Re: [PATCH v2 02/17] arm64/lib: copy_page: avoid x18 register in assembler code

From: Mark Rutland
Date: Fri Oct 25 2019 - 05:41:45 EST


On Thu, Oct 24, 2019 at 03:51:17PM -0700, samitolvanen@xxxxxxxxxx wrote:
> From: Ard Biesheuvel <ard.biesheuvel@xxxxxxxxxx>
>
> Register x18 will no longer be used as a caller save register in the
> future, so stop using it in the copy_page() code.
>
> Link: https://patchwork.kernel.org/patch/9836869/
> Signed-off-by: Ard Biesheuvel <ard.biesheuvel@xxxxxxxxxx>
> Signed-off-by: Sami Tolvanen <samitolvanen@xxxxxxxxxx>
> ---
> arch/arm64/lib/copy_page.S | 38 +++++++++++++++++++-------------------
> 1 file changed, 19 insertions(+), 19 deletions(-)
>
> diff --git a/arch/arm64/lib/copy_page.S b/arch/arm64/lib/copy_page.S
> index bbb8562396af..8b562264c165 100644
> --- a/arch/arm64/lib/copy_page.S
> +++ b/arch/arm64/lib/copy_page.S
> @@ -34,45 +34,45 @@ alternative_else_nop_endif
> ldp x14, x15, [x1, #96]
> ldp x16, x17, [x1, #112]
>
> - mov x18, #(PAGE_SIZE - 128)
> + add x0, x0, #256
> add x1, x1, #128
> 1:
> - subs x18, x18, #128
> + tst x0, #(PAGE_SIZE - 1)
>
> alternative_if ARM64_HAS_NO_HW_PREFETCH
> prfm pldl1strm, [x1, #384]
> alternative_else_nop_endif
>
> - stnp x2, x3, [x0]
> + stnp x2, x3, [x0, #-256]
> ldp x2, x3, [x1]
> - stnp x4, x5, [x0, #16]
> + stnp x4, x5, [x0, #-240]
> ldp x4, x5, [x1, #16]

For legibility, could we make the offset and bias explicit in the STNPs
so that these line up? e.g.

stnp x4, x5, [x0, #16 - 256]
ldp x4, x5, [x1, #16]

... that'd make it much easier to see by eye that this is sound, much as
I trust my mental arithmetic. ;)

> - stnp x6, x7, [x0, #32]
> + stnp x6, x7, [x0, #-224]
> ldp x6, x7, [x1, #32]
> - stnp x8, x9, [x0, #48]
> + stnp x8, x9, [x0, #-208]
> ldp x8, x9, [x1, #48]
> - stnp x10, x11, [x0, #64]
> + stnp x10, x11, [x0, #-192]
> ldp x10, x11, [x1, #64]
> - stnp x12, x13, [x0, #80]
> + stnp x12, x13, [x0, #-176]
> ldp x12, x13, [x1, #80]
> - stnp x14, x15, [x0, #96]
> + stnp x14, x15, [x0, #-160]
> ldp x14, x15, [x1, #96]
> - stnp x16, x17, [x0, #112]
> + stnp x16, x17, [x0, #-144]
> ldp x16, x17, [x1, #112]
>
> add x0, x0, #128
> add x1, x1, #128
>
> - b.gt 1b
> + b.ne 1b
>
> - stnp x2, x3, [x0]
> - stnp x4, x5, [x0, #16]
> - stnp x6, x7, [x0, #32]
> - stnp x8, x9, [x0, #48]
> - stnp x10, x11, [x0, #64]
> - stnp x12, x13, [x0, #80]
> - stnp x14, x15, [x0, #96]
> - stnp x16, x17, [x0, #112]
> + stnp x2, x3, [x0, #-256]
> + stnp x4, x5, [x0, #-240]
> + stnp x6, x7, [x0, #-224]
> + stnp x8, x9, [x0, #-208]
> + stnp x10, x11, [x0, #-192]
> + stnp x12, x13, [x0, #-176]
> + stnp x14, x15, [x0, #-160]
> + stnp x16, x17, [x0, #-144]

... likewise here:

stnp xt1, xt2, [x0, #off - 256]

I don't see a nicer way to write this sequence without using an
additional register, so with those changes:

Reviewed-by: Mark Rutland <mark.rutland@xxxxxxx>

Thanks,
Mark.

>
> ret
> ENDPROC(copy_page)
> --
> 2.24.0.rc0.303.g954a862665-goog
>