Re: [PATCH] x86/retpoline: Fill return stack buffer on vmexit

["Woodhouse, David" <dwmw@amazon.co.uk>]

[-- Attachment #1: Type: text/plain, Size: 1118 bytes --]

On Thu, 2018-01-11 at 10:47 +0100, Borislav Petkov wrote:
> On Thu, Jan 11, 2018 at 10:32:31AM +0100, Peter Zijlstra wrote:
> > 
> > can't you do lovely things like:
> > 
> > 	volatile asm ("call __fill_rsb_thunk_%1" : : "r" (dummy))
> > 
> > which would still let gcc select the register ?

I've had to do that for the __x86_indirect_thunk_\reg thunks and
provide all variants thereof, and export them to modules.

I'd much rather have this one inline. If I couldn't do that (which I
can, and have), then my next choice would probably have been to just
hard-code it to use %eax/%r12 and emit a simple call to that from the
call sites.

> Calling a function from asm is nasty because you need to pay attention
> to clobbered registers as gcc doesn't see the function.
>
> What one can do, I *think*, is do a non-inlined wrapping function and do
> all the alternative_call() fun inside. There you can do all the fun and
> have callee-clobbered regs which you can use.

That's OK; it's not really a C function and it only clobbers the *one*
register which is easy enough to tell __asm__() about.

[-- Attachment #2: smime.p7s --]
[-- Type: application/x-pkcs7-signature, Size: 5210 bytes --]