Hi Dave, This feature is used on Virtualization environment to send errors of amdgpu initialization fail from Guest side to Host side via mailbox (function xgpu_ai_mailbox_trans_msg in file mxgpu_ai.c).
In Virtualization environment, it should not use multi GPUs. So I just created a global array not to impacts amdgpu structure. The array size is 16, operated as ring buffer. It is bigger enough to save the errors of amdgpu initialization fail. You are right, I should move struct amdgpu_vf_error_buffer to struct amdgpu_device OR struct amdgpu_virt. Which one is better? Yes, there are a lot typoes adm->amd and I will fix them. Thanks your very much. Best Regards, Gavin -----Original Message----- From: Dave Airlie [mailto:airl...@gmail.com] Sent: Wednesday, September 27, 2017 7:17 PM To: amd-gfx mailing list <amd-gfx@lists.freedesktop.org>; Wan, Gavin <gavin....@amd.com> Subject: Re: amdgpu_vf_errors On 28 September 2017 at 09:14, Dave Airlie <airl...@gmail.com> wrote: > I've no idea what this is used for, virtual function errors? > > but why does it have no locking, and why is there a global array, (not per > gpu?) > > Alex, you reviewed it, please rewrite/remove as necessary. Oh it also typoes adm->amd in lots of places. Dave. > > Dave. _______________________________________________ amd-gfx mailing list amd-gfx@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/amd-gfx
