On 2/27/26 10:22, Khatri, Sunil wrote: > > On 27-02-2026 02:43 pm, Michel Dänzer wrote: >> On 2/26/26 17:05, Tvrtko Ursulin wrote: >>> On 26/02/2026 15:51, Sunil Khatri wrote: >>>> update the type for num_syncobj_handles from __u64 to _u16 with >>>> required padding. >>>> >>>> Signed-off-by: Sunil Khatri <[email protected]> >>>> --- >>>> include/uapi/drm/amdgpu_drm.h | 4 +++- >>>> 1 file changed, 3 insertions(+), 1 deletion(-) >>>> >>>> diff --git a/include/uapi/drm/amdgpu_drm.h b/include/uapi/drm/amdgpu_drm.h >>>> index 33e8738ce91f..42b4c0f6746b 100644 >>>> --- a/include/uapi/drm/amdgpu_drm.h >>>> +++ b/include/uapi/drm/amdgpu_drm.h >>>> @@ -479,7 +479,9 @@ struct drm_amdgpu_userq_signal { >>>> * @num_syncobj_handles: A count that represents the number of >>>> syncobj handles in >>>> * @syncobj_handles. >>>> */ >>>> - __u64 num_syncobj_handles; >>>> + __u16 num_syncobj_handles; >>> Could probably cause build warnings/errors for something which builds fine >>> today. Hm on big endian it actually breaks completely. >> Yeah this is a clear UAPI break, no go. > > Yeah i am holding on this change for a cleaner solution. I acknowledge the > UAPI breaking here.
This uAPI break was intentional. The functionality is used only as experimental feature behind module parameter and new FW anyway. But we didn't wanted to break existing Mesa binaries for easier testing and so kept the ABI for x86 intact with the padding. Regards, Christian. > > Regards > > Sunil khatri > >> >> >>>> + __u16 pad0; >>>> + __u32 pad1; >>>> /** >>>> * @bo_read_handles: The list of BO handles that the submitted user >>>> queue job >>>> * is using for read only. This will update BO fences in the kernel. >>
