On Tue, Mar 22, 2022 at 8:30 AM Kui-Feng Lee <kuifeng@xxxxxx> wrote:
>
> On Mon, 2022-03-21 at 16:08 -0700, Andrii Nakryiko wrote:
> > On Tue, Mar 15, 2022 at 5:44 PM Kui-Feng Lee <kuifeng@xxxxxx> wrote:
> > >
> > > @@ -1291,6 +1294,7 @@ struct bpf_cg_run_ctx {
> > > struct bpf_trace_run_ctx {
> > > struct bpf_run_ctx run_ctx;
> > > u64 bpf_cookie;
> > > + struct bpf_run_ctx *saved_run_ctx;
> > > };
> >
> > oh, and bpf_trace_run_ctx is used for kprobe/uprobe/tracepoint, let's
> > add a new struct bpf_tramp_run_ctx which would reflect that it is
> > used
> > for BPF trampoline-based BPF programs. Otherwise it's confusing to
> > have saved_run_ctx for kprobe where we don't use that. Similarly, if
> > we move "start" timestamp, it will be a bit off. Not end of the
> > world,
> > but I think keeping them separate would make sense over long run.
>
> Ok!
>
We discussed this start timestamp with Alexei offline and concluded
that it's better to leave start as is, as we don't really store start
on the stack (we keep it in register), so moving it into run_ctx won't
give us any benefits.
