Hi Josh,

On Wed, 28 Aug 2019 11:34:33 -0500
Josh Poimboeuf <jpoimboe@redhat.com> wrote:

> On Wed, Aug 28, 2019 at 11:13:31AM -0500, Josh Poimboeuf wrote:
> > Turns out this patch does break something:
> > 
> >   arch/x86/xen/enlighten_pv.o: warning: objtool: xen_cpuid()+0x25: can't find jump dest instruction at .text+0x9c
> > 
> > I'll need to figure out a better way to whitelist that
> > XEN_EMULATE_PREFIX fake instruction thing.  I'll probably just teach
> > the objtool decoder about it.
> 
> Hi Masami,
> 
> Is it possible for the kernel x86 decoder to recognize the
> XEN_EMULATE_PREFIX prefix?
> 
>         asm(XEN_EMULATE_PREFIX "cpuid"
>                 : "=a" (*ax),
>                   "=b" (*bx),
>                   "=c" (*cx),
>                   "=d" (*dx)
>                 : "0" (*ax), "2" (*cx));
> 
> is disassembled to:
> 
>       33:       0f 0b                   ud2
>       35:       78 65                   js     9c <xen_store_tr+0xc>
>       37:       6e                      outsb  %ds:(%rsi),(%dx)
>       38:       0f a2                   cpuid
> 
> which confuses objtool.  Presumably that would confuse other users of
> the decoder as well.

Good catch! It should be problematic, since x86 decoder sanity test is
based on objtool. But I don't want to change the test code itself,
because this problem is highly depending on Xen.

> That's a highly unlikely sequence of instructions, maybe the kernel
> decoder should recognize it as a single instruction.

OK, it is better to be done in decoder (only for CONFIG_XEN_PVHVM)

BTW, could you also share what test case would you using?

And what about attached patch? (just compile checked with/without CONFIG_XEN_PVHVM)

Thank you,


-- 
Masami Hiramatsu <mhiramat@kernel.org>