Re: [PATCH -final] x86/clear_user: Make it faster

[Alexey Dobriyan <adobriyan@gmail.com>]

On Mon, Jul 11, 2022 at 12:33:20PM +0200, Borislav Petkov wrote:
> On Wed, Jul 06, 2022 at 12:24:12PM +0300, Alexey Dobriyan wrote:
> > On Tue, Jul 05, 2022 at 07:01:06PM +0200, Borislav Petkov wrote:
> > 
> > > +	asm volatile(
> > > +		"1:\n\t"
> > > +		ALTERNATIVE_3("rep stosb",
> > > +			      "call clear_user_erms",	  ALT_NOT(X86_FEATURE_FSRM),
> > > +			      "call clear_user_rep_good", ALT_NOT(X86_FEATURE_ERMS),
> > > +			      "call clear_user_original", ALT_NOT(X86_FEATURE_REP_GOOD))
> > > +		"2:\n"
> > > +	       _ASM_EXTABLE_UA(1b, 2b)
> > > +	       : "+&c" (size), "+&D" (addr), ASM_CALL_CONSTRAINT
> > > +	       : "a" (0)
> > > +		/* rep_good clobbers %rdx */
> > > +	       : "rdx");
> > 
> > "+c" and "+D" should be enough for 1 instruction assembly?
> 
> I'm looking at
> 
>   e0a96129db57 ("x86: use early clobbers in usercopy*.c")
> 
> which introduced the early clobbers and I'm thinking we want them
> because "this operand is an earlyclobber operand, which is written
> before the instruction is finished using the input operands" and we have
> exception handling.
> 
> But maybe you need to be more verbose as to what you mean exactly...

This is the original code:

-#define __do_strncpy_from_user(dst,src,count,res)                         \
-do {                                                                      \
-       long __d0, __d1, __d2;                                             \
-       might_fault();                                                     \
-       __asm__ __volatile__(                                              \
-               "       testq %1,%1\n"                                     \
-               "       jz 2f\n"                                           \
-               "0:     lodsb\n"                                           \
-               "       stosb\n"                                           \
-               "       testb %%al,%%al\n"                                 \
-               "       jz 1f\n"                                           \
-               "       decq %1\n"                                         \
-               "       jnz 0b\n"                                          \
-               "1:     subq %1,%0\n"                                      \
-               "2:\n"                                                     \
-               ".section .fixup,\"ax\"\n"                                 \
-               "3:     movq %5,%0\n"                                      \
-               "       jmp 2b\n"                                          \
-               ".previous\n"                                              \
-               _ASM_EXTABLE(0b,3b)                                        \
-               : "=&r"(res), "=&c"(count), "=&a" (__d0), "=&S" (__d1),    \
-                 "=&D" (__d2)                                             \
-               : "i"(-EFAULT), "0"(count), "1"(count), "3"(src), "4"(dst) \
-               : "memory");                                               \
-} while (0)

I meant to say that earlyclobber is necessary only because the asm body
is more than 1 instruction so there is possibility of writing to some
outputs before all inputs are consumed.

If asm body is 1 insn there is no such possibility at all.
Now "rep stosb" is 1 instruction and two alterantive functions masquarade
as single instruction.