Quelle  mmhub_v4_1_0.c   Sprache: C

/*
 * Copyright 2023 Advanced Micro Devices, Inc.
 *
 * Permission is hereby granted, free of charge, to any person obtaining a
 * copy of this software and associated documentation files (the "Software"),
 * to deal in the Software without restriction, including without limitation
 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
 * and/or sell copies of the Software, and to permit persons to whom the
 * Software is furnished to do so, subject to the following conditions:
 *
 * The above copyright notice and this permission notice shall be included in
 * all copies or substantial portions of the Software.
 *
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
 * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
 * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
 * OTHER DEALINGS IN THE SOFTWARE.
 *
 */

#include "amdgpu.h"
#include "mmhub_v4_1_0.h"

#include "mmhub/mmhub_4_1_0_offset.h"
#include "mmhub/mmhub_4_1_0_sh_mask.h"

#include "soc15_common.h"
#include "soc24_enum.h"

#define regMMVM_L2_CNTL3_DEFAULT    0x80100007
#define regMMVM_L2_CNTL4_DEFAULT    0x000000c1
#define regMMVM_L2_CNTL5_DEFAULT    0x00003fe0

static const char *mmhub_client_ids_v4_1_0[][2] = {
 [0][0] = "VMC",
 [4][0] = "DCEDMC",
 [6][0] = "MP0",
 [7][0] = "MP1",
 [8][0] = "MPIO",
 [16][0] = "LSDMA",
 [17][0] = "JPEG",
 [19][0] = "VCNU",
 [22][0] = "VSCH",
 [23][0] = "HDP",
 [32+23][0] = "VCNRD",
 [3][1] = "DCEDWB",
 [4][1] = "DCEDMC",
 [6][1] = "MP0",
 [7][1] = "MP1",
 [8][1] = "MPIO",
 [10][1] = "DBGU0",
 [11][1] = "DBGU1",
 [12][1] = "DBGUNBIO",
 [14][1] = "XDP",
 [15][1] = "OSSSYS",
 [16][1] = "LSDMA",
 [17][1] = "JPEG",
 [18][1] = "VCNWR",
 [19][1] = "VCNU",
 [22][1] = "VSCH",
 [23][1] = "HDP",
};

static uint32_t mmhub_v4_1_0_get_invalidate_req(unsigned int vmid,
      uint32_t flush_type)
{
 u32 req = 0;

 /* invalidate using legacy mode on vmid*/
 req = REG_SET_FIELD(req, MMVM_INVALIDATE_ENG0_REQ,
       PER_VMID_INVALIDATE_REQ, 1 << vmid);
 /* Only use legacy inv on mmhub side */
 req = REG_SET_FIELD(req, MMVM_INVALIDATE_ENG0_REQ, FLUSH_TYPE, 0);
 req = REG_SET_FIELD(req, MMVM_INVALIDATE_ENG0_REQ, INVALIDATE_L2_PTES, 1);
 req = REG_SET_FIELD(req, MMVM_INVALIDATE_ENG0_REQ, INVALIDATE_L2_PDE0, 1);
 req = REG_SET_FIELD(req, MMVM_INVALIDATE_ENG0_REQ, INVALIDATE_L2_PDE1, 1);
 req = REG_SET_FIELD(req, MMVM_INVALIDATE_ENG0_REQ, INVALIDATE_L2_PDE2, 1);
 req = REG_SET_FIELD(req, MMVM_INVALIDATE_ENG0_REQ, INVALIDATE_L1_PTES, 1);
 req = REG_SET_FIELD(req, MMVM_INVALIDATE_ENG0_REQ,
       CLEAR_PROTECTION_FAULT_STATUS_ADDR, 0);

 return req;
}

static void
mmhub_v4_1_0_print_l2_protection_fault_status(struct amdgpu_device *adev,
           uint32_t status)
{
 uint32_t cid, rw;
 const char *mmhub_cid = NULL;

 cid = REG_GET_FIELD(status,
       MMVM_L2_PROTECTION_FAULT_STATUS_LO32, CID);
 rw = REG_GET_FIELD(status,
      MMVM_L2_PROTECTION_FAULT_STATUS_LO32, RW);

 dev_err(adev->dev,
  "MMVM_L2_PROTECTION_FAULT_STATUS_LO32:0x%08X\n",
  status);
 switch (amdgpu_ip_version(adev, MMHUB_HWIP, 0)) {
 case IP_VERSION(4, 1, 0):
  mmhub_cid = mmhub_client_ids_v4_1_0[cid][rw];
  break;
 default:
  mmhub_cid = NULL;
  break;
 }
 dev_err(adev->dev, "\t Faulty UTCL2 client ID: %s (0x%x)\n",
  mmhub_cid ? mmhub_cid : "unknown", cid);
 dev_err(adev->dev, "\t MORE_FAULTS: 0x%lx\n",
  REG_GET_FIELD(status,
  MMVM_L2_PROTECTION_FAULT_STATUS_LO32, MORE_FAULTS));
 dev_err(adev->dev, "\t WALKER_ERROR: 0x%lx\n",
  REG_GET_FIELD(status,
  MMVM_L2_PROTECTION_FAULT_STATUS_LO32, WALKER_ERROR));
 dev_err(adev->dev, "\t PERMISSION_FAULTS: 0x%lx\n",
  REG_GET_FIELD(status,
  MMVM_L2_PROTECTION_FAULT_STATUS_LO32, PERMISSION_FAULTS));
 dev_err(adev->dev, "\t MAPPING_ERROR: 0x%lx\n",
  REG_GET_FIELD(status,
  MMVM_L2_PROTECTION_FAULT_STATUS_LO32, MAPPING_ERROR));
 dev_err(adev->dev, "\t RW: 0x%x\n", rw);
}

static void mmhub_v4_1_0_setup_vm_pt_regs(struct amdgpu_device *adev,
       uint32_t vmid, uint64_t page_table_base)
{
 struct amdgpu_vmhub *hub = &adev->vmhub[AMDGPU_MMHUB0(0)];

 WREG32_SOC15_OFFSET(MMHUB, 0, regMMVM_CONTEXT0_PAGE_TABLE_BASE_ADDR_LO32,
       hub->ctx_addr_distance * vmid,
       lower_32_bits(page_table_base));

 WREG32_SOC15_OFFSET(MMHUB, 0, regMMVM_CONTEXT0_PAGE_TABLE_BASE_ADDR_HI32,
       hub->ctx_addr_distance * vmid,
       upper_32_bits(page_table_base));
}

static void mmhub_v4_1_0_init_gart_aperture_regs(struct amdgpu_device *adev)
{
 uint64_t pt_base = amdgpu_gmc_pd_addr(adev->gart.bo);

 mmhub_v4_1_0_setup_vm_pt_regs(adev, 0, pt_base);

 WREG32_SOC15(MMHUB, 0, regMMVM_CONTEXT0_PAGE_TABLE_START_ADDR_LO32,
       (u32)(adev->gmc.gart_start >> 12));
 WREG32_SOC15(MMHUB, 0, regMMVM_CONTEXT0_PAGE_TABLE_START_ADDR_HI32,
       (u32)(adev->gmc.gart_start >> 44));

 WREG32_SOC15(MMHUB, 0, regMMVM_CONTEXT0_PAGE_TABLE_END_ADDR_LO32,
       (u32)(adev->gmc.gart_end >> 12));
 WREG32_SOC15(MMHUB, 0, regMMVM_CONTEXT0_PAGE_TABLE_END_ADDR_HI32,
       (u32)(adev->gmc.gart_end >> 44));
}

static void mmhub_v4_1_0_init_system_aperture_regs(struct amdgpu_device *adev)
{
 uint64_t value;
 uint32_t tmp;

 /*
 * the new L1 policy will block SRIOV guest from writing
 * these regs, and they will be programed at host.
 * so skip programing these regs.
 */
 if (amdgpu_sriov_vf(adev))
  return;

 /* Program the AGP BAR */
 WREG32_SOC15(MMHUB, 0, regMMMC_VM_AGP_BASE, 0);
 WREG32_SOC15(MMHUB, 0, regMMMC_VM_AGP_BOT, adev->gmc.agp_start >> 24);
 WREG32_SOC15(MMHUB, 0, regMMMC_VM_AGP_TOP, adev->gmc.agp_end >> 24);

 /* Program the system aperture low logical page number. */
 WREG32_SOC15(MMHUB, 0, regMMMC_VM_SYSTEM_APERTURE_LOW_ADDR,
       min(adev->gmc.fb_start, adev->gmc.agp_start) >> 18);
 WREG32_SOC15(MMHUB, 0, regMMMC_VM_SYSTEM_APERTURE_HIGH_ADDR,
       max(adev->gmc.fb_end, adev->gmc.agp_end) >> 18);

 /* Set default page address. */
 value = adev->mem_scratch.gpu_addr - adev->gmc.vram_start +
  adev->vm_manager.vram_base_offset;
 WREG32_SOC15(MMHUB, 0, regMMMC_VM_SYSTEM_APERTURE_DEFAULT_ADDR_LSB,
       (u32)(value >> 12));
 WREG32_SOC15(MMHUB, 0, regMMMC_VM_SYSTEM_APERTURE_DEFAULT_ADDR_MSB,
       (u32)(value >> 44));

 /* Program "protection fault". */
 WREG32_SOC15(MMHUB, 0, regMMVM_L2_PROTECTION_FAULT_DEFAULT_ADDR_LO32,
       (u32)(adev->dummy_page_addr >> 12));
 WREG32_SOC15(MMHUB, 0, regMMVM_L2_PROTECTION_FAULT_DEFAULT_ADDR_HI32,
       (u32)((u64)adev->dummy_page_addr >> 44));

 tmp = RREG32_SOC15(MMHUB, 0, regMMVM_L2_PROTECTION_FAULT_CNTL2);
 tmp = REG_SET_FIELD(tmp, MMVM_L2_PROTECTION_FAULT_CNTL2,
       ACTIVE_PAGE_MIGRATION_PTE_READ_RETRY, 1);
 WREG32_SOC15(MMHUB, 0, regMMVM_L2_PROTECTION_FAULT_CNTL2, tmp);
}

static void mmhub_v4_1_0_init_tlb_regs(struct amdgpu_device *adev)
{
 uint32_t tmp;

 /* Setup TLB control */
 tmp = RREG32_SOC15(MMHUB, 0, regMMMC_VM_MX_L1_TLB_CNTL);

 tmp = REG_SET_FIELD(tmp, MMMC_VM_MX_L1_TLB_CNTL, ENABLE_L1_TLB, 1);
 tmp = REG_SET_FIELD(tmp, MMMC_VM_MX_L1_TLB_CNTL, SYSTEM_ACCESS_MODE, 3);
 tmp = REG_SET_FIELD(tmp, MMMC_VM_MX_L1_TLB_CNTL,
       ENABLE_ADVANCED_DRIVER_MODEL, 1);
 tmp = REG_SET_FIELD(tmp, MMMC_VM_MX_L1_TLB_CNTL,
       SYSTEM_APERTURE_UNMAPPED_ACCESS, 0);
 tmp = REG_SET_FIELD(tmp, MMMC_VM_MX_L1_TLB_CNTL, ECO_BITS, 0);
 tmp = REG_SET_FIELD(tmp, MMMC_VM_MX_L1_TLB_CNTL,
       MTYPE, MTYPE_UC); /* UC, uncached */

 WREG32_SOC15(MMHUB, 0, regMMMC_VM_MX_L1_TLB_CNTL, tmp);
}

static void mmhub_v4_1_0_init_cache_regs(struct amdgpu_device *adev)
{
 uint32_t tmp;

 /* These registers are not accessible to VF-SRIOV.
 * The PF will program them instead.
 */
 if (amdgpu_sriov_vf(adev))
  return;

 /* Setup L2 cache */
 tmp = RREG32_SOC15(MMHUB, 0, regMMVM_L2_CNTL);
 tmp = REG_SET_FIELD(tmp, MMVM_L2_CNTL, ENABLE_L2_CACHE, 1);
 tmp = REG_SET_FIELD(tmp, MMVM_L2_CNTL, ENABLE_L2_FRAGMENT_PROCESSING, 0);
 tmp = REG_SET_FIELD(tmp, MMVM_L2_CNTL,
       ENABLE_DEFAULT_PAGE_OUT_TO_SYSTEM_MEMORY, 1);
 /* XXX for emulation, Refer to closed source code.*/
 tmp = REG_SET_FIELD(tmp, MMVM_L2_CNTL, L2_PDE0_CACHE_TAG_GENERATION_MODE,
       0);
 tmp = REG_SET_FIELD(tmp, MMVM_L2_CNTL, PDE_FAULT_CLASSIFICATION, 0);
 tmp = REG_SET_FIELD(tmp, MMVM_L2_CNTL, CONTEXT1_IDENTITY_ACCESS_MODE, 1);
 tmp = REG_SET_FIELD(tmp, MMVM_L2_CNTL, IDENTITY_MODE_FRAGMENT_SIZE, 0);
 WREG32_SOC15(MMHUB, 0, regMMVM_L2_CNTL, tmp);

 tmp = RREG32_SOC15(MMHUB, 0, regMMVM_L2_CNTL2);
 tmp = REG_SET_FIELD(tmp, MMVM_L2_CNTL2, INVALIDATE_ALL_L1_TLBS, 1);
 tmp = REG_SET_FIELD(tmp, MMVM_L2_CNTL2, INVALIDATE_L2_CACHE, 1);
 WREG32_SOC15(MMHUB, 0, regMMVM_L2_CNTL2, tmp);

 tmp = regMMVM_L2_CNTL3_DEFAULT;
 if (adev->gmc.translate_further) {
  tmp = REG_SET_FIELD(tmp, MMVM_L2_CNTL3, BANK_SELECT, 12);
  tmp = REG_SET_FIELD(tmp, MMVM_L2_CNTL3,
        L2_CACHE_BIGK_FRAGMENT_SIZE, 9);
 } else {
  tmp = REG_SET_FIELD(tmp, MMVM_L2_CNTL3, BANK_SELECT, 9);
  tmp = REG_SET_FIELD(tmp, MMVM_L2_CNTL3,
        L2_CACHE_BIGK_FRAGMENT_SIZE, 6);
 }
 WREG32_SOC15(MMHUB, 0, regMMVM_L2_CNTL3, tmp);

 tmp = regMMVM_L2_CNTL4_DEFAULT;
 tmp = REG_SET_FIELD(tmp, MMVM_L2_CNTL4, VMC_TAP_PDE_REQUEST_PHYSICAL, 0);
 tmp = REG_SET_FIELD(tmp, MMVM_L2_CNTL4, VMC_TAP_PTE_REQUEST_PHYSICAL, 0);
 WREG32_SOC15(MMHUB, 0, regMMVM_L2_CNTL4, tmp);

 tmp = regMMVM_L2_CNTL5_DEFAULT;
 tmp = REG_SET_FIELD(tmp, MMVM_L2_CNTL5, L2_CACHE_SMALLK_FRAGMENT_SIZE, 0);
 WREG32_SOC15(MMHUB, 0, regMMVM_L2_CNTL5, tmp);
}

static void mmhub_v4_1_0_enable_system_domain(struct amdgpu_device *adev)
{
 uint32_t tmp;

 tmp = RREG32_SOC15(MMHUB, 0, regMMVM_CONTEXT0_CNTL);
 tmp = REG_SET_FIELD(tmp, MMVM_CONTEXT0_CNTL, ENABLE_CONTEXT, 1);
 tmp = REG_SET_FIELD(tmp, MMVM_CONTEXT0_CNTL, PAGE_TABLE_DEPTH, 0);
 tmp = REG_SET_FIELD(tmp, MMVM_CONTEXT0_CNTL,
       RETRY_PERMISSION_OR_INVALID_PAGE_FAULT, 0);
 WREG32_SOC15(MMHUB, 0, regMMVM_CONTEXT0_CNTL, tmp);
}

static void mmhub_v4_1_0_disable_identity_aperture(struct amdgpu_device *adev)
{
 /* These registers are not accessible to VF-SRIOV.
 * The PF will program them instead.
 */
 if (amdgpu_sriov_vf(adev))
  return;

 WREG32_SOC15(MMHUB, 0,
       regMMVM_L2_CONTEXT1_IDENTITY_APERTURE_LOW_ADDR_LO32,
       0xFFFFFFFF);
 WREG32_SOC15(MMHUB, 0,
       regMMVM_L2_CONTEXT1_IDENTITY_APERTURE_LOW_ADDR_HI32,
       0x0000000F);

 WREG32_SOC15(MMHUB, 0,
       regMMVM_L2_CONTEXT1_IDENTITY_APERTURE_HIGH_ADDR_LO32, 0);
 WREG32_SOC15(MMHUB, 0,
       regMMVM_L2_CONTEXT1_IDENTITY_APERTURE_HIGH_ADDR_HI32, 0);

 WREG32_SOC15(MMHUB, 0, regMMVM_L2_CONTEXT_IDENTITY_PHYSICAL_OFFSET_LO32,
       0);
 WREG32_SOC15(MMHUB, 0, regMMVM_L2_CONTEXT_IDENTITY_PHYSICAL_OFFSET_HI32,
       0);
}

static void mmhub_v4_1_0_setup_vmid_config(struct amdgpu_device *adev)
{
 struct amdgpu_vmhub *hub = &adev->vmhub[AMDGPU_MMHUB0(0)];
 int i;
 uint32_t tmp;

 for (i = 0; i <= 14; i++) {
  tmp = RREG32_SOC15_OFFSET(MMHUB, 0, regMMVM_CONTEXT1_CNTL, i);
  tmp = REG_SET_FIELD(tmp, MMVM_CONTEXT1_CNTL, ENABLE_CONTEXT, 1);
  tmp = REG_SET_FIELD(tmp, MMVM_CONTEXT1_CNTL, PAGE_TABLE_DEPTH,
        adev->vm_manager.num_level);
  tmp = REG_SET_FIELD(tmp, MMVM_CONTEXT1_CNTL,
        RANGE_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
  tmp = REG_SET_FIELD(tmp, MMVM_CONTEXT1_CNTL,
        DUMMY_PAGE_PROTECTION_FAULT_ENABLE_DEFAULT,
        1);
  tmp = REG_SET_FIELD(tmp, MMVM_CONTEXT1_CNTL,
        PDE0_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
  tmp = REG_SET_FIELD(tmp, MMVM_CONTEXT1_CNTL,
        VALID_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
  tmp = REG_SET_FIELD(tmp, MMVM_CONTEXT1_CNTL,
        READ_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
  tmp = REG_SET_FIELD(tmp, MMVM_CONTEXT1_CNTL,
        WRITE_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
  tmp = REG_SET_FIELD(tmp, MMVM_CONTEXT1_CNTL,
        EXECUTE_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
  tmp = REG_SET_FIELD(tmp, MMVM_CONTEXT1_CNTL,
        PAGE_TABLE_BLOCK_SIZE,
        adev->vm_manager.block_size - 9);
  /* Send no-retry XNACK on fault to suppress VM fault storm. */
  tmp = REG_SET_FIELD(tmp, MMVM_CONTEXT1_CNTL,
        RETRY_PERMISSION_OR_INVALID_PAGE_FAULT,
        !amdgpu_noretry);
  WREG32_SOC15_OFFSET(MMHUB, 0, regMMVM_CONTEXT1_CNTL,
        i * hub->ctx_distance, tmp);
  WREG32_SOC15_OFFSET(MMHUB, 0, regMMVM_CONTEXT1_PAGE_TABLE_START_ADDR_LO32,
        i * hub->ctx_addr_distance, 0);
  WREG32_SOC15_OFFSET(MMHUB, 0, regMMVM_CONTEXT1_PAGE_TABLE_START_ADDR_HI32,
        i * hub->ctx_addr_distance, 0);
  WREG32_SOC15_OFFSET(MMHUB, 0, regMMVM_CONTEXT1_PAGE_TABLE_END_ADDR_LO32,
        i * hub->ctx_addr_distance,
        lower_32_bits(adev->vm_manager.max_pfn - 1));
  WREG32_SOC15_OFFSET(MMHUB, 0, regMMVM_CONTEXT1_PAGE_TABLE_END_ADDR_HI32,
        i * hub->ctx_addr_distance,
        upper_32_bits(adev->vm_manager.max_pfn - 1));
 }

 hub->vm_cntx_cntl = tmp;
}

static void mmhub_v4_1_0_program_invalidation(struct amdgpu_device *adev)
{
 struct amdgpu_vmhub *hub = &adev->vmhub[AMDGPU_MMHUB0(0)];
 unsigned i;

 for (i = 0; i < 18; ++i) {
  WREG32_SOC15_OFFSET(MMHUB, 0, regMMVM_INVALIDATE_ENG0_ADDR_RANGE_LO32,
        i * hub->eng_addr_distance, 0xffffffff);
  WREG32_SOC15_OFFSET(MMHUB, 0, regMMVM_INVALIDATE_ENG0_ADDR_RANGE_HI32,
        i * hub->eng_addr_distance, 0x1f);
 }
}

static int mmhub_v4_1_0_gart_enable(struct amdgpu_device *adev)
{
 /* GART Enable. */
 mmhub_v4_1_0_init_gart_aperture_regs(adev);
 mmhub_v4_1_0_init_system_aperture_regs(adev);
 mmhub_v4_1_0_init_tlb_regs(adev);
 mmhub_v4_1_0_init_cache_regs(adev);

 mmhub_v4_1_0_enable_system_domain(adev);
 mmhub_v4_1_0_disable_identity_aperture(adev);
 mmhub_v4_1_0_setup_vmid_config(adev);
 mmhub_v4_1_0_program_invalidation(adev);

 return 0;
}

static void mmhub_v4_1_0_gart_disable(struct amdgpu_device *adev)
{
 struct amdgpu_vmhub *hub = &adev->vmhub[AMDGPU_MMHUB0(0)];
 u32 tmp;
 u32 i;

 /* Disable all tables */
 for (i = 0; i < 16; i++)
  WREG32_SOC15_OFFSET(MMHUB, 0, regMMVM_CONTEXT0_CNTL,
        i * hub->ctx_distance, 0);

 /* Setup TLB control */
 tmp = RREG32_SOC15(MMHUB, 0, regMMMC_VM_MX_L1_TLB_CNTL);
 tmp = REG_SET_FIELD(tmp, MMMC_VM_MX_L1_TLB_CNTL, ENABLE_L1_TLB, 0);
 tmp = REG_SET_FIELD(tmp, MMMC_VM_MX_L1_TLB_CNTL,
       ENABLE_ADVANCED_DRIVER_MODEL, 0);
 WREG32_SOC15(MMHUB, 0, regMMMC_VM_MX_L1_TLB_CNTL, tmp);

 /* Setup L2 cache */
 tmp = RREG32_SOC15(MMHUB, 0, regMMVM_L2_CNTL);
 tmp = REG_SET_FIELD(tmp, MMVM_L2_CNTL, ENABLE_L2_CACHE, 0);
 WREG32_SOC15(MMHUB, 0, regMMVM_L2_CNTL, tmp);
 WREG32_SOC15(MMHUB, 0, regMMVM_L2_CNTL3, 0);
}

/**
 * mmhub_v4_1_0_set_fault_enable_default - update GART/VM fault handling
 *
 * @adev: amdgpu_device pointer
 * @value: true redirects VM faults to the default page
 */
static void
mmhub_v4_1_0_set_fault_enable_default(struct amdgpu_device *adev, bool value)
{
 u32 tmp;

 /* These registers are not accessible to VF-SRIOV.
 * The PF will program them instead.
 */
 if (amdgpu_sriov_vf(adev))
  return;

 tmp = RREG32_SOC15(MMHUB, 0, regMMVM_L2_PROTECTION_FAULT_CNTL);
 tmp = REG_SET_FIELD(tmp, MMVM_L2_PROTECTION_FAULT_CNTL,
       RANGE_PROTECTION_FAULT_ENABLE_DEFAULT, value);
 tmp = REG_SET_FIELD(tmp, MMVM_L2_PROTECTION_FAULT_CNTL,
       PDE0_PROTECTION_FAULT_ENABLE_DEFAULT, value);
 tmp = REG_SET_FIELD(tmp, MMVM_L2_PROTECTION_FAULT_CNTL,
       PDE1_PROTECTION_FAULT_ENABLE_DEFAULT, value);
 tmp = REG_SET_FIELD(tmp, MMVM_L2_PROTECTION_FAULT_CNTL,
       PDE2_PROTECTION_FAULT_ENABLE_DEFAULT, value);
 tmp = REG_SET_FIELD(tmp, MMVM_L2_PROTECTION_FAULT_CNTL,
       TRANSLATE_FURTHER_PROTECTION_FAULT_ENABLE_DEFAULT,
       value);
 tmp = REG_SET_FIELD(tmp, MMVM_L2_PROTECTION_FAULT_CNTL,
       NACK_PROTECTION_FAULT_ENABLE_DEFAULT, value);
 tmp = REG_SET_FIELD(tmp, MMVM_L2_PROTECTION_FAULT_CNTL,
       DUMMY_PAGE_PROTECTION_FAULT_ENABLE_DEFAULT, value);
 tmp = REG_SET_FIELD(tmp, MMVM_L2_PROTECTION_FAULT_CNTL,
       VALID_PROTECTION_FAULT_ENABLE_DEFAULT, value);
 tmp = REG_SET_FIELD(tmp, MMVM_L2_PROTECTION_FAULT_CNTL,
       READ_PROTECTION_FAULT_ENABLE_DEFAULT, value);
 tmp = REG_SET_FIELD(tmp, MMVM_L2_PROTECTION_FAULT_CNTL,
       WRITE_PROTECTION_FAULT_ENABLE_DEFAULT, value);
 tmp = REG_SET_FIELD(tmp, MMVM_L2_PROTECTION_FAULT_CNTL,
       EXECUTE_PROTECTION_FAULT_ENABLE_DEFAULT, value);
 if (!value) {
  tmp = REG_SET_FIELD(tmp, MMVM_L2_PROTECTION_FAULT_CNTL,
    CRASH_ON_NO_RETRY_FAULT, 1);
  tmp = REG_SET_FIELD(tmp, MMVM_L2_PROTECTION_FAULT_CNTL,
    CRASH_ON_RETRY_FAULT, 1);
 }
 WREG32_SOC15(MMHUB, 0, regMMVM_L2_PROTECTION_FAULT_CNTL, tmp);
}

static const struct amdgpu_vmhub_funcs mmhub_v4_1_0_vmhub_funcs = {
 .print_l2_protection_fault_status = mmhub_v4_1_0_print_l2_protection_fault_status,
 .get_invalidate_req = mmhub_v4_1_0_get_invalidate_req,
};

static void mmhub_v4_1_0_init(struct amdgpu_device *adev)
{
 struct amdgpu_vmhub *hub = &adev->vmhub[AMDGPU_MMHUB0(0)];

 hub->ctx0_ptb_addr_lo32 =
  SOC15_REG_OFFSET(MMHUB, 0,
     regMMVM_CONTEXT0_PAGE_TABLE_BASE_ADDR_LO32);
 hub->ctx0_ptb_addr_hi32 =
  SOC15_REG_OFFSET(MMHUB, 0,
     regMMVM_CONTEXT0_PAGE_TABLE_BASE_ADDR_HI32);
 hub->vm_inv_eng0_sem =
  SOC15_REG_OFFSET(MMHUB, 0, regMMVM_INVALIDATE_ENG0_SEM);
 hub->vm_inv_eng0_req =
  SOC15_REG_OFFSET(MMHUB, 0, regMMVM_INVALIDATE_ENG0_REQ);
 hub->vm_inv_eng0_ack =
  SOC15_REG_OFFSET(MMHUB, 0, regMMVM_INVALIDATE_ENG0_ACK);
 hub->vm_context0_cntl =
  SOC15_REG_OFFSET(MMHUB, 0, regMMVM_CONTEXT0_CNTL);
 hub->vm_l2_pro_fault_status =
  SOC15_REG_OFFSET(MMHUB, 0, regMMVM_L2_PROTECTION_FAULT_STATUS_LO32);
 hub->vm_l2_pro_fault_cntl =
  SOC15_REG_OFFSET(MMHUB, 0, regMMVM_L2_PROTECTION_FAULT_CNTL);

 hub->ctx_distance = regMMVM_CONTEXT1_CNTL - regMMVM_CONTEXT0_CNTL;
 hub->ctx_addr_distance = regMMVM_CONTEXT1_PAGE_TABLE_BASE_ADDR_LO32 -
  regMMVM_CONTEXT0_PAGE_TABLE_BASE_ADDR_LO32;
 hub->eng_distance = regMMVM_INVALIDATE_ENG1_REQ -
  regMMVM_INVALIDATE_ENG0_REQ;
 hub->eng_addr_distance = regMMVM_INVALIDATE_ENG1_ADDR_RANGE_LO32 -
  regMMVM_INVALIDATE_ENG0_ADDR_RANGE_LO32;

 hub->vm_cntx_cntl_vm_fault = MMVM_CONTEXT1_CNTL__RANGE_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
  MMVM_CONTEXT1_CNTL__DUMMY_PAGE_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
  MMVM_CONTEXT1_CNTL__PDE0_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
  MMVM_CONTEXT1_CNTL__VALID_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
  MMVM_CONTEXT1_CNTL__READ_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
  MMVM_CONTEXT1_CNTL__WRITE_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK |
  MMVM_CONTEXT1_CNTL__EXECUTE_PROTECTION_FAULT_ENABLE_INTERRUPT_MASK;

 hub->vm_l2_bank_select_reserved_cid2 =
  SOC15_REG_OFFSET(MMHUB, 0, regMMVM_L2_BANK_SELECT_RESERVED_CID2);

 hub->vm_contexts_disable =
  SOC15_REG_OFFSET(MMHUB, 0, regMMVM_CONTEXTS_DISABLE);

 hub->vmhub_funcs = &mmhub_v4_1_0_vmhub_funcs;
}

static u64 mmhub_v4_1_0_get_fb_location(struct amdgpu_device *adev)
{
 u64 base;

 base = RREG32_SOC15(MMHUB, 0, regMMMC_VM_FB_LOCATION_BASE);

 base &= MMMC_VM_FB_LOCATION_BASE__FB_BASE_MASK;
 base <<= 24;

 return base;
}

static u64 mmhub_v4_1_0_get_mc_fb_offset(struct amdgpu_device *adev)
{
 return (u64)RREG32_SOC15(MMHUB, 0, regMMMC_VM_FB_OFFSET) << 24;
}

static void
mmhub_v4_1_0_update_medium_grain_clock_gating(struct amdgpu_device *adev,
           bool enable)
{
#if 0
 uint32_t def, data;
#endif
 uint32_t def1, data1, def2 = 0, data2 = 0;
#if 0
 def  = data  = RREG32_SOC15(MMHUB, 0, regMM_ATC_L2_MISC_CG);
#endif
 def1 = data1 = RREG32_SOC15(MMHUB, 0, regDAGB0_CNTL_MISC2);
 def2 = data2 = RREG32_SOC15(MMHUB, 0, regDAGB1_CNTL_MISC2);

 if (enable) {
#if 0
  data |= MM_ATC_L2_MISC_CG__ENABLE_MASK;
#endif
  data1 &= ~(DAGB0_CNTL_MISC2__DISABLE_RDRET_TAP_CHAIN_FGCG_MASK |
      DAGB0_CNTL_MISC2__DISABLE_WRRET_TAP_CHAIN_FGCG_MASK);

  data2 &= ~(DAGB1_CNTL_MISC2__DISABLE_RDRET_TAP_CHAIN_FGCG_MASK |
      DAGB1_CNTL_MISC2__DISABLE_WRRET_TAP_CHAIN_FGCG_MASK);
 } else {
#if 0
  data &= ~MM_ATC_L2_MISC_CG__ENABLE_MASK;
#endif
  data1 |= (DAGB0_CNTL_MISC2__DISABLE_RDRET_TAP_CHAIN_FGCG_MASK |
     DAGB0_CNTL_MISC2__DISABLE_WRRET_TAP_CHAIN_FGCG_MASK);

  data2 |= (DAGB1_CNTL_MISC2__DISABLE_RDRET_TAP_CHAIN_FGCG_MASK |
     DAGB1_CNTL_MISC2__DISABLE_WRRET_TAP_CHAIN_FGCG_MASK);
 }

#if 0
 if (def != data)
  WREG32_SOC15(MMHUB, 0, regMM_ATC_L2_MISC_CG, data);
#endif
 if (def1 != data1)
  WREG32_SOC15(MMHUB, 0, regDAGB0_CNTL_MISC2, data1);

 if (def2 != data2)
  WREG32_SOC15(MMHUB, 0, regDAGB1_CNTL_MISC2, data2);
}

static void
mmhub_v4_1_0_update_medium_grain_light_sleep(struct amdgpu_device *adev,
          bool enable)
{
#if 0
 uint32_t def, data;

 def = data = RREG32_SOC15(MMHUB, 0, regMM_ATC_L2_MISC_CG);

 if (enable)
  data |= MM_ATC_L2_MISC_CG__MEM_LS_ENABLE_MASK;
 else
  data &= ~MM_ATC_L2_MISC_CG__MEM_LS_ENABLE_MASK;

 if (def != data)
  WREG32_SOC15(MMHUB, 0, regMM_ATC_L2_MISC_CG, data);
#endif
}

static int mmhub_v4_1_0_set_clockgating(struct amdgpu_device *adev,
     enum amd_clockgating_state state)
{
 if (amdgpu_sriov_vf(adev))
  return 0;

 if (adev->cg_flags & AMD_CG_SUPPORT_MC_MGCG)
  mmhub_v4_1_0_update_medium_grain_clock_gating(adev,
    state == AMD_CG_STATE_GATE);

 if (adev->cg_flags & AMD_CG_SUPPORT_MC_LS)
  mmhub_v4_1_0_update_medium_grain_light_sleep(adev,
    state == AMD_CG_STATE_GATE);

 return 0;
}

static void mmhub_v4_1_0_get_clockgating(struct amdgpu_device *adev, u64 *flags)
{
#if 0
 int data;

 if (amdgpu_sriov_vf(adev))
  *flags = 0;

 data = RREG32_SOC15(MMHUB, 0, regMM_ATC_L2_MISC_CG);

 /* AMD_CG_SUPPORT_MC_MGCG */
 if (data & MM_ATC_L2_MISC_CG__ENABLE_MASK)
  *flags |= AMD_CG_SUPPORT_MC_MGCG;

 /* AMD_CG_SUPPORT_MC_LS */
 if (data & MM_ATC_L2_MISC_CG__MEM_LS_ENABLE_MASK)
  *flags |= AMD_CG_SUPPORT_MC_LS;
#endif
}

const struct amdgpu_mmhub_funcs mmhub_v4_1_0_funcs = {
 .init = mmhub_v4_1_0_init,
 .get_fb_location = mmhub_v4_1_0_get_fb_location,
 .get_mc_fb_offset = mmhub_v4_1_0_get_mc_fb_offset,
 .gart_enable = mmhub_v4_1_0_gart_enable,
 .set_fault_enable_default = mmhub_v4_1_0_set_fault_enable_default,
 .gart_disable = mmhub_v4_1_0_gart_disable,
 .set_clockgating = mmhub_v4_1_0_set_clockgating,
 .get_clockgating = mmhub_v4_1_0_get_clockgating,
 .setup_vm_pt_regs = mmhub_v4_1_0_setup_vm_pt_regs,
};