perf trace: Make "--call-graph" affect just "raw_syscalls:sys_exit"

We don't need the callchains at the syscall enter tracepoint, just when
finishing it at syscall exit, so reduce the overhead by asking for
callchains just at syscall exit.
Suggested-by: Milian Wolff <milian.wolff@kdab.com>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Wang Nan <wangnan0@huawei.com>
Link: http://lkml.kernel.org/n/tip-fja1ods5vqpg42mdz09xcz3r@git.kernel.orgSigned-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

tools/perf/builtin-trace.c

@@ -2749,7 +2749,27 @@ static int trace__run(struct trace *trace, int argc, const char **argv)
 		goto out_delete_evlist;
 	}
 
-	perf_evlist__config(evlist, &trace->opts, &callchain_param);
+	perf_evlist__config(evlist, &trace->opts, NULL);
+
+	if (trace->opts.callgraph_set && trace->syscalls.events.sys_exit) {
+		perf_evsel__config_callchain(trace->syscalls.events.sys_exit,
+					     &trace->opts, &callchain_param);
+               /*
+                * Now we have evsels with different sample_ids, use
+                * PERF_SAMPLE_IDENTIFIER to map from sample to evsel
+                * from a fixed position in each ring buffer record.
+                *
+                * As of this the changeset introducing this comment, this
+                * isn't strictly needed, as the fields that can come before
+                * PERF_SAMPLE_ID are all used, but we'll probably disable
+                * some of those for things like copying the payload of
+                * pointer syscall arguments, and for vfs_getname we don't
+                * need PERF_SAMPLE_ADDR and PERF_SAMPLE_IP, so do this
+                * here as a warning we need to use PERF_SAMPLE_IDENTIFIER.
+                */
+		perf_evlist__set_sample_bit(evlist, IDENTIFIER);
+		perf_evlist__reset_sample_bit(evlist, ID);
+	}
 
 	signal(SIGCHLD, sig_handler);
 	signal(SIGINT, sig_handler);