Re: [PATCH] x86/xsave: Robustify and merge macros

[Quentin Casasnovas <quentin.casasnovas@oracle.com>]

On Thu, Apr 02, 2015 at 03:11:22PM +0200, Borislav Petkov wrote:
> From: Borislav Petkov <bp@suse.de>
> 
> Previously, we did call an XSAVE/XRSTOR variant through alternatives
> and did potential exception handling resulting from the instruction
> execution in a second inline asm. Which was misleading and error prone,
> see
> 
>   06c8173eb92b ("x86/fpu/xsaves: Fix improper uses of __ex_table")
> 
> for an example.
> 
> Add single macros which combine the alternatives and the exception
> handling.

FWIW I think this looks much nicer!  I have a couple of comments though,
apologies in advance if they aren't relevant :)

> 
> While at it, remove the SYSTEM_BOOTING checks in favor of
> static_cpu_has_safe() which works regardless of system state.
>

I thought the SYSTEM_BOOTING checks were present to make sure we call these
functions only when the alternative instructions had *not* been applied
(i.e. when SYSTEM_BOOTING).  We could have added the opposite checks in
xsave_state()/xrstor_state() to make sure the alternative instructions are
applied when these are called (i.e. when !SYSTEM_BOOTING).

In the unlikely event where I'm not wrong about this, having a nicely named
helper altinstr_are_applied() instead of manually checking the system_state
variable would probably help!

But maybe we're pretty confident this will not happen anyway?

> Cleanup comments.
> 
> Signed-off-by: Borislav Petkov <bp@suse.de>
> Cc: "H. Peter Anvin" <hpa@zytor.com>
> Cc: Ingo Molnar <mingo@kernel.org>
> Cc: Thomas Gleixner <tglx@linutronix.de>
> Cc: Quentin Casasnovas <quentin.casasnovas@oracle.com>
> Cc: Oleg Nesterov <oleg@redhat.com>
> Cc: Andy Lutomirski <luto@amacapital.net>
> ---
>  arch/x86/include/asm/xsave.h | 141 ++++++++++++++++++++++---------------------
>  1 file changed, 73 insertions(+), 68 deletions(-)
> 
> diff --git a/arch/x86/include/asm/xsave.h b/arch/x86/include/asm/xsave.h
> index c9a6d68b8d62..e6c7986c95df 100644
> --- a/arch/x86/include/asm/xsave.h
> +++ b/arch/x86/include/asm/xsave.h
> @@ -67,6 +67,66 @@ extern int init_fpu(struct task_struct *child);
>  			_ASM_EXTABLE(1b, 3b)		\
>  			: [err] "=r" (err)
>  
> +#define XSTATE_OP(op, st, lmask, hmask, err)				\
> +	asm volatile("1:" op "\n\t"					\
> +		     "2:\n\t"						\
> +		     "xor %[err], %[err]\n"				\

Are you not invariably clearing err here?  If the instruction fault, we go
to label '3' which does 'err = -1; goto 2', which clears err.  Same remark
for XSTATE_XSAVE()/XSTATE_RESTORE().

Probably missing something..

Also, tiny consistency nit, maybe use "\n\t" everywhere?

> +		     ".pushsection .fixup,\"ax\"\n\t"			\
> +		     "3: movl $-1,%[err]\n\t"				\
> +		     "jmp 2b\n\t"					\
> +		     ".popsection\n\t"					\
> +		     _ASM_EXTABLE(1b, 3b)				\
> +		     : [err] "=r" (err)					\
> +		     : "D" (st), "m" (*st), "a" (lmask), "d" (hmask)	\
> +		     : "memory")
> +

I've tried compiling this on top of v4.0-rc5 and I get a compile error
because alt_end_marker isn't defined.  Which other patches should I take to
test this?

Quentin