diff options
| author | David Francis <David.Francis@amd.com> | 2023-07-21 11:14:24 -0400 |
|---|---|---|
| committer | Alex Deucher <alexander.deucher@amd.com> | 2023-09-20 16:24:06 -0400 |
| commit | 5f248462c69d5ea7b54a88c62edb73f626d1528d (patch) | |
| tree | 8840e1bb0fcb471cf1c480513b3390c8e1b1b428 /include/uapi | |
| parent | d92e55565c53eff6d0c549090487d0a1ef704e1c (diff) | |
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: David Yat Sin <David.YatSin@amd.com>
Signed-off-by: David Francis <David.Francis@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Diffstat (limited to 'include/uapi')
| -rw-r--r-- | include/uapi/drm/amdgpu_drm.h | 10 | ||||
| -rw-r--r-- | include/uapi/linux/kfd_ioctl.h | 3 |
2 files changed, 12 insertions, 1 deletions
diff --git a/include/uapi/drm/amdgpu_drm.h b/include/uapi/drm/amdgpu_drm.h index ec437120b083..984fc16577ca 100644 --- a/include/uapi/drm/amdgpu_drm.h +++ b/include/uapi/drm/amdgpu_drm.h @@ -150,7 +150,7 @@ extern "C" { */ #define AMDGPU_GEM_CREATE_DISCARDABLE (1 << 12) /* Flag that BO is shared coherently between multiple devices or CPU threads. - * May depend on GPU instructions to flush caches explicitly + * May depend on GPU instructions to flush caches to system scope explicitly. * * This influences the choice of MTYPE in the PTEs on GFXv9 and later GPUs and * may override the MTYPE selected in AMDGPU_VA_OP_MAP. @@ -163,6 +163,14 @@ extern "C" { * may override the MTYPE selected in AMDGPU_VA_OP_MAP. */ #define AMDGPU_GEM_CREATE_UNCACHED (1 << 14) +/* Flag that BO should be coherent across devices when using device-level + * atomics. May depend on GPU instructions to flush caches to device scope + * explicitly, promoting them to system scope automatically. + * + * This influences the choice of MTYPE in the PTEs on GFXv9 and later GPUs and + * may override the MTYPE selected in AMDGPU_VA_OP_MAP. + */ +#define AMDGPU_GEM_CREATE_EXT_COHERENT (1 << 15) struct drm_amdgpu_gem_create_in { /** the requested memory size */ diff --git a/include/uapi/linux/kfd_ioctl.h b/include/uapi/linux/kfd_ioctl.h index eeb2fdcbdcb7..f0ed68974c54 100644 --- a/include/uapi/linux/kfd_ioctl.h +++ b/include/uapi/linux/kfd_ioctl.h @@ -405,6 +405,7 @@ struct kfd_ioctl_acquire_vm_args { #define KFD_IOC_ALLOC_MEM_FLAGS_AQL_QUEUE_MEM (1 << 27) #define KFD_IOC_ALLOC_MEM_FLAGS_COHERENT (1 << 26) #define KFD_IOC_ALLOC_MEM_FLAGS_UNCACHED (1 << 25) +#define KFD_IOC_ALLOC_MEM_FLAGS_EXT_COHERENT (1 << 24) /* Allocate memory for later SVM (shared virtual memory) mapping. * @@ -659,6 +660,8 @@ 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 +/* Fine grained coherency between all devices using device-scope atomics */ +#define KFD_IOCTL_SVM_FLAG_EXT_COHERENT 0x00000080 /** * kfd_ioctl_svm_op - SVM ioctl operations |