Re: Re [PATCH] x86/CET: Fix build following c/s 43b98e7190

[Jan Beulich <jbeulich@xxxxxxxx>]

On 02.06.2020 19:15, Andrew Cooper wrote:
> On 02/06/2020 15:21, Jan Beulich wrote:
>>> OSSTest reports:
>>>
>>>   x86_64.S: Assembler messages:
>>>   x86_64.S:57: Error: no such instruction: `setssbsy'
>>>   /home/osstest/build.150510.build-amd64/xen/xen/Rules.mk:183: recipe for 
>>> target 'head.o' failed
>>>   make[4]: Leaving directory 
>>> '/home/osstest/build.150510.build-amd64/xen/xen/arch/x86/boot'
>>>   make[4]: *** [head.o] Error 1
>>>
>>> All use of CET instructions, even those inside alternative blocks, needs to 
>>> be
>>> behind CONFIG_XEN_SHSTK, as it indicates suitable toolchain support.
>>>
>>> Signed-off-by: Andrew Cooper <andrew.cooper3@xxxxxxxxxx>
>> That's quite a bit of #ifdef-ary here. Simple (operand-less) insns
>> like SETSSBSY could easily be made available via .byte directives.
>> Would you be amenable to to ack-ing a patch to replace some of the
>> #ifdef-s (at least the ones at the lstar, cstar, and sysenter
>> entry points), after 4.14?
> 
> Yeah - that was a bit of a mess in the end.  (But given the
> circumstances, and that I've got past form typo'ing the SETSSBSY opcode,
> it probably was the right move even in hindsight).
> 
> Reducing it to .byte should be fine so long as some form of /* setssbsy
> */ comment appears.

Sure.

> One other option would be to introduce a SETSSBSY macro, but that hides
> the alternative so is something I'd prefer to avoid.

With this you mean you'd rather not see us go the CLAC/STAC route?
I was instead thinking of a pure assembly macro named "setssbsy".
In fact we could switch the CLAC/STAC ugliness to some such, if we
end up being happy with the model.

Jan