On Tue, Sep 1, 2020 at 10:45 PM Alexei Starovoitov
<alexei.starovoitov@xxxxxxxxx> wrote:
>
> On Mon, Aug 31, 2020 at 06:50:01PM -0700, Andrii Nakryiko wrote:
> > diff --git a/tools/testing/selftests/bpf/progs/pyperf.h b/tools/testing/selftests/bpf/progs/pyperf.h
> > index cc615b82b56e..13998aee887f 100644
> > --- a/tools/testing/selftests/bpf/progs/pyperf.h
> > +++ b/tools/testing/selftests/bpf/progs/pyperf.h
> > @@ -67,7 +67,7 @@ typedef struct {
> > void* co_name; // PyCodeObject.co_name
> > } FrameData;
> >
> > -static __always_inline void *get_thread_state(void *tls_base, PidData *pidData)
> > +static __noinline void *get_thread_state(void *tls_base, PidData *pidData)
> > {
> > void* thread_state;
> > int key;
> > @@ -154,12 +154,10 @@ struct {
> > __uint(value_size, sizeof(long long) * 127);
> > } stackmap SEC(".maps");
> >
> > -#ifdef GLOBAL_FUNC
> > -__attribute__((noinline))
> > -#else
> > -static __always_inline
> > +#ifndef GLOBAL_FUNC
> > +static
> > #endif
> > -int __on_event(struct bpf_raw_tracepoint_args *ctx)
> > +__noinline int __on_event(struct bpf_raw_tracepoint_args *ctx)
> > {
> > uint64_t pid_tgid = bpf_get_current_pid_tgid();
> > pid_t pid = (pid_t)(pid_tgid >> 32);
> > diff --git a/tools/testing/selftests/bpf/progs/strobemeta.h b/tools/testing/selftests/bpf/progs/strobemeta.h
> > index ad61b722a9de..d307c67ce52e 100644
> > --- a/tools/testing/selftests/bpf/progs/strobemeta.h
> > +++ b/tools/testing/selftests/bpf/progs/strobemeta.h
> > @@ -266,8 +266,7 @@ struct tls_index {
> > uint64_t offset;
> > };
> >
> > -static __always_inline void *calc_location(struct strobe_value_loc *loc,
> > - void *tls_base)
> > +static __noinline void *calc_location(struct strobe_value_loc *loc, void *tls_base)
>
> hmm. this reduces the existing test coverage. Unless I'm misreading it.
> Could you keep existing strobemta tests and add new one?
> With new ifdefs. Like this GLOBAL_FUNC.
Oh, you mean testing single BPF program complexity when everything is
inlined? Yeah, haven't thought about that. Ok, I'll add new variants
with or without subprogram calls.
