Linux Virtualization - Nov 2017 - [PATCH 03/13] x86/paravirt: Convert native patch assembly code strings to macros

On Fri, Nov 17, 2017 at 08:10:13PM +0100, Juergen Gross
wrote:
> On 17/11/17 19:07, Borislav Petkov wrote:
> > On Wed, Oct 04, 2017 at 10:58:24AM -0500, Josh Poimboeuf wrote:
> >> Convert the hard-coded native patch assembly code strings to
macros to
> >> facilitate sharing common code between 32-bit and 64-bit.
> >>
> >> These macros will also be used by a future patch which requires
the GCC
> >> extended asm syntax of two '%' characters instead of one
when specifying
> >> a register name.
> >>
> >> Signed-off-by: Josh Poimboeuf <jpoimboe at redhat.com>
> >> ---
> >>  arch/x86/include/asm/special_insns.h | 24
++++++++++++++++++++++++
> >>  arch/x86/kernel/paravirt_patch_32.c  | 21 +++++++++++----------
> >>  arch/x86/kernel/paravirt_patch_64.c  | 29
+++++++++++++++--------------
> >>  3 files changed, 50 insertions(+), 24 deletions(-)
> >>
> >> diff --git a/arch/x86/include/asm/special_insns.h
b/arch/x86/include/asm/special_insns.h
> >> index ac402c6fc24b..0549c5f2c1b3 100644
> >> --- a/arch/x86/include/asm/special_insns.h
> >> +++ b/arch/x86/include/asm/special_insns.h
> >> @@ -6,6 +6,30 @@
> >>  
> >>  #include <asm/nops.h>
> >>  
> >> +#ifdef CONFIG_X86_64
> >> +# define _REG_ARG1			"%rdi"
> >> +# define NATIVE_IDENTITY_32		"mov %edi, %eax"
> > 
> > Yeah, that "identity" looks strange. How about NATIVE_NOOP
and
> > NATIVE_NOOP_32 ?
> 
> Those are not NOPs. They return the identical value which was passed to
> them. So identity isn't a bad name after all.
Right, like the math identity function:

  https://en.wikipedia.org/wiki/Identity_function
> >> +# define NATIVE_USERGS_SYSRET64		"swapgs; sysretq"
> >> +#else
> >> +# define _REG_ARG1			"%eax"
> >> +#endif
> >> +
> >> +#define _REG_RET			"%" _ASM_AX
> >> +
> >> +#define NATIVE_ZERO			"xor " _REG_ARG1 ", "
_REG_ARG1
> > 
> > NATIVE_ZERO_OUT
> > 
> > I guess. NATIVE_ZERO reads like the native representation of 0 :-)
> 
> NATIVE_ZERO_ARG1?
On a slight tangent, does anybody know why it zeros the arg?

The only place it's used is here:

#if defined(CONFIG_PARAVIRT_SPINLOCKS)
DEF_NATIVE(pv_lock_ops,	queued_spin_unlock,	NATIVE_QUEUED_SPIN_UNLOCK);
DEF_NATIVE(pv_lock_ops,	vcpu_is_preempted,	NATIVE_ZERO);
#endif

Isn't that a bug?  Seems like it should _return_ zero.  Zeroing the arg
shouldn't have any effect.

If I'm right, we could call it NATIVE_FALSE.

-- 
Josh

On 17/11/17 20:42, Josh Poimboeuf wrote:
> On Fri, Nov 17, 2017 at 08:10:13PM +0100, Juergen Gross wrote:
>> On 17/11/17 19:07, Borislav Petkov wrote:
>>> On Wed, Oct 04, 2017 at 10:58:24AM -0500, Josh Poimboeuf wrote:
>>>> Convert the hard-coded native patch assembly code strings to
macros to
>>>> facilitate sharing common code between 32-bit and 64-bit.
>>>>
>>>> These macros will also be used by a future patch which requires
the GCC
>>>> extended asm syntax of two '%' characters instead of
one when specifying
>>>> a register name.
>>>>
>>>> Signed-off-by: Josh Poimboeuf <jpoimboe at redhat.com>
>>>> ---
>>>>  arch/x86/include/asm/special_insns.h | 24
++++++++++++++++++++++++
>>>>  arch/x86/kernel/paravirt_patch_32.c  | 21
+++++++++++----------
>>>>  arch/x86/kernel/paravirt_patch_64.c  | 29
+++++++++++++++--------------
>>>>  3 files changed, 50 insertions(+), 24 deletions(-)
>>>>
>>>> diff --git a/arch/x86/include/asm/special_insns.h
b/arch/x86/include/asm/special_insns.h
>>>> index ac402c6fc24b..0549c5f2c1b3 100644
>>>> --- a/arch/x86/include/asm/special_insns.h
>>>> +++ b/arch/x86/include/asm/special_insns.h
>>>> @@ -6,6 +6,30 @@
>>>>  
>>>>  #include <asm/nops.h>
>>>>  
>>>> +#ifdef CONFIG_X86_64
>>>> +# define _REG_ARG1			"%rdi"
>>>> +# define NATIVE_IDENTITY_32		"mov %edi, %eax"
>>>
>>> Yeah, that "identity" looks strange. How about
NATIVE_NOOP and
>>> NATIVE_NOOP_32 ?
>>
>> Those are not NOPs. They return the identical value which was passed to
>> them. So identity isn't a bad name after all.
> 
> Right, like the math identity function:
> 
>   https://en.wikipedia.org/wiki/Identity_function
> 
>>>> +# define NATIVE_USERGS_SYSRET64		"swapgs; sysretq"
>>>> +#else
>>>> +# define _REG_ARG1			"%eax"
>>>> +#endif
>>>> +
>>>> +#define _REG_RET			"%" _ASM_AX
>>>> +
>>>> +#define NATIVE_ZERO			"xor " _REG_ARG1 ",
" _REG_ARG1
>>>
>>> NATIVE_ZERO_OUT
>>>
>>> I guess. NATIVE_ZERO reads like the native representation of 0 :-)
>>
>> NATIVE_ZERO_ARG1?
> 
> On a slight tangent, does anybody know why it zeros the arg?
Why are _you_ asking? You've introduced it.
> The only place it's used is here:
> 
> #if defined(CONFIG_PARAVIRT_SPINLOCKS)
> DEF_NATIVE(pv_lock_ops,	queued_spin_unlock,	NATIVE_QUEUED_SPIN_UNLOCK);
> DEF_NATIVE(pv_lock_ops,	vcpu_is_preempted,	NATIVE_ZERO);
> #endif
> 
> Isn't that a bug?  Seems like it should _return_ zero.  Zeroing the arg
> shouldn't have any effect.
Right. Before that patch it _did_ return zero instead of zeroing arg1.
> If I'm right, we could call it NATIVE_FALSE.
I'd prefer NATIVE_ZERO, as it will be usable for non-boolean cases, too.


Juergen

On Sat, Nov 18, 2017 at 11:20:06AM +0100, Juergen Gross
wrote:
> >>>> +#define NATIVE_ZERO			"xor " _REG_ARG1 ",
" _REG_ARG1
> >>>
> >>> NATIVE_ZERO_OUT
> >>>
> >>> I guess. NATIVE_ZERO reads like the native representation of 0
:-)
> >>
> >> NATIVE_ZERO_ARG1?
> > 
> > On a slight tangent, does anybody know why it zeros the arg?
> 
> Why are _you_ asking? You've introduced it.
So I did.  Touch?!
> > The only place it's used is here:
> > 
> > #if defined(CONFIG_PARAVIRT_SPINLOCKS)
> > DEF_NATIVE(pv_lock_ops,	queued_spin_unlock,
NATIVE_QUEUED_SPIN_UNLOCK);
> > DEF_NATIVE(pv_lock_ops,	vcpu_is_preempted,	NATIVE_ZERO);
> > #endif
> > 
> > Isn't that a bug?  Seems like it should _return_ zero.  Zeroing
the arg
> > shouldn't have any effect.
> 
> Right. Before that patch it _did_ return zero instead of zeroing arg1.
Oops!
> > If I'm right, we could call it NATIVE_FALSE.
> 
> I'd prefer NATIVE_ZERO, as it will be usable for non-boolean cases,
too.
NATIVE_ZERO works for me.

-- 
Josh