Re: [PATCH bpf-next v3 3/5] bpf: Inline calls to bpf_loop when callback is known

[Eduard Zingerman <eddyz87@gmail.com>]

Hi Joanne,

> Looking at arch/arm64/net/bpf_jit_comp.c, I see this diagram
> 
>         /*
>          * BPF prog stack layout
>          *
>          *                         high
>          * original A64_SP =>   0:+-----+ BPF prologue
>          *                        |FP/LR|
>          * current A64_FP =>  -16:+-----+
>          *                        | ... | callee saved registers
>          * BPF fp register => -64:+-----+ <= (BPF_FP)
>          *                        |     |
>          *                        | ... | BPF prog stack
>          *                        |     |
>          *                        +-----+ <= (BPF_FP - prog->aux->stack_depth)
>          *                        |RSVD | padding
>          * current A64_SP =>      +-----+ <= (BPF_FP - ctx->stack_size)
>          *                        |     |
>          *                        | ... | Function call stack
>          *                        |     |
>          *                        +-----+
>          *                          low
>          *
>          */
> It looks like prog->aux->stack_depth is used for the "BPF prog stack",
> which is the stack for the main bpf program (subprog 0)

Thanks for looking into this. Also note the verifier.c:jit_subprogs
function which essentially "promotes" every sub-program to sub-program #0
before calling bpf_jit_comp.c:bpf_int_jit_compile for that sub-program.

> I'm not sure either whether MAX_BPF_STACK is a hard limit or a soft
> limit. I'm curious to know as well.

Alexei had commented in a sibling thread that MAX_BPF_STACK should be
considered a soft limit.

Thanks,
Eduard