Re: [PATCH 03/20] x86/uaccess: Move copy_user_handle_tail into asm

[Peter Zijlstra <peterz@infradead.org>]

On Fri, Mar 08, 2019 at 01:53:02PM -0600, Josh Poimboeuf wrote:
> On Fri, Mar 08, 2019 at 08:48:35PM +0100, Peter Zijlstra wrote:
> > On Fri, Mar 08, 2019 at 12:53:21PM -0600, Josh Poimboeuf wrote:
> > > On Thu, Mar 07, 2019 at 12:45:14PM +0100, Peter Zijlstra wrote:
> > > >  /*
> > > > + * Try to copy last bytes and clear the rest if needed.
> > > > + * Since protection fault in copy_from/to_user is not a normal situation,
> > > > + * it is not necessary to optimize tail handling.
> > > > + *
> > > > + * Input:
> > > > + * rdi destination
> > > > + * rsi source
> > > > + * rdx count
> > > > + *
> > > > + * Output:
> > > > + * eax uncopied bytes or 0 if successful.
> > > > + */
> > > > +ALIGN;
> > > > +copy_user_handle_tail:
> > > > +	movl %edx,%ecx
> > > > +1:	rep movsb
> > > > +2:	mov %ecx,%eax
> > > > +	ASM_CLAC
> > > > +	ret
> > > > +
> > > > +	_ASM_EXTABLE_UA(1b, 2b)
> > > > +ENDPROC(copy_user_handle_tail)
> > > 
> > > This is an unstructured piece of code rather than a callable function,
> > > END would probably be more appropriate.  Or maybe it should just be a
> > > local label (.Lcopy_user_handle_tail) because I don't think the
> > > alignment and ELF symbol size are even needed.
> > 
> > ENDPROC makes it STT_FUNC and gets us stricter AC tests.
> 
> How so?  I would have thought the opposite.  Doesn't objtool only follow
> a jump if its destination is to a non-function?  Otherwise it's
> considered a sibling call.

Normally yes, but we don't do that for .fixup I think. And by setting
STT_FUNC we enable the 'redundant CLAC' warning, which is ignored for
!STT_FUNC.