[PATCH] riscv, bpf: Optimize zextw insn with Zba extension

[Pu Lehui]

On 2024/5/7 18:45, Xiao Wang wrote:
> The Zba extension provides add.uw insn which can be used to implement
> zext.w with rs2 set as ZERO.
> 
> Signed-off-by: Xiao Wang <xiao.w.wang at intel.com>
> ---
>   arch/riscv/Kconfig       | 19 +++++++++++++++++++
>   arch/riscv/net/bpf_jit.h | 18 ++++++++++++++++++
>   2 files changed, 37 insertions(+)
> 
> diff --git a/arch/riscv/Kconfig b/arch/riscv/Kconfig
> index 6bec1bce6586..0679127cc0ea 100644
> --- a/arch/riscv/Kconfig
> +++ b/arch/riscv/Kconfig
> @@ -586,6 +586,14 @@ config RISCV_ISA_V_PREEMPTIVE
>   	  preemption. Enabling this config will result in higher memory
>   	  consumption due to the allocation of per-task's kernel Vector context.
>   
> +config TOOLCHAIN_HAS_ZBA
> +	bool
> +	default y
> +	depends on !64BIT || $(cc-option,-mabi=lp64 -march=rv64ima_zba)
> +	depends on !32BIT || $(cc-option,-mabi=ilp32 -march=rv32ima_zba)
> +	depends on LLD_VERSION >= 150000 || LD_VERSION >= 23900
> +	depends on AS_HAS_OPTION_ARCH
> +
>   config TOOLCHAIN_HAS_ZBB
>   	bool
>   	default y
> @@ -601,6 +609,17 @@ config TOOLCHAIN_HAS_VECTOR_CRYPTO
>   	def_bool $(as-instr, .option arch$(comma) +v$(comma) +zvkb)
>   	depends on AS_HAS_OPTION_ARCH
>   
> +config RISCV_ISA_ZBA
> +	bool "Zba extension support for bit manipulation instructions"
> +	depends on TOOLCHAIN_HAS_ZBA
> +	depends on RISCV_ALTERNATIVE
> +	default y
> +	help
> +	   Adds support to dynamically detect the presence of the ZBA
> +	   extension (address generation acceleration) and enable its usage.

It would be better to add Zba's function description like Zbb.

> +
> +	   If you don't know what to do here, say Y.
> +
>   config RISCV_ISA_ZBB
>   	bool "Zbb extension support for bit manipulation instructions"
>   	depends on TOOLCHAIN_HAS_ZBB
> diff --git a/arch/riscv/net/bpf_jit.h b/arch/riscv/net/bpf_jit.h
> index f4b6b3b9edda..18a7885ba95e 100644
> --- a/arch/riscv/net/bpf_jit.h
> +++ b/arch/riscv/net/bpf_jit.h
> @@ -18,6 +18,11 @@ static inline bool rvc_enabled(void)
>   	return IS_ENABLED(CONFIG_RISCV_ISA_C);
>   }
>   
> +static inline bool rvzba_enabled(void)
> +{
> +	return IS_ENABLED(CONFIG_RISCV_ISA_ZBA) && riscv_has_extension_likely(RISCV_ISA_EXT_ZBA);
> +}
> +
>   static inline bool rvzbb_enabled(void)
>   {
>   	return IS_ENABLED(CONFIG_RISCV_ISA_ZBB) && riscv_has_extension_likely(RISCV_ISA_EXT_ZBB);
> @@ -937,6 +942,14 @@ static inline u16 rvc_sdsp(u32 imm9, u8 rs2)
>   	return rv_css_insn(0x7, imm, rs2, 0x2);
>   }
>   
> +/* RV64-only ZBA instructions. */
> +
> +static inline u32 rvzba_zextw(u8 rd, u8 rs1)
> +{
> +	/* add.uw rd, rs1, ZERO */
> +	return rv_r_insn(0x04, RV_REG_ZERO, rs1, 0, rd, 0x3b);
> +}
> +
>   #endif /* __riscv_xlen == 64 */
>   
>   /* Helper functions that emit RVC instructions when possible. */
> @@ -1159,6 +1172,11 @@ static inline void emit_zexth(u8 rd, u8 rs, struct rv_jit_context *ctx)
>   
>   static inline void emit_zextw(u8 rd, u8 rs, struct rv_jit_context *ctx)
>   {
> +	if (rvzba_enabled()) {
> +		emit(rvzba_zextw(rd, rs), ctx);
> +		return;
> +	}

Looks good to me. It seems that Zba has fewer uses in rv64 bpf jit.

> +
>   	emit_slli(rd, rs, 32, ctx);
>   	emit_srli(rd, rd, 32, ctx);
>   }