On Tue, Oct 20, 2020 at 05:48:12PM +0200, Joerg Roedel wrote: > On Tue, Oct 20, 2020 at 10:12:59AM -0400, Arvind Sankar wrote: > > On Tue, Oct 20, 2020 at 02:18:54PM +0200, Joerg Roedel wrote: > > Why use r10-r12 rather than the caller-save registers? Even for the head > > code where you need to perserve the cr3 value you can just return it in > > rax? > > It can surely be optimized, but it makes the code less robust. This > function is only called from assembly so the standard x86-64 calling > conventions might not be followed strictly. I think its better to make > as few assumptions as possible about the calling code to avoid > regressions. Changes to the head code are not necessarily tested with > SEV/SEV-ES guests by developers. > > Regards, > > Joerg
This is called from both assembly and C, but anyway, you're already assuming r10 and r11 can be clobbered safely, and you just took out the save/restores in set_sev_encryption_mask, which is actually called only from assembly.
