From: Rajneesh Bhardwaj <[email protected]> Recently introduced commit "drm/amdgpu: Set cache coherency for GC 9.4.3" did not update the settings applicable for svm ranges. Add the coherence settings for svm ranges for GFX IP 9.4.3.
Reviewed-by: Amber Lin <[email protected]> Signed-off-by: Rajneesh Bhardwaj <[email protected]> Signed-off-by: Alex Deucher <[email protected]> --- drivers/gpu/drm/amd/amdkfd/kfd_svm.c | 17 +++++++++++++++++ include/uapi/linux/kfd_ioctl.h | 2 ++ 2 files changed, 19 insertions(+) diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_svm.c b/drivers/gpu/drm/amd/amdkfd/kfd_svm.c index 9f0932d0c93f..8023be9a9cc1 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_svm.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_svm.c @@ -1156,6 +1156,7 @@ svm_range_get_pte_flags(struct amdgpu_device *adev, struct svm_range *prange, uint64_t pte_flags; bool snoop = (domain != SVM_RANGE_VRAM_DOMAIN); bool coherent = flags & KFD_IOCTL_SVM_FLAG_COHERENT; + bool uncached = flags & KFD_IOCTL_SVM_FLAG_UNCACHED; if (domain == SVM_RANGE_VRAM_DOMAIN) bo_adev = amdgpu_ttm_adev(prange->svm_bo->bo->tbo.bdev); @@ -1195,6 +1196,22 @@ svm_range_get_pte_flags(struct amdgpu_device *adev, struct svm_range *prange, AMDGPU_VM_MTYPE_UC : AMDGPU_VM_MTYPE_NC; } break; + case IP_VERSION(9, 4, 3): + //TODO: Need more work for handling multiple memory partitions + //e.g. NPS4. Current approch is only applicable without memory + //partitions. + snoop = true; + if (uncached) + mapping_flags |= AMDGPU_VM_MTYPE_UC; + /* local HBM region close to partition*/ + else if (bo_adev == adev) + mapping_flags |= AMDGPU_VM_MTYPE_RW; + /* local HBM region far from partition or remote XGMI GPU or + * system memory + */ + else + mapping_flags |= AMDGPU_VM_MTYPE_NC; + break; default: mapping_flags |= coherent ? AMDGPU_VM_MTYPE_UC : AMDGPU_VM_MTYPE_NC; diff --git a/include/uapi/linux/kfd_ioctl.h b/include/uapi/linux/kfd_ioctl.h index 2da5c3ad71bd..2a9671e1ddb5 100644 --- a/include/uapi/linux/kfd_ioctl.h +++ b/include/uapi/linux/kfd_ioctl.h @@ -623,6 +623,8 @@ enum kfd_mmio_remap { #define KFD_IOCTL_SVM_FLAG_GPU_READ_MOSTLY 0x00000020 /* Keep GPU memory mapping always valid as if XNACK is disable */ #define KFD_IOCTL_SVM_FLAG_GPU_ALWAYS_MAPPED 0x00000040 +/* Uncached access to memory */ +#define KFD_IOCTL_SVM_FLAG_UNCACHED 0x00000080 /** * kfd_ioctl_svm_op - SVM ioctl operations -- 2.39.2
