Re: [PATCH 25/32] s390/pci: Convert inline assembly to EX_TABLE_INSN

[Niklas Schnelle <schnelle@linux.ibm.com>]

On Thu, 2026-10-01 at 12:32 +0200, Heiko Carstens wrote:
> Convert all pci exception table entries to use EX_TABLE_INSN.
> With EX_TABLE_INSN entries always point to the faulting instruction,
> so duplicate entries are removed.
> 
> Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
> ---
>  arch/s390/pci/pci_clp.c  | 12 ++++++------
>  arch/s390/pci/pci_insn.c | 36 ++++++++++++++++++------------------
>  arch/s390/pci/pci_mmio.c | 28 +++++++++++++++-------------
>  3 files changed, 39 insertions(+), 37 deletions(-)
--- snip ---
> diff --git a/arch/s390/pci/pci_mmio.c b/arch/s390/pci/pci_mmio.c
> index f3f79ba78410..8b49175b0031 100644
> --- a/arch/s390/pci/pci_mmio.c
> +++ b/arch/s390/pci/pci_mmio.c
> @@ -39,10 +39,10 @@ static inline int __pcistb_mio_inuser(
>  	asm_inline volatile (
>  		"	sacf	256\n"
>  		"0:	.insn	rsy,0xeb00000000d4,%[len],%[ioaddr],%[src]\n"
> -		"1:	lhi	%[exc],0\n"
> -		"2:	sacf	768\n"
> +		"	lhi	%[exc],0\n"
> +		"1:	sacf	768\n"
>  		CC_IPM(cc)
> -		EX_TABLE(0b, 2b) EX_TABLE(1b, 2b)
> +		EX_TABLE_INSN(0b, 1b)
>  		: CC_OUT(cc, cc), [len] "+d" (len), [exc] "+d" (exception)
>  		: [ioaddr] "a" (ioaddr), [src] "Q" (*((u8 __force *)src))
>  		: CC_CLOBBER_LIST("memory"));
> @@ -72,15 +72,16 @@ static inline int __pcistg_mio_inuser(
>  	asm_inline volatile (
>  		"	sacf	256\n"
>  		"0:	llgc	%[tmp],0(%[src])\n"
> -		"4:	sllg	%[val],%[val],8\n"
> +		"	sllg	%[val],%[val],8\n"
>  		"	aghi	%[src],1\n"
>  		"	ogr	%[val],%[tmp]\n"
>  		"	brctg	%[cnt],0b\n"
>  		"1:	.insn	rre,0xb9d40000,%[val],%[ioaddr_len]\n"
> -		"2:	lhi	%[exc],0\n"
> -		"3:	sacf	768\n"
> +		"	lhi	%[exc],0\n"
> +		"2:	sacf	768\n"
>  		CC_IPM(cc)
> -		EX_TABLE(0b, 3b) EX_TABLE(4b, 3b) EX_TABLE(1b, 3b) EX_TABLE(2b, 3b)
> +		EX_TABLE_INSN(0b, 2b)
> +		EX_TABLE_INSN(1b, 2b)

:) this and the one in __pcilg_mio_inuser() must be one of the best
examples why the new approach is cleaner and easier to read. Well done!

>  		: [src] "+a" (src), [cnt] "+d" (cnt),
>  		  [val] "+d" (val), [tmp] "=d" (tmp), [exc] "+d" (exception),
>  		  CC_OUT(cc, cc), [ioaddr_len] "+&d" (ioaddr_len.pair)
> @@ -227,13 +228,13 @@ static inline int __pcilg_mio_inuser(
>  	asm_inline volatile (
>  		"	sacf	256\n"
>  		"0:	.insn	rre,0xb9d60000,%[val],%[ioaddr_len]\n"
> -		"1:	lhi	%[exc],0\n"
> +		"	lhi	%[exc],0\n"
>  		"	jne	4f\n"
> -		"2:	ahi	%[shift],-8\n"
> +		"1:	ahi	%[shift],-8\n"
>  		"	srlg	%[tmp],%[val],0(%[shift])\n"
> -		"3:	stc	%[tmp],0(%[dst])\n"
> -		"5:	aghi	%[dst],1\n"
> -		"	brctg	%[cnt],2b\n"
> +		"2:	stc	%[tmp],0(%[dst])\n"
> +		"	aghi	%[dst],1\n"
> +		"	brctg	%[cnt],1b\n"
>  		/*
>  		 * Use xr to clear exc and set condition code to zero
>  		 * to ensure flag output is correct for this branch.
> @@ -241,7 +242,8 @@ static inline int __pcilg_mio_inuser(
>  		"	xr	%[exc],%[exc]\n"
>  		"4:	sacf	768\n"
>  		CC_IPM(cc)
> -		EX_TABLE(0b, 4b) EX_TABLE(1b, 4b) EX_TABLE(3b, 4b) EX_TABLE(5b, 4b)
> +		EX_TABLE_INSN(0b, 4b)
> +		EX_TABLE_INSN(2b, 4b)
>  		: [ioaddr_len] "+&d" (ioaddr_len.pair), [exc] "+d" (exception),
>  		  CC_OUT(cc, cc), [val] "=d" (val),
>  		  [dst] "+a" (dst), [cnt] "+d" (cnt), [tmp] "=d" (tmp),

I went through the adjustments and they all look correct under the
explanations you gave in the cover letter and prerequisite patches.

So feel free to add my:

Reviewed-by: Niklas Schnelle <schnelle@linux.ibm.com>

Thanks,
Niklas