[PATCH] libperf: Remove BUG_ON() from library code in get_group_fd(). was Re: [PATCH 6/7] libperF: Add group support to perf_evsel__open

[Arnaldo Carvalho de Melo <acme@kernel.org>]

Em Wed, Jul 07, 2021 at 08:15:38PM +0200, Jiri Olsa escreveu:
> On Wed, Jul 07, 2021 at 02:43:07PM -0300, Arnaldo Carvalho de Melo wrote:
> > Em Tue, Jul 06, 2021 at 05:17:03PM +0200, Jiri Olsa escreveu:
> > > +static int get_group_fd(struct perf_evsel *evsel, int cpu, int thread)
> > > +{
> > > +	struct perf_evsel *leader = evsel->leader;
> > > +	int fd;
> > > +
> > > +	if (evsel == leader)
> > > +		return -1;
> > > +
> > > +	/*
> > > +	 * Leader must be already processed/open,
> > > +	 * if not it's a bug.
> > > +	 */
> > > +	BUG_ON(!leader->fd);

> > Humm, having panics in library code looks ugly, why can't we just return
> > some errno and let the whatever is using the library to fail gracefully?
 
> true, I took it from perf code, did not realize this,
> I'll check what can we do in here

So, I've added this as a follow up patch:

commit 0ec138125eaea5f15157adcecc3e0def1ad2ed22
Author: Arnaldo Carvalho de Melo <acme@redhat.com>
Date:   Fri Jul 9 14:52:16 2021 -0300

    libperf: Remove BUG_ON() from library code in get_group_fd()
    
    We shouldn't just panic, return a value that doesn't clash with what
    perf_evsel__open() was already returning in case of error, i.e. errno
    when sys_perf_event_open() fails.
    
    Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
    Cc: Ian Rogers <irogers@google.com>
    Cc: Jiri Olsa <jolsa@kernel.org>
    Cc: Mark Rutland <mark.rutland@arm.com>
    Cc: Michael Petlan <mpetlan@redhat.com>
    Cc: Namhyung Kim <namhyung@kernel.org>
    Cc: Peter Zijlstra <peterz@infradead.org>
    Cc: Shunsuke Nakamura <nakamura.shun@fujitsu.com>
    Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

diff --git a/tools/lib/perf/evsel.c b/tools/lib/perf/evsel.c
index 9ebf7122d4766c5e..d8886720e83d8dfe 100644
--- a/tools/lib/perf/evsel.c
+++ b/tools/lib/perf/evsel.c
@@ -77,23 +77,30 @@ sys_perf_event_open(struct perf_event_attr *attr,
 	return syscall(__NR_perf_event_open, attr, pid, cpu, group_fd, flags);
 }
 
-static int get_group_fd(struct perf_evsel *evsel, int cpu, int thread)
+static int get_group_fd(struct perf_evsel *evsel, int cpu, int thread, int *group_fd)
 {
 	struct perf_evsel *leader = evsel->leader;
 	int fd;
 
-	if (evsel == leader)
-		return -1;
+	if (evsel == leader) {
+		*group_fd = -1;
+		return 0;
+	}
 
 	/*
 	 * Leader must be already processed/open,
 	 * if not it's a bug.
 	 */
-	BUG_ON(!leader->fd);
+	if (!leader->fd)
+		return -ENOTCONN;
 
 	fd = FD(leader, cpu, thread);
-	BUG_ON(fd == -1);
-	return fd;
+	if (fd == -1)
+		return -EBADF;
+
+	*group_fd = fd;
+
+	return 0;
 }
 
 int perf_evsel__open(struct perf_evsel *evsel, struct perf_cpu_map *cpus,
@@ -133,7 +140,9 @@ int perf_evsel__open(struct perf_evsel *evsel, struct perf_cpu_map *cpus,
 		for (thread = 0; thread < threads->nr; thread++) {
 			int fd, group_fd;
 
-			group_fd = get_group_fd(evsel, cpu, thread);
+			err = get_group_fd(evsel, cpu, thread, &group_fd);
+			if (err < 0)
+				return err;
 
 			fd = sys_perf_event_open(&evsel->attr,
 						 threads->map[thread].pid,