Re: [PATCH 09/17] m68k: Implement xor_unlock_is_negative_byte

[Greg Ungerer]

On 3/10/23 06:07, Matthew Wilcox wrote:
> On Wed, Sep 20, 2023 at 05:22:33PM +1000, Greg Ungerer wrote:
> > On 20/9/23 01:14, Matthew Wilcox wrote:
> > > I have a 68020 book; what I don't have is a Coldfire manual.
> >
> > You can find it here: https://www.nxp.com/docs/en/reference-manual/CFPRM.pdf
>
> Thanks, Greg.  This is almost good:
>
> static inline bool xor_unlock_is_negative_byte(unsigned long mask,
>                  volatile unsigned long *p)
> {
> #ifdef CONFIG_COLDFIRE
>          __asm__ __volatile__ ("eorl %1, %0"
>                  : "+m" (*p)
>                  : "d" (mask)
>                  : "memory");
>          return *p & (1 << 7);
> #else
>          char result;
>          char *cp = (char *)p + 3;       /* m68k is big-endian */
>
>          __asm__ __volatile__ ("eor.b %1, %2; smi %0"
>                  : "=d" (result)
>                  : "di" (mask), "o" (*cp)
>                  : "memory");
>          return result;
> #endif
> }
>
> folio_end_read() does about as well as can be expected:
>
> 00000708 <folio_end_read>:
>       708:       206f 0004       moveal %sp@(4),%a0
>       70c:       7009            moveq #9,%d0
>       70e:       4a2f 000b       tstb %sp@(11)
>       712:       6602            bnes 716 <folio_end_read+0xe>
>       714:       7001            moveq #1,%d0
>       716:       b190            eorl %d0,%a0@
>       718:       2010            movel %a0@,%d0
>       71a:       4a00            tstb %d0
>       71c:       6a0c            bpls 72a <folio_end_read+0x22>
>       71e:       42af 0008       clrl %sp@(8)
>       722:       2f48 0004       movel %a0,%sp@(4)
>       726:       6000 fcfe       braw 426 <folio_wake_bit>
>       72a:       4e75            rts
>
> However, it seems that folio_unlock() could shave off an instruction:
>
> 00000918 <folio_unlock>:
>       918:       206f 0004       moveal %sp@(4),%a0
>       91c:       7001            moveq #1,%d0
>       91e:       b190            eorl %d0,%a0@
>       920:       2010            movel %a0@,%d0
>       922:       4a00            tstb %d0
>       924:       6a0a            bpls 930 <folio_unlock+0x18>
>       926:       42a7            clrl %sp@-
>       928:       2f08            movel %a0,%sp@-
>       92a:       4eba fafa       jsr %pc@(426 <folio_wake_bit>)
>       92e:       508f            addql #8,%sp
>       930:       4e75            rts
>
> We could use eori instead of eorl, at least according to table 3-9 on
> page 3-8:
>
> EOR Dy,<ea>x L Source ^ Destination → Destination ISA_A
> EORI #<data>,Dx L Immediate Data ^ Destination → Destination ISA_A
>
> but gas is unhappy with everything I've tried to use eori.  I'm building

I can't seem to get it to always use it either. This comes close:

        __asm__ __volatile__ ("eorl %1, %0"
                : "+d" (*p)
                : "di" (mask)
                : "memory");
        return *p & (1 << 7);

Using eoril for folio_unlock, but not for folio_end_read:

400413e6 <folio_unlock>:
400413e6:       206f 0004       moveal %sp@(4),%a0
400413ea:       2010            movel %a0@,%d0
400413ec:       0a80 0000 0001  eoril #1,%d0
400413f2:       2080            movel %d0,%a0@
400413f4:       2010            movel %a0@,%d0
400413f6:       4a00            tstb %d0
400413f8:       6c0a            bges 40041404 <folio_unlock+0x1e>
400413fa:       42a7            clrl %sp@-
400413fc:       2f08            movel %a0,%sp@-
400413fe:       4eba ff30       jsr %pc@(40041330 <folio_wake_bit>)
40041402:       508f            addql #8,%sp
40041404:       4e75            rts

But that is still worse anyway.

> with stmark2_defconfig, which I assume should work.

Yes, or any of amcore, m5208evb, m5249evb, m5272c3, m5275evb, m5307c3, m5407c3.

Regards
Greg