Re: [PATCH v3 5/7] x86: guard against straight-line speculation past RET

[Roger Pau Monné <roger.pau@xxxxxxxxxx>]

On Wed, Nov 11, 2020 at 02:33:34PM +0100, Jan Beulich wrote:
> On 11.11.2020 12:15, Roger Pau Monné wrote:
> > On Fri, Oct 23, 2020 at 10:38:04AM +0200, Jan Beulich wrote:
> >> Under certain conditions CPUs can speculate into the instruction stream
> >> past a RET instruction. Guard against this just like 3b7dab93f240
> >> ("x86/spec-ctrl: Protect against CALL/JMP straight-line speculation")
> >> did - by inserting an "INT $3" insn. It's merely the mechanics of how to
> >> achieve this that differ: A set of macros gets introduced to post-
> >> process RET insns issued by the compiler (or living in assembly files).
> >>
> >> Unfortunately for clang this requires further features their built-in
> >> assembler doesn't support: We need to be able to override insn mnemonics
> >> produced by the compiler (which may be impossible, if internally
> >> assembly mnemonics never get generated), and we want to use \(text)
> >> escaping / quoting in the auxiliary macro.
> >>
> >> Signed-off-by: Jan Beulich <jbeulich@xxxxxxxx>
> >> Acked-by: Roger Pau Monné <roger.pau@xxxxxxxxxx>
> >> ---
> >> TBD: Would be nice to avoid the additions in .init.text, but a query to
> >>      the binutils folks regarding the ability to identify the section
> >>      stuff is in (by Peter Zijlstra over a year ago:
> >>      https://sourceware.org/pipermail/binutils/2019-July/107528.html)
> >>      has been left without helpful replies.
> >> ---
> >> v3: Use .byte 0xc[23] instead of the nested macros.
> >> v2: Fix build with newer clang. Use int3 mnemonic. Also override retq.
> >>
> >> --- a/xen/Makefile
> >> +++ b/xen/Makefile
> >> @@ -145,7 +145,15 @@ t2 = $(call as-insn,$(CC) -I$(BASEDIR)/i
> >>  # https://bugs.llvm.org/show_bug.cgi?id=36110
> >>  t3 = $(call as-insn,$(CC),".macro FOO;.endm"$(close); asm volatile 
> >> $(open)".macro FOO;.endm",-no-integrated-as)
> >>  
> >> -CLANG_FLAGS += $(call or,$(t1),$(t2),$(t3))
> >> +# Check whether \(text) escaping in macro bodies is supported.
> >> +t4 = $(call as-insn,$(CC),".macro m ret:req; \\(ret) $$\\ret; .endm; m 
> >> 8",,-no-integrated-as)
> >> +
> >> +# Check whether macros can override insn mnemonics in inline assembly.
> >> +t5 = $(call as-insn,$(CC),".macro ret; .error; .endm; .macro retq; 
> >> .error; .endm",-no-integrated-as)
> > 
> > I was going over this to post a bug report to LLVM, but it seems like
> > gcc also doesn't overwrite ret when using the above snippet:
> > 
> > https://godbolt.org/z/oqsPTv
> 
> I can't see what's different from
> 
> void test(void) {
>       asm volatile (".macro ret; .error; .endm; .macro retq; .error; .endm");
> }
> 
> but this one produces "Error: .error directive invoked in source file"
> for me with both old and new gcc.

You are right, I think godbolt is somehow busted?

I can reproduce your results with my version of gcc, so will just
report to LLVM.

Roger.