                 amdgpu_vm_get_task_info(adev, entry->pasid, &task_info);
1459
1460                 dev_err(adev->dev, "GPU fault detected: %d 0x%08x for process %s pid %d thread %s pid %d\n",
1461                         entry->src_id, entry->src_data[0], task_info.process_name,
1462                         task_info.tgid, task_info.task_name, task_info.pid);
1463                 dev_err(adev->dev, "  VM_CONTEXT1_PROTECTION_FAULT_ADDR   0x%08X\n",
1464                         addr);
1465                 dev_err(adev->dev, "  VM_CONTEXT1_PROTECTION_FAULT_STATUS 0x%08X\n",
1466                         status);
1467                 gmc_v8_0_vm_decode_fault(adev, status, addr, mc_client,
1468                                          entry->pasid);
1469         }
1470
1471         vmid = REG_GET_FIELD(status, VM_CONTEXT1_PROTECTION_FAULT_STATUS,
1472                              VMID);
1473         if (amdgpu_amdkfd_is_kfd_vmid(adev, vmid)
1474                 && !atomic_read(&adev->gmc.vm_fault_info_updated)) {
1475                 struct kfd_vm_fault_info *info = adev->gmc.vm_fault_info;
1476                 u32 protections = REG_GET_FIELD(status,
1477                                         VM_CONTEXT1_PROTECTION_FAULT_STATUS,
1478                                         PROTECTIONS);
1479
1480                 info->vmid = vmid;
1481                 info->mc_id = REG_GET_FIELD(status,
1482                                             VM_CONTEXT1_PROTECTION_FAULT_STATUS,
1483                                             MEMORY_CLIENT_ID);
1484                 info->status = status;
1485                 info->page_addr = addr;
1486                 info->prot_valid = protections & 0x7 ? true : false;
1487                 info->prot_read = protections & 0x8 ? true : false;
1488                 info->prot_write = protections & 0x10 ? true : false;
1489                 info->prot_exec = protections & 0x20 ? true : false;
1490                 mb();
1491                 atomic_set(&adev->gmc.vm_fault_info_updated, 1);
1492         }
1493
1494         return 0;
1495 }
1496
1497 static void fiji_update_mc_medium_grain_clock_gating(struct amdgpu_device *adev,
1498                                                      bool enable)
1499 {
1500         uint32_t data;
1501
1502         if (enable && (adev->cg_flags & AMD_CG_SUPPORT_MC_MGCG)) {
1503                 data = RREG32(mmMC_HUB_MISC_HUB_CG);
1504                 data |= MC_HUB_MISC_HUB_CG__ENABLE_MASK;
1505                 WREG32(mmMC_HUB_MISC_HUB_CG, data);
1506
1507                 data = RREG32(mmMC_HUB_MISC_SIP_CG);
1508                 data |= MC_HUB_MISC_SIP_CG__ENABLE_MASK;
1509                 WREG32(mmMC_HUB_MISC_SIP_CG, data);
1510
1511                 data = RREG32(mmMC_HUB_MISC_VM_CG);
1512                 data |= MC_HUB_MISC_VM_CG__ENABLE_MASK;
1513                 WREG32(mmMC_HUB_MISC_VM_CG, data);
1514
1515                 data = RREG32(mmMC_XPB_CLK_GAT);
1516                 data |= MC_XPB_CLK_GAT__ENABLE_MASK;
1517                 WREG32(mmMC_XPB_CLK_GAT, data);
1518
1519                 data = RREG32(mmATC_MISC_CG);
1520                 data |= ATC_MISC_CG__ENABLE_MASK;
1521                 WREG32(mmATC_MISC_CG, data);
1522
1523                 data = RREG32(mmMC_CITF_MISC_WR_CG);
1524                 data |= MC_CITF_MISC_WR_CG__ENABLE_MASK;
1525                 WREG32(mmMC_CITF_MISC_WR_CG, data);
1526
1527                 data = RREG32(mmMC_CITF_MISC_RD_CG);
1528                 data |= MC_CITF_MISC_RD_CG__ENABLE_MASK;
1529                 WREG32(mmMC_CITF_MISC_RD_CG, data);
1530
1531                 data = RREG32(mmMC_CITF_MISC_VM_CG);
1532                 data |= MC_CITF_MISC_VM_CG__ENABLE_MASK;
1533                 WREG32(mmMC_CITF_MISC_VM_CG, data);
1534
1535                 data = RREG32(mmVM_L2_CG);
1536                 data |= VM_L2_CG__ENABLE_MASK;
1537                 WREG32(mmVM_L2_CG, data);
1538         } else {
1539                 data = RREG32(mmMC_HUB_MISC_HUB_CG);
1540                 data &= ~MC_HUB_MISC_HUB_CG__ENABLE_MASK;
1541                 WREG32(mmMC_HUB_MISC_HUB_CG, data);
1542
1543                 data = RREG32(mmMC_HUB_MISC_SIP_CG);
1544                 data &= ~MC_HUB_MISC_SIP_CG__ENABLE_MASK;
1545                 WREG32(mmMC_HUB_MISC_SIP_CG, data);
1546
1547                 data = RREG32(mmMC_HUB_MISC_VM_CG);
1548                 data &= ~MC_HUB_MISC_VM_CG__ENABLE_MASK;
1549                 WREG32(mmMC_HUB_MISC_VM_CG, data);
1550
1551                 data = RREG32(mmMC_XPB_CLK_GAT);
1552                 data &= ~MC_XPB_CLK_GAT__ENABLE_MASK;
1553                 WREG32(mmMC_XPB_CLK_GAT, data);
1554
1555                 data = RREG32(mmATC_MISC_CG);
1556                 data &= ~ATC_MISC_CG__ENABLE_MASK;
1557                 WREG32(mmATC_MISC_CG, data);
1558
1559                 data = RREG32(mmMC_CITF_MISC_WR_CG);
1560                 data &= ~MC_CITF_MISC_WR_CG__ENABLE_MASK;
1561                 WREG32(mmMC_CITF_MISC_WR_CG, data);
1562
1563                 data = RREG32(mmMC_CITF_MISC_RD_CG);
1564                 data &= ~MC_CITF_MISC_RD_CG__ENABLE_MASK;
1565                 WREG32(mmMC_CITF_MISC_RD_CG, data);
1566
1567                 data = RREG32(mmMC_CITF_MISC_VM_CG);
1568                 data &= ~MC_CITF_MISC_VM_CG__ENABLE_MASK;
1569                 WREG32(mmMC_CITF_MISC_VM_CG, data);
1570
1571                 data = RREG32(mmVM_L2_CG);
1572                 data &= ~VM_L2_CG__ENABLE_MASK;
1573                 WREG32(mmVM_L2_CG, data);
1574         }
1575 }
1576
1577 static void fiji_update_mc_light_sleep(struct amdgpu_device *adev,
1578                                        bool enable)
1579 {
1580         uint32_t data;
1581
1582         if (enable && (adev->cg_flags & AMD_CG_SUPPORT_MC_LS)) {
1583                 data = RREG32(mmMC_HUB_MISC_HUB_CG);
1584                 data |= MC_HUB_MISC_HUB_CG__MEM_LS_ENABLE_MASK;
1585                 WREG32(mmMC_HUB_MISC_HUB_CG, data);
1586
1587                 data = RREG32(mmMC_HUB_MISC_SIP_CG);
1588                 data |= MC_HUB_MISC_SIP_CG__MEM_LS_ENABLE_MASK;
1589                 WREG32(mmMC_HUB_MISC_SIP_CG, data);
1590
1591                 data = RREG32(mmMC_HUB_MISC_VM_CG);
1592                 data |= MC_HUB_MISC_VM_CG__MEM_LS_ENABLE_MASK;
1593                 WREG32(mmMC_HUB_MISC_VM_CG, data);
1594
1595                 data = RREG32(mmMC_XPB_CLK_GAT);
1596                 data |= MC_XPB_CLK_GAT__MEM_LS_ENABLE_MASK;
1597                 WREG32(mmMC_XPB_CLK_GAT, data);
1598
1599                 data = RREG32(mmATC_MISC_CG);
1600                 data |= ATC_MISC_CG__MEM_LS_ENABLE_MASK;
1601                 WREG32(mmATC_MISC_CG, data);
1602
1603                 data = RREG32(mmMC_CITF_MISC_WR_CG);
1604                 data |= MC_CITF_MISC_WR_CG__MEM_LS_ENABLE_MASK;
1605                 WREG32(mmMC_CITF_MISC_WR_CG, data);
1606
1607                 data = RREG32(mmMC_CITF_MISC_RD_CG);
1608                 data |= MC_CITF_MISC_RD_CG__MEM_LS_ENABLE_MASK;
1609                 WREG32(mmMC_CITF_MISC_RD_CG, data);
1610
1611                 data = RREG32(mmMC_CITF_MISC_VM_CG);
1612                 data |= MC_CITF_MISC_VM_CG__MEM_LS_ENABLE_MASK;
1613                 WREG32(mmMC_CITF_MISC_VM_CG, data);
1614
1615                 data = RREG32(mmVM_L2_CG);
1616                 data |= VM_L2_CG__MEM_LS_ENABLE_MASK;
1617                 WREG32(mmVM_L2_CG, data);
1618         } else {
1619                 data = RREG32(mmMC_HUB_MISC_HUB_CG);
1620                 data &= ~MC_HUB_MISC_HUB_CG__MEM_LS_ENABLE_MASK;
1621                 WREG32(mmMC_HUB_MISC_HUB_CG, data);
1622
1623                 data = RREG32(mmMC_HUB_MISC_SIP_CG);
1624                 data &= ~MC_HUB_MISC_SIP_CG__MEM_LS_ENABLE_MASK;
1625                 WREG32(mmMC_HUB_MISC_SIP_CG, data);
1626
1627                 data = RREG32(mmMC_HUB_MISC_VM_CG);
1628                 data &= ~MC_HUB_MISC_VM_CG__MEM_LS_ENABLE_MASK;
1629                 WREG32(mmMC_HUB_MISC_VM_CG, data);
1630
1631                 data = RREG32(mmMC_XPB_CLK_GAT);
1632                 data &= ~MC_XPB_CLK_GAT__MEM_LS_ENABLE_MASK;
1633                 WREG32(mmMC_XPB_CLK_GAT, data);
1634
1635                 data = RREG32(mmATC_MISC_CG);
1636                 data &= ~ATC_MISC_CG__MEM_LS_ENABLE_MASK;
1637                 WREG32(mmATC_MISC_CG, data);
1638
1639                 data = RREG32(mmMC_CITF_MISC_WR_CG);
1640                 data &= ~MC_CITF_MISC_WR_CG__MEM_LS_ENABLE_MASK;
1641                 WREG32(mmMC_CITF_MISC_WR_CG, data);
1642
1643                 data = RREG32(mmMC_CITF_MISC_RD_CG);
1644                 data &= ~MC_CITF_MISC_RD_CG__MEM_LS_ENABLE_MASK;
1645                 WREG32(mmMC_CITF_MISC_RD_CG, data);
1646
1647                 data = RREG32(mmMC_CITF_MISC_VM_CG);
1648                 data &= ~MC_CITF_MISC_VM_CG__MEM_LS_ENABLE_MASK;
1649                 WREG32(mmMC_CITF_MISC_VM_CG, data);
1650
1651                 data = RREG32(mmVM_L2_CG);
1652                 data &= ~VM_L2_CG__MEM_LS_ENABLE_MASK;
1653                 WREG32(mmVM_L2_CG, data);
1654         }
1655 }
1656
1657 static int gmc_v8_0_set_clockgating_state(void *handle,
1658                                           enum amd_clockgating_state state)
1659 {
1660         struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1661
1662         if (amdgpu_sriov_vf(adev))
1663                 return 0;
1664
1665         switch (adev->asic_type) {
1666         case CHIP_FIJI:
1667                 fiji_update_mc_medium_grain_clock_gating(adev,
1668                                 state == AMD_CG_STATE_GATE);
1669                 fiji_update_mc_light_sleep(adev,
1670                                 state == AMD_CG_STATE_GATE);
1671                 break;
1672         default:
1673                 break;
1674         }
1675         return 0;
1676 }
1677
1678 static int gmc_v8_0_set_powergating_state(void *handle,
1679                                           enum amd_powergating_state state)
1680 {
1681         return 0;
1682 }
1683
1684 static void gmc_v8_0_get_clockgating_state(void *handle, u64 *flags)
1685 {
1686         struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1687         int data;
1688
1689         if (amdgpu_sriov_vf(adev))
1690                 *flags = 0;
1691
1692         /* AMD_CG_SUPPORT_MC_MGCG */
1693         data = RREG32(mmMC_HUB_MISC_HUB_CG);
1694         if (data & MC_HUB_MISC_HUB_CG__ENABLE_MASK)
1695                 *flags |= AMD_CG_SUPPORT_MC_MGCG;
1696
1697         /* AMD_CG_SUPPORT_MC_LS */
1698         if (data & MC_HUB_MISC_HUB_CG__MEM_LS_ENABLE_MASK)
1699                 *flags |= AMD_CG_SUPPORT_MC_LS;
1700 }
1701
1702 static const struct amd_ip_funcs gmc_v8_0_ip_funcs = {
1703         .name = "gmc_v8_0",
1704         .early_init = gmc_v8_0_early_init,
1705         .late_init = gmc_v8_0_late_init,
1706         .sw_init = gmc_v8_0_sw_init,
1707         .sw_fini = gmc_v8_0_sw_fini,
1708         .hw_init = gmc_v8_0_hw_init,
1709         .hw_fini = gmc_v8_0_hw_fini,
1710         .suspend = gmc_v8_0_suspend,
1711         .resume = gmc_v8_0_resume,
1712         .is_idle = gmc_v8_0_is_idle,
1713         .wait_for_idle = gmc_v8_0_wait_for_idle,
1714         .check_soft_reset = gmc_v8_0_check_soft_reset,
1715         .pre_soft_reset = gmc_v8_0_pre_soft_reset,
1716         .soft_reset = gmc_v8_0_soft_reset,
1717         .post_soft_reset = gmc_v8_0_post_soft_reset,
1718         .set_clockgating_state = gmc_v8_0_set_clockgating_state,
1719         .set_powergating_state = gmc_v8_0_set_powergating_state,
1720         .get_clockgating_state = gmc_v8_0_get_clockgating_state,
1721 };
1722
1723 static const struct amdgpu_gmc_funcs gmc_v8_0_gmc_funcs = {
1724         .flush_gpu_tlb = gmc_v8_0_flush_gpu_tlb,
1725         .flush_gpu_tlb_pasid = gmc_v8_0_flush_gpu_tlb_pasid,
1726         .emit_flush_gpu_tlb = gmc_v8_0_emit_flush_gpu_tlb,
1727         .emit_pasid_mapping = gmc_v8_0_emit_pasid_mapping,
1728         .set_prt = gmc_v8_0_set_prt,
1729         .get_vm_pde = gmc_v8_0_get_vm_pde,
1730         .get_vm_pte = gmc_v8_0_get_vm_pte,
1731         .get_vbios_fb_size = gmc_v8_0_get_vbios_fb_size,
1732 };
1733
1734 static const struct amdgpu_irq_src_funcs gmc_v8_0_irq_funcs = {
1735         .set = gmc_v8_0_vm_fault_interrupt_state,
1736         .process = gmc_v8_0_process_interrupt,
1737 };
1738
1739 static void gmc_v8_0_set_gmc_funcs(struct amdgpu_device *adev)
1740 {
1741         adev->gmc.gmc_funcs = &gmc_v8_0_gmc_funcs;
1742 }
1743
1744 static void gmc_v8_0_set_irq_funcs(struct amdgpu_device *adev)
1745 {
1746         adev->gmc.vm_fault.num_types = 1;
1747         adev->gmc.vm_fault.funcs = &gmc_v8_0_irq_funcs;
1748 }
1749
1750 const struct amdgpu_ip_block_version gmc_v8_0_ip_block =
1751 {
1752         .type = AMD_IP_BLOCK_TYPE_GMC,
1753         .major = 8,
1754         .minor = 0,
1755         .rev = 0,
1756         .funcs = &gmc_v8_0_ip_funcs,
1757 };
1758
1759 const struct amdgpu_ip_block_version gmc_v8_1_ip_block =
1760 {
1761         .type = AMD_IP_BLOCK_TYPE_GMC,
1762         .major = 8,
1763         .minor = 1,
1764         .rev = 0,
1765         .funcs = &gmc_v8_0_ip_funcs,
1766 };
1767
1768 const struct amdgpu_ip_block_version gmc_v8_5_ip_block =
1769 {
1770         .type = AMD_IP_BLOCK_TYPE_GMC,
1771         .major = 8,
1772         .minor = 5,
1773         .rev = 0,
1774         .funcs = &gmc_v8_0_ip_funcs,
1775 };