Re: [RFT PATCH] riscv: mremap speedup - enable HAVE_MOVE_PUD and HAVE_MOVE_PMD

[Palmer Dabbelt]

On Fri, 16 Apr 2021 09:37:22 PDT (-0700), jszhang3@xxxxxxxxxxxxxxxx wrote:
> From: Jisheng Zhang <jszhang@xxxxxxxxxx>
>
> HAVE_MOVE_PUD enables remapping pages at the PUD level if both the source
> and destination addresses are PUD-aligned.
> HAVE_MOVE_PMD does similar speedup on the PMD level.
>
> With HAVE_MOVE_PUD enabled, there is about a 143x improvement on qemu
> With HAVE_MOVE_PMD enabled, there is about a 5x improvement on qemu
>
> Signed-off-by: Jisheng Zhang <jszhang@xxxxxxxxxx>
> ---
> Hi all,
>
> After reading the risc-v privileged spec, I think it's safe to support
> the HAVE_MOVE_PUD and HAVE_MOVE_PMD optimization on riscv. The patch passed
> the mremap_test on QEMU. However this can't be tested on real HW due to lack
> of HW boards, this is the reason why I mark this patch as RFT. I hope I can
> buy a sipeed Allwiner D1 SoC powered SBC soon, at least I can test
> HAVE_MOVE_PMD. Helping me to test on real HW is appreciated!
>
> Thanks
>
>  arch/riscv/Kconfig               |  2 ++
>  arch/riscv/include/asm/pgtable.h | 12 ++++++++++++
>  2 files changed, 14 insertions(+)
>
> diff --git a/arch/riscv/Kconfig b/arch/riscv/Kconfig
> index e8074d248457..37660de68b00 100644
> --- a/arch/riscv/Kconfig
> +++ b/arch/riscv/Kconfig
> @@ -80,6 +80,8 @@ config RISCV
>  	select HAVE_KPROBES
>  	select HAVE_KPROBES_ON_FTRACE
>  	select HAVE_KRETPROBES
> +	select HAVE_MOVE_PMD
> +	select HAVE_MOVE_PUD
>  	select HAVE_PCI
>  	select HAVE_PERF_EVENTS
>  	select HAVE_PERF_REGS
> diff --git a/arch/riscv/include/asm/pgtable.h b/arch/riscv/include/asm/pgtable.h
> index ebf817c1bdf4..287733b95d10 100644
> --- a/arch/riscv/include/asm/pgtable.h
> +++ b/arch/riscv/include/asm/pgtable.h
> @@ -360,6 +360,18 @@ static inline void set_pte_at(struct mm_struct *mm,
>  	set_pte(ptep, pteval);
>  }
>
> +static inline void set_pmd_at(struct mm_struct *mm, unsigned long addr,
> +			      pmd_t *pmdp, pmd_t pmd)
> +{
> +	set_pmd(pmdp, pmd);
> +}
> +
> +static inline void set_pud_at(struct mm_struct *mm, unsigned long addr,
> +			      pud_t *pudp, pud_t pud)
> +{
> +	set_pud(pudp, pud);
> +}
> +
>  static inline void pte_clear(struct mm_struct *mm,
>  	unsigned long addr, pte_t *ptep)
>  {

Thanks, this is on for-next.