Commit 5f248462 authored by David Francis's avatar David Francis Committed by Alex Deucher
Browse files

drm/amdgpu: Add EXT_COHERENT memory allocation flags 

These flags (for GEM and SVM allocations) allocate
memory that allows for system-scope atomic semantics.

On GFX943 these flags cause caches to be avoided on
non-local memory.

On all other ASICs they are identical in functionality to the
equivalent COHERENT flags.

Corresponding Thunk patch is at
https://github.com/RadeonOpenCompute/ROCT-Thunk-Interface/pull/88



Reviewed-by: default avatarDavid Yat Sin <David.YatSin@amd.com>
Signed-off-by: default avatarDavid Francis <David.Francis@amd.com>
Signed-off-by: default avatarAlex Deucher <alexander.deucher@amd.com>
parent d92e5556

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

+2 −0
Original line number Diff line number Diff line
@@ -1691,6 +1691,8 @@ int amdgpu_amdkfd_gpuvm_alloc_memory_of_gpu( 


	if (flags & KFD_IOC_ALLOC_MEM_FLAGS_COHERENT)

		alloc_flags |= AMDGPU_GEM_CREATE_COHERENT;

	if (flags & KFD_IOC_ALLOC_MEM_FLAGS_EXT_COHERENT)

		alloc_flags |= AMDGPU_GEM_CREATE_EXT_COHERENT;

	if (flags & KFD_IOC_ALLOC_MEM_FLAGS_UNCACHED)

		alloc_flags |= AMDGPU_GEM_CREATE_UNCACHED;

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

+1 −0
Original line number Diff line number Diff line
@@ -331,6 +331,7 @@ amdgpu_dma_buf_create_obj(struct drm_device *dev, struct dma_buf *dma_buf) 


		flags |= other->flags & (AMDGPU_GEM_CREATE_CPU_GTT_USWC |

					 AMDGPU_GEM_CREATE_COHERENT |

					 AMDGPU_GEM_CREATE_EXT_COHERENT |

					 AMDGPU_GEM_CREATE_UNCACHED);

	}

drivers/gpu/drm/amd/amdgpu/gmc_v10_0.c

+1 −0
Original line number Diff line number Diff line
@@ -635,6 +635,7 @@ static void gmc_v10_0_get_vm_pte(struct amdgpu_device *adev, 
	}



	if (bo && bo->flags & (AMDGPU_GEM_CREATE_COHERENT |

			       AMDGPU_GEM_CREATE_EXT_COHERENT |

			       AMDGPU_GEM_CREATE_UNCACHED))

		*flags = (*flags & ~AMDGPU_PTE_MTYPE_NV10_MASK) |

			 AMDGPU_PTE_MTYPE_NV10(MTYPE_UC);

drivers/gpu/drm/amd/amdgpu/gmc_v11_0.c

+1 −0
Original line number Diff line number Diff line
@@ -543,6 +543,7 @@ static void gmc_v11_0_get_vm_pte(struct amdgpu_device *adev, 
	}



	if (bo && bo->flags & (AMDGPU_GEM_CREATE_COHERENT |

			       AMDGPU_GEM_CREATE_EXT_COHERENT |

			       AMDGPU_GEM_CREATE_UNCACHED))

		*flags = (*flags & ~AMDGPU_PTE_MTYPE_NV10_MASK) |

			 AMDGPU_PTE_MTYPE_NV10(MTYPE_UC);

drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c

+4 −1
Original line number Diff line number Diff line
@@ -1187,7 +1187,8 @@ static void gmc_v9_0_get_coherence_flags(struct amdgpu_device *adev, 
{

	struct amdgpu_device *bo_adev = amdgpu_ttm_adev(bo->tbo.bdev);

	bool is_vram = bo->tbo.resource->mem_type == TTM_PL_VRAM;

	bool coherent = bo->flags & AMDGPU_GEM_CREATE_COHERENT;

	bool coherent = bo->flags & (AMDGPU_GEM_CREATE_COHERENT | AMDGPU_GEM_CREATE_EXT_COHERENT);

	bool ext_coherent = bo->flags & AMDGPU_GEM_CREATE_EXT_COHERENT;

	bool uncached = bo->flags & AMDGPU_GEM_CREATE_UNCACHED;

	struct amdgpu_vm *vm = mapping->bo_va->base.vm;

	unsigned int mtype_local, mtype;
@@ -1257,6 +1258,8 @@ static void gmc_v9_0_get_coherence_flags(struct amdgpu_device *adev, 
		snoop = true;

		if (uncached) {

			mtype = MTYPE_UC;

		} else if (ext_coherent) {

			mtype = is_local ? MTYPE_CC : MTYPE_UC;

		} else if (adev->flags & AMD_IS_APU) {

			mtype = is_local ? mtype_local : MTYPE_NC;

		} else {