On Mon, Sep 13, 2021 at 10:03 PM Rafael David Tinoco <rafaeldtinoco@xxxxxxxxx> wrote: > > > >> Allow kprobe tracepoint events creation through legacy interface, as the > >> kprobe dynamic PMUs support, used by default, was only created in v4.17. > >> > >> After commit "bpf: implement minimal BPF perf link", it was allowed that > >> some extra - to the link - information is accessed through container_of > >> struct bpf_link. This allows the tracing perf event legacy name, and > >> information whether it is a retprobe, to be saved outside bpf_link > >> structure, which would not be optimal. > >> > >> This enables CO-RE support for older kernels. > >> > >> Cc: Andrii Nakryiko <andrii.nakryiko@xxxxxxxxx> > >> Signed-off-by: Rafael David Tinoco <rafaeldtinoco@xxxxxxxxx> > >> --- > > > > I've adjusted the commit message a bit (this has nothing to do with > > CO-RE per se, so I dropped that, for example). Also see my comments > > below, I've applied all that to your patch while applying, please > > check them out. > > Thanks. I'm assuming you don't need a v6 based on your adjustment comments, let me know if you do please. > Nope, I've applied it to bpf-next. > ... > > >> > >> - if (ioctl(perf_link->perf_event_fd, PERF_EVENT_IOC_DISABLE, 0) < 0) > >> - err = -errno; > >> + ioctl(perf_link->perf_event_fd, PERF_EVENT_IOC_DISABLE, 0); > > > > what's the reason for dropping the error check? I kept it, but please > > let me know if there is any reason to drop it > > From: _perf_ioctl() -> case PERF_EVENT_IOC_DISABLE: func = _perf_event_disable; > > _perf_ioctl() will always return 0 and func is void (*func)(struct perf_event *). > And what about all the future kernels? This is an unnecessary assumption that it will always succeed.
