Ok, its a bad comment :).
The point is that, the hypervisor cannot depend on the OSes above
us to use VRSAVE at all, so we have to S&R the whole shebang.
Ah, much clearer already.
Do we really have to? Can't we, for example, define the supervisor
vs. hypervisor ABI wrt. VMX to be "if the hypervisor wants to use
any VMX regs, it better S&R those itself"? And when switching
supervisor context, just swap-em all out. Although perhaps this is
just what the full code is doing, souldn't tell from the patch alone :-)
I'll write a clearer comment.
/* r0 as the second operand is considered 0 */
NOT AT ALL. Binary pattern 0b00000 for RA means literal 0 instead
of GPR0, sure. And the assembler you use just uses CCCP to
translate "r0" into a literal 0. But some other assemblers
do not. So please just write "0" when you mean "0".
No, _thank you_. Let's exterminate this disease before it spreads
even further :-)
Xen-ppc-devel mailing list