Re: [PATCH perf/core 3/5] perf tools: Move libbpf init in libbpf_init function

[Jiri Olsa <olsajiri@gmail.com>]

On Fri, Apr 22, 2022 at 02:03:24PM -0300, Arnaldo Carvalho de Melo wrote:
> Em Fri, Apr 22, 2022 at 12:00:23PM +0200, Jiri Olsa escreveu:
> > Moving the libbpf init code into single function,
> > so we have single place doing that.
> 
> Cherry picked this one, waiting for Andrii to chime in wrt the libbpf
> changes, if its acceptable, how to proceed, i.e. in what tree to carry
> these?

I think at this point it's ok with either yours perf/core or bpf-next/master,
I waited for them to be in sync for this ;-)

but as you pointed out there's issue with perf linked with dynamic libbpf,
because the current version does not have the libbpf_register_prog_handler
api that perf needs now.. also it needs the fix and api introduced in this
patchset

I'll check and perhaps we can temporirly disable perf/bpf prologue generation
for dynamic linking..? until the libbpf release has all the needed changes

jirka

> 
> - Arnaldo
>  
> > Signed-off-by: Jiri Olsa <jolsa@kernel.org>
> > ---
> >  tools/perf/util/bpf-loader.c | 27 ++++++++++++++++++---------
> >  1 file changed, 18 insertions(+), 9 deletions(-)
> > 
> > diff --git a/tools/perf/util/bpf-loader.c b/tools/perf/util/bpf-loader.c
> > index b72cef1ae959..f8ad581ea247 100644
> > --- a/tools/perf/util/bpf-loader.c
> > +++ b/tools/perf/util/bpf-loader.c
> > @@ -99,16 +99,26 @@ static int bpf_perf_object__add(struct bpf_object *obj)
> >  	return perf_obj ? 0 : -ENOMEM;
> >  }
> >  
> > +static int libbpf_init(void)
> > +{
> > +	if (libbpf_initialized)
> > +		return 0;
> > +
> > +	libbpf_set_print(libbpf_perf_print);
> > +	libbpf_initialized = true;
> > +	return 0;
> > +}
> > +
> >  struct bpf_object *
> >  bpf__prepare_load_buffer(void *obj_buf, size_t obj_buf_sz, const char *name)
> >  {
> >  	LIBBPF_OPTS(bpf_object_open_opts, opts, .object_name = name);
> >  	struct bpf_object *obj;
> > +	int err;
> >  
> > -	if (!libbpf_initialized) {
> > -		libbpf_set_print(libbpf_perf_print);
> > -		libbpf_initialized = true;
> > -	}
> > +	err = libbpf_init();
> > +	if (err)
> > +		return ERR_PTR(err);
> >  
> >  	obj = bpf_object__open_mem(obj_buf, obj_buf_sz, &opts);
> >  	if (IS_ERR_OR_NULL(obj)) {
> > @@ -135,14 +145,13 @@ struct bpf_object *bpf__prepare_load(const char *filename, bool source)
> >  {
> >  	LIBBPF_OPTS(bpf_object_open_opts, opts, .object_name = filename);
> >  	struct bpf_object *obj;
> > +	int err;
> >  
> > -	if (!libbpf_initialized) {
> > -		libbpf_set_print(libbpf_perf_print);
> > -		libbpf_initialized = true;
> > -	}
> > +	err = libbpf_init();
> > +	if (err)
> > +		return ERR_PTR(err);
> >  
> >  	if (source) {
> > -		int err;
> >  		void *obj_buf;
> >  		size_t obj_buf_sz;
> >  
> > -- 
> > 2.35.1
> 
> -- 
> 
> - Arnaldo