Re: [PATCH] Complete PPC INLINE_SYSCALL macro

[Franz Sirl <Franz dot Sirl-kernel at lauterbach dot com>]

On Tuesday 24 September 2002 22:55, Jakub Jelinek wrote:
> On Tue, Sep 24, 2002 at 10:45:51PM +0200, Franz Sirl wrote:
> > Hi,
> >
> > after comparing the INLINE_SYSCALL macro with the stuff we have in
> > asm/unistd.h, I noticed a few missing bits. Main thing are the missing
> > register clobbers which may lead to subtle bugs.
>
> Doesn't look right. E.g. with INLINE_SYSCALL(name, 2, ...) you
> have r4 both in clobbers and as input which is not allowed.

Huh? has this changed post-3.2? Or hasn't the documentation been updated? In 
3.2 it explicitly allows clobbering the inputs of a parallel, which is how 
the asm ends up as:

(insn 18 17 20 (parallel[
            (set (reg/v:SI 3 r3)
                (asm_operands:SI ("sc
        mfcr      %1
        ") ("=&r") 0[
                        (reg/v:SI 0 r0)
                        (reg/v:SI 3 r3)
                        (reg/v:SI 4 r4)
                    ]
                    [
                        (asm_input:SI ("1"))
                        (asm_input:SI ("0"))
                        (asm_input:SI ("r"))
                    ]  ("test-syscall.c") 56))
            (set (reg/v:SI 0 r0)
                (asm_operands:SI ("sc
        mfcr      %1
        ") ("=&r") 1[
                        (reg/v:SI 0 r0)
                        (reg/v:SI 3 r3)
                        (reg/v:SI 4 r4)
                    ]
                    [
                        (asm_input:SI ("1"))
                        (asm_input:SI ("0"))
                        (asm_input:SI ("r"))
                    ]  ("test-syscall.c") 56))
            (clobber (mem:BLK (scratch) [0 A8]))
            (clobber (reg:QI 68 cr0))
            (clobber (reg:QI 4 r4))
        ] ) -1 (nil)
    (nil))

So unless this has changed, the macro is valid, or?

Franz.