Re: [PATCH bpf-next] bpf, x86: inline bpf_get_current_task() for x86_64

[Eduard Zingerman <eddyz87@gmail.com>]

On Thu, 2025-12-25 at 18:44 +0800, Menglong Dong wrote:
> Inline bpf_get_current_task() and bpf_get_current_task_btf() for x86_64
> to obtain better performance. The instruction we use here is:
> 
>   65 48 8B 04 25 [offset] // mov rax, gs:[offset]
> 
> Not sure if there is any side effect here.
> 
> Signed-off-by: Menglong Dong <dongml2@chinatelecom.cn>
> ---

The change makes sense to me.
Could you please address the compilation error reported by kernel test robot?
Could you please also add a tests case using __jited annotation like
in verifier_ldsx.c?

>  arch/x86/net/bpf_jit_comp.c | 33 +++++++++++++++++++++++++++++++++
>  1 file changed, 33 insertions(+)
> 
> diff --git a/arch/x86/net/bpf_jit_comp.c b/arch/x86/net/bpf_jit_comp.c
> index b69dc7194e2c..7f38481816f0 100644
> --- a/arch/x86/net/bpf_jit_comp.c
> +++ b/arch/x86/net/bpf_jit_comp.c
> @@ -1300,6 +1300,19 @@ static void emit_st_r12(u8 **pprog, u32 size, u32 dst_reg, int off, int imm)
>  	emit_st_index(pprog, size, dst_reg, X86_REG_R12, off, imm);
>  }
>  
> +static void emit_ldx_percpu_r0(u8 **pprog, const void __percpu *ptr)
> +{
> +	u8 *prog = *pprog;
> +
> +	/* mov rax, gs:[offset] */
> +	EMIT2(0x65, 0x48);
> +	EMIT2(0x8B, 0x04);
> +	EMIT1(0x25);
> +	EMIT((u32)(unsigned long)ptr, 4);
> +
> +	*pprog = prog;
> +}
> +
>  static int emit_atomic_rmw(u8 **pprog, u32 atomic_op,
>  			   u32 dst_reg, u32 src_reg, s16 off, u8 bpf_size)
>  {
> @@ -2435,6 +2448,15 @@ st:			if (is_imm8(insn->off))
>  		case BPF_JMP | BPF_CALL: {
>  			u8 *ip = image + addrs[i - 1];
>  
> +			if (insn->src_reg == 0 && (insn->imm == BPF_FUNC_get_current_task ||
> +						   insn->imm == BPF_FUNC_get_current_task_btf)) {
> +				if (IS_ENABLED(CONFIG_USE_X86_SEG_SUPPORT))
> +					emit_ldx_percpu_r0(&prog, &const_current_task);
> +				else
> +					emit_ldx_percpu_r0(&prog, &current_task);

Nit: arch/x86/include/asm/current.h says that current_task and const_current_task are aliases.
     In that case, why would we need two branches here?

> +				break;
> +			}
> +
>  			func = (u8 *) __bpf_call_base + imm32;
>  			if (src_reg == BPF_PSEUDO_CALL && tail_call_reachable) {
>  				LOAD_TAIL_CALL_CNT_PTR(stack_depth);
> @@ -4067,3 +4089,14 @@ bool bpf_jit_supports_timed_may_goto(void)
>  {
>  	return true;
>  }
> +
> +bool bpf_jit_inlines_helper_call(s32 imm)
> +{
> +	switch (imm) {
> +	case BPF_FUNC_get_current_task:
> +	case BPF_FUNC_get_current_task_btf:
> +		return true;
> +	default:
> +		return false;
> +	}
> +}