On 2022/9/1 11:36, Huang, Kai wrote: > On Thu, 2022-09-01 at 08:35 +0800, Zhiquan Li wrote: >> In order to avoid unnecessary casting, rename the 'owner' field of >> struct sgx_epc_page as 'encl_owner', and update all of references. > This changelog itself doesn't explain _why_ renaming 'owner' to 'encl_owner' can > avoid the explicit castg. In fact, the reason is that you will use a 'union' > to separate the use of 'owner' of the EPC page (between SGX driver and virtual > EPC). The rename is just to make the name more clear, but cannot really avoid > casting. > > So I think there should be more sentences to explain here, such as "in order to > send SIGBUS to userspace hypervisor to allow it to inject #MC to guest, use > virtual EPC page's owner to be the userspace virtual address of the EPC page", > and after those you can say something like "in order to avoid casting, use a > union to separate the use of owner for SGX driver EPC page and virtual EPC page. > And rename owner of SGX driver EPC page to 'encl_owner' to be more specific", > etc. > > That being said, I guess you can just merge this patch with your second patch, > which actually introduces the 'union' and uses the owner for virtual EPC page. > And in changelog you explain everything above to justify the patch. > > Does this make sense? > Many thanks for your excellent explanation, I would like to add it into the commit message of this patch but keep it separately. IMHO, this will benefit cherry-pick the subsequent patch to stable tree if it bases on the new 'encl_owner' field but no related to this MCA fix. In such case just cherry-pick patch 01 and the critical fix patch is enough, patch 02 and 03 can be ignored. What do you think? Furthermore, I would like to add your “Acked-by” to this patch, is that OK for you? Thanks and Best Regards, Zhiquan
