[PATCH v2 4/6] RISC-V: Use Zicboz in clear_page when available
Conor Dooley
conor.dooley at microchip.com
Mon Jan 23 03:42:37 PST 2023
On Sun, Jan 22, 2023 at 08:13:26PM +0100, Andrew Jones wrote:
> Using memset() to zero a 4K page takes 563 total instructions
> where 20 are branches. clear_page() with Zicboz takes 150 total
> instructions where 16 are branches. We could reduce the numbers
> by further unrolling, but, since the cboz block size isn't fixed,
> we'd need a Duff device to ensure we don't execute too many
> unrolled steps. Also, cbo.zero doesn't take an offset, so each
> unrolled step requires it and an add instruction. This increases
> the chance for icache misses if we unroll many times. For these
> reasons we only unroll four times. Unrolling four times should be
> safe as it supports cboz block sizes up to 1K when used with 4K
> pages and it's only 24 to 32 bytes of unrolled instructions.
>
> Another note about the Duff device idea is that it would probably
> be best to store the number of steps needed at boot time and then
> load the value in clear_page(). Calculating it in clear_page(),
> particularly without the Zbb extension, would not be efficient.
>
> Signed-off-by: Andrew Jones <ajones at ventanamicro.com>
> ---
> arch/riscv/Kconfig | 13 +++++++++++
> arch/riscv/include/asm/insn-def.h | 4 ++++
> arch/riscv/include/asm/page.h | 6 +++++-
> arch/riscv/lib/Makefile | 1 +
> arch/riscv/lib/clear_page.S | 36 +++++++++++++++++++++++++++++++
> 5 files changed, 59 insertions(+), 1 deletion(-)
> create mode 100644 arch/riscv/lib/clear_page.S
>
> diff --git a/arch/riscv/Kconfig b/arch/riscv/Kconfig
> index 33bbdc33cef8..3759a2f6edd5 100644
> --- a/arch/riscv/Kconfig
> +++ b/arch/riscv/Kconfig
> @@ -432,6 +432,19 @@ config RISCV_ISA_ZICBOM
>
> If you don't know what to do here, say Y.
>
> +config RISCV_ISA_ZICBOZ
> + bool "Zicboz extension support for faster zeroing of memory"
> + depends on !XIP_KERNEL && MMU
> + select RISCV_ALTERNATIVE
> + default y
> + help
> + Enable the use of the ZICBOZ extension (cbo.zero instruction)
> + when available.
> +
> + The Zicboz extension is used for faster zeroing of memory.
> +
> + If you don't know what to do here, say Y.
> +
> config TOOLCHAIN_HAS_ZIHINTPAUSE
> bool
> default y
> diff --git a/arch/riscv/include/asm/insn-def.h b/arch/riscv/include/asm/insn-def.h
> index e01ab51f50d2..6960beb75f32 100644
> --- a/arch/riscv/include/asm/insn-def.h
> +++ b/arch/riscv/include/asm/insn-def.h
> @@ -192,4 +192,8 @@
> INSN_I(OPCODE_MISC_MEM, FUNC3(2), __RD(0), \
> RS1(base), SIMM12(2))
>
> +#define CBO_zero(base) \
> + INSN_I(OPCODE_MISC_MEM, FUNC3(2), __RD(0), \
> + RS1(base), SIMM12(4))
> +
Since most of the patch is the clear page implementation that I have no
comments on:
Acked-by: Conor Dooley <conor.dooley at microchip.com>
Thanks,
Conor.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 228 bytes
Desc: not available
URL: <http://lists.infradead.org/pipermail/linux-riscv/attachments/20230123/3f8f6643/attachment.sig>
More information about the linux-riscv
mailing list