uint64_t page_table_base)
{
struct amdgpu_vmhub *hub = &adev->vmhub[AMDGPU_GFXHUB_0];
+ int i;
- WREG32_SOC15_OFFSET(GC, 0, regVM_CONTEXT0_PAGE_TABLE_BASE_ADDR_LO32,
- hub->ctx_addr_distance * vmid,
- lower_32_bits(page_table_base));
+ for (i = 0; i < adev->gfx.num_xcd; i++) {
+ WREG32_SOC15_OFFSET(GC, i,
+ regVM_CONTEXT0_PAGE_TABLE_BASE_ADDR_LO32,
+ hub->ctx_addr_distance * vmid,
+ lower_32_bits(page_table_base));
- WREG32_SOC15_OFFSET(GC, 0, regVM_CONTEXT0_PAGE_TABLE_BASE_ADDR_HI32,
- hub->ctx_addr_distance * vmid,
- upper_32_bits(page_table_base));
+ WREG32_SOC15_OFFSET(GC, i,
+ regVM_CONTEXT0_PAGE_TABLE_BASE_ADDR_HI32,
+ hub->ctx_addr_distance * vmid,
+ upper_32_bits(page_table_base));
+ }
}
static void gfxhub_v1_2_init_gart_aperture_regs(struct amdgpu_device *adev)
{
uint64_t pt_base;
+ int i;
if (adev->gmc.pdb0_bo)
pt_base = amdgpu_gmc_pd_addr(adev->gmc.pdb0_bo);
/* If use GART for FB translation, vmid0 page table covers both
* vram and system memory (gart)
*/
- if (adev->gmc.pdb0_bo) {
- WREG32_SOC15(GC, 0, regVM_CONTEXT0_PAGE_TABLE_START_ADDR_LO32,
- (u32)(adev->gmc.fb_start >> 12));
- WREG32_SOC15(GC, 0, regVM_CONTEXT0_PAGE_TABLE_START_ADDR_HI32,
- (u32)(adev->gmc.fb_start >> 44));
-
- WREG32_SOC15(GC, 0, regVM_CONTEXT0_PAGE_TABLE_END_ADDR_LO32,
- (u32)(adev->gmc.gart_end >> 12));
- WREG32_SOC15(GC, 0, regVM_CONTEXT0_PAGE_TABLE_END_ADDR_HI32,
- (u32)(adev->gmc.gart_end >> 44));
- } else {
- WREG32_SOC15(GC, 0, regVM_CONTEXT0_PAGE_TABLE_START_ADDR_LO32,
- (u32)(adev->gmc.gart_start >> 12));
- WREG32_SOC15(GC, 0, regVM_CONTEXT0_PAGE_TABLE_START_ADDR_HI32,
- (u32)(adev->gmc.gart_start >> 44));
-
- WREG32_SOC15(GC, 0, regVM_CONTEXT0_PAGE_TABLE_END_ADDR_LO32,
- (u32)(adev->gmc.gart_end >> 12));
- WREG32_SOC15(GC, 0, regVM_CONTEXT0_PAGE_TABLE_END_ADDR_HI32,
- (u32)(adev->gmc.gart_end >> 44));
+ for (i = 0; i < adev->gfx.num_xcd; i++) {
+ if (adev->gmc.pdb0_bo) {
+ WREG32_SOC15(GC, i,
+ regVM_CONTEXT0_PAGE_TABLE_START_ADDR_LO32,
+ (u32)(adev->gmc.fb_start >> 12));
+ WREG32_SOC15(GC, i,
+ regVM_CONTEXT0_PAGE_TABLE_START_ADDR_HI32,
+ (u32)(adev->gmc.fb_start >> 44));
+
+ WREG32_SOC15(GC, i,
+ regVM_CONTEXT0_PAGE_TABLE_END_ADDR_LO32,
+ (u32)(adev->gmc.gart_end >> 12));
+ WREG32_SOC15(GC, i,
+ regVM_CONTEXT0_PAGE_TABLE_END_ADDR_HI32,
+ (u32)(adev->gmc.gart_end >> 44));
+ } else {
+ WREG32_SOC15(GC, i,
+ regVM_CONTEXT0_PAGE_TABLE_START_ADDR_LO32,
+ (u32)(adev->gmc.gart_start >> 12));
+ WREG32_SOC15(GC, i,
+ regVM_CONTEXT0_PAGE_TABLE_START_ADDR_HI32,
+ (u32)(adev->gmc.gart_start >> 44));
+
+ WREG32_SOC15(GC, i,
+ regVM_CONTEXT0_PAGE_TABLE_END_ADDR_LO32,
+ (u32)(adev->gmc.gart_end >> 12));
+ WREG32_SOC15(GC, i,
+ regVM_CONTEXT0_PAGE_TABLE_END_ADDR_HI32,
+ (u32)(adev->gmc.gart_end >> 44));
+ }
}
}
{
uint64_t value;
uint32_t tmp;
+ int i;
- /* Program the AGP BAR */
- WREG32_SOC15_RLC(GC, 0, regMC_VM_AGP_BASE, 0);
- WREG32_SOC15_RLC(GC, 0, regMC_VM_AGP_BOT, adev->gmc.agp_start >> 24);
- WREG32_SOC15_RLC(GC, 0, regMC_VM_AGP_TOP, adev->gmc.agp_end >> 24);
-
- if (!amdgpu_sriov_vf(adev) || adev->asic_type <= CHIP_VEGA10) {
- /* Program the system aperture low logical page number. */
- WREG32_SOC15_RLC(GC, 0, regMC_VM_SYSTEM_APERTURE_LOW_ADDR,
- min(adev->gmc.fb_start, adev->gmc.agp_start) >> 18);
-
- if (adev->apu_flags & AMD_APU_IS_RAVEN2)
- /*
- * Raven2 has a HW issue that it is unable to use the
- * vram which is out of MC_VM_SYSTEM_APERTURE_HIGH_ADDR.
- * So here is the workaround that increase system
- * aperture high address (add 1) to get rid of the VM
- * fault and hardware hang.
- */
- WREG32_SOC15_RLC(GC, 0,
- regMC_VM_SYSTEM_APERTURE_HIGH_ADDR,
- max((adev->gmc.fb_end >> 18) + 0x1,
- adev->gmc.agp_end >> 18));
- else
- WREG32_SOC15_RLC(GC, 0,
- regMC_VM_SYSTEM_APERTURE_HIGH_ADDR,
- max(adev->gmc.fb_end, adev->gmc.agp_end) >> 18);
-
- /* Set default page address. */
- value = amdgpu_gmc_vram_mc2pa(adev, adev->mem_scratch.gpu_addr);
- WREG32_SOC15(GC, 0, regMC_VM_SYSTEM_APERTURE_DEFAULT_ADDR_LSB,
- (u32)(value >> 12));
- WREG32_SOC15(GC, 0, regMC_VM_SYSTEM_APERTURE_DEFAULT_ADDR_MSB,
- (u32)(value >> 44));
-
- /* Program "protection fault". */
- WREG32_SOC15(GC, 0, regVM_L2_PROTECTION_FAULT_DEFAULT_ADDR_LO32,
- (u32)(adev->dummy_page_addr >> 12));
- WREG32_SOC15(GC, 0, regVM_L2_PROTECTION_FAULT_DEFAULT_ADDR_HI32,
- (u32)((u64)adev->dummy_page_addr >> 44));
-
- tmp = RREG32_SOC15(GC, 0, regVM_L2_PROTECTION_FAULT_CNTL2);
- tmp = REG_SET_FIELD(tmp, VM_L2_PROTECTION_FAULT_CNTL2,
- ACTIVE_PAGE_MIGRATION_PTE_READ_RETRY, 1);
- WREG32_SOC15(GC, 0, regVM_L2_PROTECTION_FAULT_CNTL2, tmp);
- }
-
- /* In the case squeezing vram into GART aperture, we don't use
- * FB aperture and AGP aperture. Disable them.
- */
- if (adev->gmc.pdb0_bo) {
- WREG32_SOC15(GC, 0, regMC_VM_FB_LOCATION_TOP, 0);
- WREG32_SOC15(GC, 0, regMC_VM_FB_LOCATION_BASE, 0x00FFFFFF);
- WREG32_SOC15(GC, 0, regMC_VM_AGP_TOP, 0);
- WREG32_SOC15(GC, 0, regMC_VM_AGP_BOT, 0xFFFFFF);
- WREG32_SOC15(GC, 0, regMC_VM_SYSTEM_APERTURE_LOW_ADDR, 0x3FFFFFFF);
- WREG32_SOC15(GC, 0, regMC_VM_SYSTEM_APERTURE_HIGH_ADDR, 0);
+ for (i = 0; i < adev->gfx.num_xcd; i++) {
+ /* Program the AGP BAR */
+ WREG32_SOC15_RLC(GC, i, regMC_VM_AGP_BASE, 0);
+ WREG32_SOC15_RLC(GC, i, regMC_VM_AGP_BOT, adev->gmc.agp_start >> 24);
+ WREG32_SOC15_RLC(GC, i, regMC_VM_AGP_TOP, adev->gmc.agp_end >> 24);
+
+ if (!amdgpu_sriov_vf(adev) || adev->asic_type <= CHIP_VEGA10) {
+ /* Program the system aperture low logical page number. */
+ WREG32_SOC15_RLC(GC, i, regMC_VM_SYSTEM_APERTURE_LOW_ADDR,
+ min(adev->gmc.fb_start, adev->gmc.agp_start) >> 18);
+
+ if (adev->apu_flags & AMD_APU_IS_RAVEN2)
+ /*
+ * Raven2 has a HW issue that it is unable to use the
+ * vram which is out of MC_VM_SYSTEM_APERTURE_HIGH_ADDR.
+ * So here is the workaround that increase system
+ * aperture high address (add 1) to get rid of the VM
+ * fault and hardware hang.
+ */
+ WREG32_SOC15_RLC(GC, i,
+ regMC_VM_SYSTEM_APERTURE_HIGH_ADDR,
+ max((adev->gmc.fb_end >> 18) + 0x1,
+ adev->gmc.agp_end >> 18));
+ else
+ WREG32_SOC15_RLC(GC, i,
+ regMC_VM_SYSTEM_APERTURE_HIGH_ADDR,
+ max(adev->gmc.fb_end, adev->gmc.agp_end) >> 18);
+
+ /* Set default page address. */
+ value = amdgpu_gmc_vram_mc2pa(adev, adev->mem_scratch.gpu_addr);
+ WREG32_SOC15(GC, i, regMC_VM_SYSTEM_APERTURE_DEFAULT_ADDR_LSB,
+ (u32)(value >> 12));
+ WREG32_SOC15(GC, i, regMC_VM_SYSTEM_APERTURE_DEFAULT_ADDR_MSB,
+ (u32)(value >> 44));
+
+ /* Program "protection fault". */
+ WREG32_SOC15(GC, i, regVM_L2_PROTECTION_FAULT_DEFAULT_ADDR_LO32,
+ (u32)(adev->dummy_page_addr >> 12));
+ WREG32_SOC15(GC, i, regVM_L2_PROTECTION_FAULT_DEFAULT_ADDR_HI32,
+ (u32)((u64)adev->dummy_page_addr >> 44));
+
+ tmp = RREG32_SOC15(GC, i, regVM_L2_PROTECTION_FAULT_CNTL2);
+ tmp = REG_SET_FIELD(tmp, VM_L2_PROTECTION_FAULT_CNTL2,
+ ACTIVE_PAGE_MIGRATION_PTE_READ_RETRY, 1);
+ WREG32_SOC15(GC, i, regVM_L2_PROTECTION_FAULT_CNTL2, tmp);
+ }
+
+ /* In the case squeezing vram into GART aperture, we don't use
+ * FB aperture and AGP aperture. Disable them.
+ */
+ if (adev->gmc.pdb0_bo) {
+ WREG32_SOC15(GC, i, regMC_VM_FB_LOCATION_TOP, 0);
+ WREG32_SOC15(GC, i, regMC_VM_FB_LOCATION_BASE, 0x00FFFFFF);
+ WREG32_SOC15(GC, i, regMC_VM_AGP_TOP, 0);
+ WREG32_SOC15(GC, i, regMC_VM_AGP_BOT, 0xFFFFFF);
+ WREG32_SOC15(GC, i, regMC_VM_SYSTEM_APERTURE_LOW_ADDR, 0x3FFFFFFF);
+ WREG32_SOC15(GC, i, regMC_VM_SYSTEM_APERTURE_HIGH_ADDR, 0);
+ }
}
}
static void gfxhub_v1_2_init_tlb_regs(struct amdgpu_device *adev)
{
uint32_t tmp;
+ int i;
- /* Setup TLB control */
- tmp = RREG32_SOC15(GC, 0, regMC_VM_MX_L1_TLB_CNTL);
-
- tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL, ENABLE_L1_TLB, 1);
- tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL, SYSTEM_ACCESS_MODE, 3);
- tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL,
- ENABLE_ADVANCED_DRIVER_MODEL, 1);
- tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL,
- SYSTEM_APERTURE_UNMAPPED_ACCESS, 0);
- tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL,
- MTYPE, MTYPE_UC);/* XXX for emulation. */
- tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL, ATC_EN, 1);
-
- WREG32_SOC15_RLC(GC, 0, regMC_VM_MX_L1_TLB_CNTL, tmp);
+ for (i = 0; i < adev->gfx.num_xcd; i++) {
+ /* Setup TLB control */
+ tmp = RREG32_SOC15(GC, i, regMC_VM_MX_L1_TLB_CNTL);
+
+ tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL,
+ ENABLE_L1_TLB, 1);
+ tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL,
+ SYSTEM_ACCESS_MODE, 3);
+ tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL,
+ ENABLE_ADVANCED_DRIVER_MODEL, 1);
+ tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL,
+ SYSTEM_APERTURE_UNMAPPED_ACCESS, 0);
+ tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL,
+ MTYPE, MTYPE_UC);/* XXX for emulation. */
+ tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL, ATC_EN, 1);
+
+ WREG32_SOC15_RLC(GC, i, regMC_VM_MX_L1_TLB_CNTL, tmp);
+ }
}
static void gfxhub_v1_2_init_cache_regs(struct amdgpu_device *adev)
{
uint32_t tmp;
+ int i;
- /* Setup L2 cache */
- tmp = RREG32_SOC15(GC, 0, regVM_L2_CNTL);
- tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, ENABLE_L2_CACHE, 1);
- tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, ENABLE_L2_FRAGMENT_PROCESSING, 1);
- /* XXX for emulation, Refer to closed source code.*/
- tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, L2_PDE0_CACHE_TAG_GENERATION_MODE,
- 0);
- tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, PDE_FAULT_CLASSIFICATION, 0);
- tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, CONTEXT1_IDENTITY_ACCESS_MODE, 1);
- tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, IDENTITY_MODE_FRAGMENT_SIZE, 0);
- WREG32_SOC15_RLC(GC, 0, regVM_L2_CNTL, tmp);
-
- tmp = RREG32_SOC15(GC, 0, regVM_L2_CNTL2);
- tmp = REG_SET_FIELD(tmp, VM_L2_CNTL2, INVALIDATE_ALL_L1_TLBS, 1);
- tmp = REG_SET_FIELD(tmp, VM_L2_CNTL2, INVALIDATE_L2_CACHE, 1);
- WREG32_SOC15_RLC(GC, 0, regVM_L2_CNTL2, tmp);
-
- tmp = regVM_L2_CNTL3_DEFAULT;
- if (adev->gmc.translate_further) {
- tmp = REG_SET_FIELD(tmp, VM_L2_CNTL3, BANK_SELECT, 12);
- tmp = REG_SET_FIELD(tmp, VM_L2_CNTL3,
- L2_CACHE_BIGK_FRAGMENT_SIZE, 9);
- } else {
- tmp = REG_SET_FIELD(tmp, VM_L2_CNTL3, BANK_SELECT, 9);
- tmp = REG_SET_FIELD(tmp, VM_L2_CNTL3,
- L2_CACHE_BIGK_FRAGMENT_SIZE, 6);
- }
- WREG32_SOC15_RLC(GC, 0, regVM_L2_CNTL3, tmp);
-
- tmp = regVM_L2_CNTL4_DEFAULT;
- if (adev->gmc.xgmi.connected_to_cpu) {
- tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_PDE_REQUEST_PHYSICAL, 1);
- tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_PTE_REQUEST_PHYSICAL, 1);
- } else {
- tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_PDE_REQUEST_PHYSICAL, 0);
- tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_PTE_REQUEST_PHYSICAL, 0);
+ for (i = 0; i < adev->gfx.num_xcd; i++) {
+ /* Setup L2 cache */
+ tmp = RREG32_SOC15(GC, i, regVM_L2_CNTL);
+ tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, ENABLE_L2_CACHE, 1);
+ tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, ENABLE_L2_FRAGMENT_PROCESSING, 1);
+ /* XXX for emulation, Refer to closed source code.*/
+ tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, L2_PDE0_CACHE_TAG_GENERATION_MODE,
+ 0);
+ tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, PDE_FAULT_CLASSIFICATION, 0);
+ tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, CONTEXT1_IDENTITY_ACCESS_MODE, 1);
+ tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, IDENTITY_MODE_FRAGMENT_SIZE, 0);
+ WREG32_SOC15_RLC(GC, i, regVM_L2_CNTL, tmp);
+
+ tmp = RREG32_SOC15(GC, i, regVM_L2_CNTL2);
+ tmp = REG_SET_FIELD(tmp, VM_L2_CNTL2, INVALIDATE_ALL_L1_TLBS, 1);
+ tmp = REG_SET_FIELD(tmp, VM_L2_CNTL2, INVALIDATE_L2_CACHE, 1);
+ WREG32_SOC15_RLC(GC, i, regVM_L2_CNTL2, tmp);
+
+ tmp = regVM_L2_CNTL3_DEFAULT;
+ if (adev->gmc.translate_further) {
+ tmp = REG_SET_FIELD(tmp, VM_L2_CNTL3, BANK_SELECT, 12);
+ tmp = REG_SET_FIELD(tmp, VM_L2_CNTL3,
+ L2_CACHE_BIGK_FRAGMENT_SIZE, 9);
+ } else {
+ tmp = REG_SET_FIELD(tmp, VM_L2_CNTL3, BANK_SELECT, 9);
+ tmp = REG_SET_FIELD(tmp, VM_L2_CNTL3,
+ L2_CACHE_BIGK_FRAGMENT_SIZE, 6);
+ }
+ WREG32_SOC15_RLC(GC, i, regVM_L2_CNTL3, tmp);
+
+ tmp = regVM_L2_CNTL4_DEFAULT;
+ if (adev->gmc.xgmi.connected_to_cpu) {
+ tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_PDE_REQUEST_PHYSICAL, 1);
+ tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_PTE_REQUEST_PHYSICAL, 1);
+ } else {
+ tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_PDE_REQUEST_PHYSICAL, 0);
+ tmp = REG_SET_FIELD(tmp, VM_L2_CNTL4, VMC_TAP_PTE_REQUEST_PHYSICAL, 0);
+ }
+ WREG32_SOC15_RLC(GC, i, regVM_L2_CNTL4, tmp);
}
- WREG32_SOC15_RLC(GC, 0, regVM_L2_CNTL4, tmp);
}
static void gfxhub_v1_2_enable_system_domain(struct amdgpu_device *adev)
{
uint32_t tmp;
+ int i;
- tmp = RREG32_SOC15(GC, 0, regVM_CONTEXT0_CNTL);
- tmp = REG_SET_FIELD(tmp, VM_CONTEXT0_CNTL, ENABLE_CONTEXT, 1);
- tmp = REG_SET_FIELD(tmp, VM_CONTEXT0_CNTL, PAGE_TABLE_DEPTH,
- adev->gmc.vmid0_page_table_depth);
- tmp = REG_SET_FIELD(tmp, VM_CONTEXT0_CNTL, PAGE_TABLE_BLOCK_SIZE,
- adev->gmc.vmid0_page_table_block_size);
- tmp = REG_SET_FIELD(tmp, VM_CONTEXT0_CNTL,
- RETRY_PERMISSION_OR_INVALID_PAGE_FAULT, 0);
- WREG32_SOC15(GC, 0, regVM_CONTEXT0_CNTL, tmp);
+ for (i = 0; i < adev->gfx.num_xcd; i++) {
+ tmp = RREG32_SOC15(GC, i, regVM_CONTEXT0_CNTL);
+ tmp = REG_SET_FIELD(tmp, VM_CONTEXT0_CNTL, ENABLE_CONTEXT, 1);
+ tmp = REG_SET_FIELD(tmp, VM_CONTEXT0_CNTL, PAGE_TABLE_DEPTH,
+ adev->gmc.vmid0_page_table_depth);
+ tmp = REG_SET_FIELD(tmp, VM_CONTEXT0_CNTL, PAGE_TABLE_BLOCK_SIZE,
+ adev->gmc.vmid0_page_table_block_size);
+ tmp = REG_SET_FIELD(tmp, VM_CONTEXT0_CNTL,
+ RETRY_PERMISSION_OR_INVALID_PAGE_FAULT, 0);
+ WREG32_SOC15(GC, i, regVM_CONTEXT0_CNTL, tmp);
+ }
}
static void gfxhub_v1_2_disable_identity_aperture(struct amdgpu_device *adev)
{
- WREG32_SOC15(GC, 0, regVM_L2_CONTEXT1_IDENTITY_APERTURE_LOW_ADDR_LO32,
- 0XFFFFFFFF);
- WREG32_SOC15(GC, 0, regVM_L2_CONTEXT1_IDENTITY_APERTURE_LOW_ADDR_HI32,
- 0x0000000F);
-
- WREG32_SOC15(GC, 0, regVM_L2_CONTEXT1_IDENTITY_APERTURE_HIGH_ADDR_LO32,
- 0);
- WREG32_SOC15(GC, 0, regVM_L2_CONTEXT1_IDENTITY_APERTURE_HIGH_ADDR_HI32,
- 0);
-
- WREG32_SOC15(GC, 0, regVM_L2_CONTEXT_IDENTITY_PHYSICAL_OFFSET_LO32, 0);
- WREG32_SOC15(GC, 0, regVM_L2_CONTEXT_IDENTITY_PHYSICAL_OFFSET_HI32, 0);
+ int i;
+ for (i = 0; i < adev->gfx.num_xcd; i++) {
+ WREG32_SOC15(GC, i,
+ regVM_L2_CONTEXT1_IDENTITY_APERTURE_LOW_ADDR_LO32,
+ 0XFFFFFFFF);
+ WREG32_SOC15(GC, i,
+ regVM_L2_CONTEXT1_IDENTITY_APERTURE_LOW_ADDR_HI32,
+ 0x0000000F);
+
+ WREG32_SOC15(GC, i,
+ regVM_L2_CONTEXT1_IDENTITY_APERTURE_HIGH_ADDR_LO32,
+ 0);
+ WREG32_SOC15(GC, i,
+ regVM_L2_CONTEXT1_IDENTITY_APERTURE_HIGH_ADDR_HI32,
+ 0);
+
+ WREG32_SOC15(GC, i,
+ regVM_L2_CONTEXT_IDENTITY_PHYSICAL_OFFSET_LO32, 0);
+ WREG32_SOC15(GC, i,
+ regVM_L2_CONTEXT_IDENTITY_PHYSICAL_OFFSET_HI32, 0);
+ }
}
static void gfxhub_v1_2_setup_vmid_config(struct amdgpu_device *adev)
struct amdgpu_vmhub *hub = &adev->vmhub[AMDGPU_GFXHUB_0];
unsigned num_level, block_size;
uint32_t tmp;
- int i;
+ int i, j;
num_level = adev->vm_manager.num_level;
block_size = adev->vm_manager.block_size;
else
block_size -= 9;
- for (i = 0; i <= 14; i++) {
- tmp = RREG32_SOC15_OFFSET(GC, 0, regVM_CONTEXT1_CNTL, i);
- tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, ENABLE_CONTEXT, 1);
- tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, PAGE_TABLE_DEPTH,
- num_level);
- tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
- RANGE_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
- tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
- DUMMY_PAGE_PROTECTION_FAULT_ENABLE_DEFAULT,
- 1);
- tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
- PDE0_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
- tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
- VALID_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
- tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
- READ_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
- tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
- WRITE_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
- tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
- EXECUTE_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
- tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
- PAGE_TABLE_BLOCK_SIZE,
- block_size);
- /* Send no-retry XNACK on fault to suppress VM fault storm.
- * On Aldebaran, XNACK can be enabled in the SQ per-process.
- * Retry faults need to be enabled for that to work.
- */
- tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
- RETRY_PERMISSION_OR_INVALID_PAGE_FAULT,
- !adev->gmc.noretry ||
- adev->asic_type == CHIP_ALDEBARAN);
- WREG32_SOC15_OFFSET(GC, 0, regVM_CONTEXT1_CNTL,
- i * hub->ctx_distance, tmp);
- WREG32_SOC15_OFFSET(GC, 0,
- regVM_CONTEXT1_PAGE_TABLE_START_ADDR_LO32,
- i * hub->ctx_addr_distance, 0);
- WREG32_SOC15_OFFSET(GC, 0,
- regVM_CONTEXT1_PAGE_TABLE_START_ADDR_HI32,
- i * hub->ctx_addr_distance, 0);
- WREG32_SOC15_OFFSET(GC, 0,
- regVM_CONTEXT1_PAGE_TABLE_END_ADDR_LO32,
- i * hub->ctx_addr_distance,
- lower_32_bits(adev->vm_manager.max_pfn - 1));
- WREG32_SOC15_OFFSET(GC, 0,
- regVM_CONTEXT1_PAGE_TABLE_END_ADDR_HI32,
- i * hub->ctx_addr_distance,
- upper_32_bits(adev->vm_manager.max_pfn - 1));
+ for (j = 0; j < adev->gfx.num_xcd; j++) {
+ for (i = 0; i <= 14; i++) {
+ tmp = RREG32_SOC15_OFFSET(GC, j, regVM_CONTEXT1_CNTL, i);
+ tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, ENABLE_CONTEXT, 1);
+ tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL, PAGE_TABLE_DEPTH,
+ num_level);
+ tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
+ RANGE_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
+ tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
+ DUMMY_PAGE_PROTECTION_FAULT_ENABLE_DEFAULT,
+ 1);
+ tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
+ PDE0_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
+ tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
+ VALID_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
+ tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
+ READ_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
+ tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
+ WRITE_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
+ tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
+ EXECUTE_PROTECTION_FAULT_ENABLE_DEFAULT, 1);
+ tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
+ PAGE_TABLE_BLOCK_SIZE,
+ block_size);
+ /* Send no-retry XNACK on fault to suppress VM fault storm.
+ * On Aldebaran, XNACK can be enabled in the SQ per-process.
+ * Retry faults need to be enabled for that to work.
+ */
+ tmp = REG_SET_FIELD(tmp, VM_CONTEXT1_CNTL,
+ RETRY_PERMISSION_OR_INVALID_PAGE_FAULT,
+ !adev->gmc.noretry ||
+ adev->asic_type == CHIP_ALDEBARAN);
+ WREG32_SOC15_OFFSET(GC, j, regVM_CONTEXT1_CNTL,
+ i * hub->ctx_distance, tmp);
+ WREG32_SOC15_OFFSET(GC, j,
+ regVM_CONTEXT1_PAGE_TABLE_START_ADDR_LO32,
+ i * hub->ctx_addr_distance, 0);
+ WREG32_SOC15_OFFSET(GC, j,
+ regVM_CONTEXT1_PAGE_TABLE_START_ADDR_HI32,
+ i * hub->ctx_addr_distance, 0);
+ WREG32_SOC15_OFFSET(GC, j,
+ regVM_CONTEXT1_PAGE_TABLE_END_ADDR_LO32,
+ i * hub->ctx_addr_distance,
+ lower_32_bits(adev->vm_manager.max_pfn - 1));
+ WREG32_SOC15_OFFSET(GC, j,
+ regVM_CONTEXT1_PAGE_TABLE_END_ADDR_HI32,
+ i * hub->ctx_addr_distance,
+ upper_32_bits(adev->vm_manager.max_pfn - 1));
+ }
}
}
static void gfxhub_v1_2_program_invalidation(struct amdgpu_device *adev)
{
struct amdgpu_vmhub *hub = &adev->vmhub[AMDGPU_GFXHUB_0];
- unsigned i;
-
- for (i = 0 ; i < 18; ++i) {
- WREG32_SOC15_OFFSET(GC, 0, regVM_INVALIDATE_ENG0_ADDR_RANGE_LO32,
- i * hub->eng_addr_distance, 0xffffffff);
- WREG32_SOC15_OFFSET(GC, 0, regVM_INVALIDATE_ENG0_ADDR_RANGE_HI32,
- i * hub->eng_addr_distance, 0x1f);
+ unsigned i, j;
+
+ for (j = 0; j < adev->gfx.num_xcd; j++) {
+ for (i = 0 ; i < 18; ++i) {
+ WREG32_SOC15_OFFSET(GC, j, regVM_INVALIDATE_ENG0_ADDR_RANGE_LO32,
+ i * hub->eng_addr_distance, 0xffffffff);
+ WREG32_SOC15_OFFSET(GC, j, regVM_INVALIDATE_ENG0_ADDR_RANGE_HI32,
+ i * hub->eng_addr_distance, 0x1f);
+ }
}
}
static int gfxhub_v1_2_gart_enable(struct amdgpu_device *adev)
{
- if (amdgpu_sriov_vf(adev) && adev->asic_type != CHIP_ARCTURUS) {
+ int i;
+
+ for (i = 0; i < adev->gfx.num_xcd; i++) {
+ if (amdgpu_sriov_vf(adev)) {
/*
* MC_VM_FB_LOCATION_BASE/TOP is NULL for VF, becuase they are
* VF copy registers so vbios post doesn't program them, for
* SRIOV driver need to program them
*/
- WREG32_SOC15_RLC(GC, 0, regMC_VM_FB_LOCATION_BASE,
- adev->gmc.vram_start >> 24);
- WREG32_SOC15_RLC(GC, 0, regMC_VM_FB_LOCATION_TOP,
- adev->gmc.vram_end >> 24);
+ WREG32_SOC15_RLC(GC, i, regMC_VM_FB_LOCATION_BASE,
+ adev->gmc.vram_start >> 24);
+ WREG32_SOC15_RLC(GC, i, regMC_VM_FB_LOCATION_TOP,
+ adev->gmc.vram_end >> 24);
+ }
}
/* GART Enable. */
{
struct amdgpu_vmhub *hub = &adev->vmhub[AMDGPU_GFXHUB_0];
u32 tmp;
- u32 i;
-
- /* Disable all tables */
- for (i = 0; i < 16; i++)
- WREG32_SOC15_OFFSET(GC, 0, regVM_CONTEXT0_CNTL,
- i * hub->ctx_distance, 0);
-
- /* Setup TLB control */
- tmp = RREG32_SOC15(GC, 0, regMC_VM_MX_L1_TLB_CNTL);
- tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL, ENABLE_L1_TLB, 0);
- tmp = REG_SET_FIELD(tmp,
- MC_VM_MX_L1_TLB_CNTL,
- ENABLE_ADVANCED_DRIVER_MODEL,
- 0);
- WREG32_SOC15_RLC(GC, 0, regMC_VM_MX_L1_TLB_CNTL, tmp);
-
- /* Setup L2 cache */
- tmp = RREG32_SOC15(GC, 0, regVM_L2_CNTL);
- tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, ENABLE_L2_CACHE, 0);
- WREG32_SOC15(GC, 0, regVM_L2_CNTL, tmp);
- WREG32_SOC15(GC, 0, regVM_L2_CNTL3, 0);
+ u32 i, j;
+
+ for (j = 0; j < adev->gfx.num_xcd; j++) {
+ /* Disable all tables */
+ for (i = 0; i < 16; i++)
+ WREG32_SOC15_OFFSET(GC, j, regVM_CONTEXT0_CNTL,
+ i * hub->ctx_distance, 0);
+
+ /* Setup TLB control */
+ tmp = RREG32_SOC15(GC, j, regMC_VM_MX_L1_TLB_CNTL);
+ tmp = REG_SET_FIELD(tmp, MC_VM_MX_L1_TLB_CNTL, ENABLE_L1_TLB, 0);
+ tmp = REG_SET_FIELD(tmp,
+ MC_VM_MX_L1_TLB_CNTL,
+ ENABLE_ADVANCED_DRIVER_MODEL,
+ 0);
+ WREG32_SOC15_RLC(GC, j, regMC_VM_MX_L1_TLB_CNTL, tmp);
+
+ /* Setup L2 cache */
+ tmp = RREG32_SOC15(GC, j, regVM_L2_CNTL);
+ tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, ENABLE_L2_CACHE, 0);
+ WREG32_SOC15(GC, j, regVM_L2_CNTL, tmp);
+ WREG32_SOC15(GC, j, regVM_L2_CNTL3, 0);
+ }
}
/**
bool value)
{
u32 tmp;
- tmp = RREG32_SOC15(GC, 0, regVM_L2_PROTECTION_FAULT_CNTL);
- tmp = REG_SET_FIELD(tmp, VM_L2_PROTECTION_FAULT_CNTL,
- RANGE_PROTECTION_FAULT_ENABLE_DEFAULT, value);
- tmp = REG_SET_FIELD(tmp, VM_L2_PROTECTION_FAULT_CNTL,
- PDE0_PROTECTION_FAULT_ENABLE_DEFAULT, value);
- tmp = REG_SET_FIELD(tmp, VM_L2_PROTECTION_FAULT_CNTL,
- PDE1_PROTECTION_FAULT_ENABLE_DEFAULT, value);
- tmp = REG_SET_FIELD(tmp, VM_L2_PROTECTION_FAULT_CNTL,
- PDE2_PROTECTION_FAULT_ENABLE_DEFAULT, value);
- tmp = REG_SET_FIELD(tmp,
- VM_L2_PROTECTION_FAULT_CNTL,
- TRANSLATE_FURTHER_PROTECTION_FAULT_ENABLE_DEFAULT,
- value);
- tmp = REG_SET_FIELD(tmp, VM_L2_PROTECTION_FAULT_CNTL,
- NACK_PROTECTION_FAULT_ENABLE_DEFAULT, value);
- tmp = REG_SET_FIELD(tmp, VM_L2_PROTECTION_FAULT_CNTL,
- DUMMY_PAGE_PROTECTION_FAULT_ENABLE_DEFAULT, value);
- tmp = REG_SET_FIELD(tmp, VM_L2_PROTECTION_FAULT_CNTL,
- VALID_PROTECTION_FAULT_ENABLE_DEFAULT, value);
- tmp = REG_SET_FIELD(tmp, VM_L2_PROTECTION_FAULT_CNTL,
- READ_PROTECTION_FAULT_ENABLE_DEFAULT, value);
- tmp = REG_SET_FIELD(tmp, VM_L2_PROTECTION_FAULT_CNTL,
- WRITE_PROTECTION_FAULT_ENABLE_DEFAULT, value);
- tmp = REG_SET_FIELD(tmp, VM_L2_PROTECTION_FAULT_CNTL,
- EXECUTE_PROTECTION_FAULT_ENABLE_DEFAULT, value);
- if (!value) {
+ int i;
+
+ for (i = 0; i < adev->gfx.num_xcd; i++) {
+ tmp = RREG32_SOC15(GC, i, regVM_L2_PROTECTION_FAULT_CNTL);
+ tmp = REG_SET_FIELD(tmp, VM_L2_PROTECTION_FAULT_CNTL,
+ RANGE_PROTECTION_FAULT_ENABLE_DEFAULT, value);
+ tmp = REG_SET_FIELD(tmp, VM_L2_PROTECTION_FAULT_CNTL,
+ PDE0_PROTECTION_FAULT_ENABLE_DEFAULT, value);
+ tmp = REG_SET_FIELD(tmp, VM_L2_PROTECTION_FAULT_CNTL,
+ PDE1_PROTECTION_FAULT_ENABLE_DEFAULT, value);
+ tmp = REG_SET_FIELD(tmp, VM_L2_PROTECTION_FAULT_CNTL,
+ PDE2_PROTECTION_FAULT_ENABLE_DEFAULT, value);
+ tmp = REG_SET_FIELD(tmp,
+ VM_L2_PROTECTION_FAULT_CNTL,
+ TRANSLATE_FURTHER_PROTECTION_FAULT_ENABLE_DEFAULT,
+ value);
+ tmp = REG_SET_FIELD(tmp, VM_L2_PROTECTION_FAULT_CNTL,
+ NACK_PROTECTION_FAULT_ENABLE_DEFAULT, value);
+ tmp = REG_SET_FIELD(tmp, VM_L2_PROTECTION_FAULT_CNTL,
+ DUMMY_PAGE_PROTECTION_FAULT_ENABLE_DEFAULT, value);
+ tmp = REG_SET_FIELD(tmp, VM_L2_PROTECTION_FAULT_CNTL,
+ VALID_PROTECTION_FAULT_ENABLE_DEFAULT, value);
+ tmp = REG_SET_FIELD(tmp, VM_L2_PROTECTION_FAULT_CNTL,
+ READ_PROTECTION_FAULT_ENABLE_DEFAULT, value);
tmp = REG_SET_FIELD(tmp, VM_L2_PROTECTION_FAULT_CNTL,
- CRASH_ON_NO_RETRY_FAULT, 1);
+ WRITE_PROTECTION_FAULT_ENABLE_DEFAULT, value);
tmp = REG_SET_FIELD(tmp, VM_L2_PROTECTION_FAULT_CNTL,
- CRASH_ON_RETRY_FAULT, 1);
+ EXECUTE_PROTECTION_FAULT_ENABLE_DEFAULT, value);
+ if (!value) {
+ tmp = REG_SET_FIELD(tmp, VM_L2_PROTECTION_FAULT_CNTL,
+ CRASH_ON_NO_RETRY_FAULT, 1);
+ tmp = REG_SET_FIELD(tmp, VM_L2_PROTECTION_FAULT_CNTL,
+ CRASH_ON_RETRY_FAULT, 1);
+ }
+ WREG32_SOC15(GC, i, regVM_L2_PROTECTION_FAULT_CNTL, tmp);
}
- WREG32_SOC15(GC, 0, regVM_L2_PROTECTION_FAULT_CNTL, tmp);
}
static void gfxhub_v1_2_init(struct amdgpu_device *adev)
projects / platform / kernel / linux-starfive.git / commitdiff