Re: [RFC PATCH 2/5] bpf: Pass trace_probe to bpf_prog for variable fetching

[Alexei Starovoitov <ast@plumgrid.com>]

On 5/17/15 10:30 PM, He Kuang wrote:
> Add new structure bpf_pt_regs, which contains both original
> 'ctx'(pt_regs) and trabe_probe pointer, and pass this new pointer to bpf
> prog for variable fetching.
>
> Signed-off-by: He Kuang <hekuang@huawei.com>
> ---
>   kernel/trace/trace_kprobe.c | 11 +++++++++--
>   kernel/trace/trace_probe.h  |  5 +++++
>   2 files changed, 14 insertions(+), 2 deletions(-)
>
> diff --git a/kernel/trace/trace_kprobe.c b/kernel/trace/trace_kprobe.c
> index d0ce590..cee0b28 100644
> --- a/kernel/trace/trace_kprobe.c
> +++ b/kernel/trace/trace_kprobe.c
> @@ -1141,8 +1141,15 @@ kprobe_perf_func(struct trace_kprobe *tk, struct pt_regs *regs)
>   	int size, __size, dsize;
>   	int rctx;
>
> -	if (prog && !trace_call_bpf(prog, regs))
> -		return;
> +	if (prog) {
> +		struct bpf_pt_regs bpf_pt_regs;
> +
> +		bpf_pt_regs.pt_regs = *regs;
> +		bpf_pt_regs.tp = &tk->tp;
...
> +struct bpf_pt_regs {
> +	struct pt_regs pt_regs;
> +	struct trace_probe *tp;
> +};

that is a massive overhead.
On x64 it means copying 168 bytes for every call.
imo that's a wrong trade off.