perf unwind: Check the target platform before assigning unwind methods

Currently, 'perf script' uses host unwind methods to parse perf.data
callchain info without taking the target architecture into account, i.e.
assuming the perf.data file was generated on the same machine where the
analysis is being performed. So we get wrong result without any warnings
when unwinding callchains of x86(32-bit) on x86(64-bit) machine.

This patch adds an extra step that checks the target platform before
assigning unwind methods. In later patches in this series, we can use
this info to assign the right unwind methods for supported platforms.

Committer note:

After fixing it to register the local unwinder for live mode tools
('perf trace', 'perf top'), i.e. tools that don't use a perf.data file,
it works as intended and passes the 'perf test unwind' test:

  # perf trace -e nanosleep --call dwarf usleep 1
     0.328 ( 0.058 ms): usleep/11115 nanosleep(rqtp: 0x7fff083fa480) = 0
                                       __nanosleep_nocancel+0x7 (/usr/lib64/libc-2.22.so)
                                       usleep+0x34 (/usr/lib64/libc-2.22.so)
                                       main+0x1eb (/usr/bin/usleep)
                                       __libc_start_main+0xf0 (/usr/lib64/libc-2.22.so)
                                       _start+0x29 (/usr/bin/usleep)
  # perf test 48
  48: Test dwarf unwind         : Ok
  #

Signed-off-by: He Kuang <hekuang@huawei.com>
Acked-by: Jiri Olsa <jolsa@kernel.org>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Ekaterina Tumanova <tumanova@linux.vnet.ibm.com>
Cc: Josh Poimboeuf <jpoimboe@redhat.com>
Cc: Kan Liang <kan.liang@intel.com>
Cc: Masami Hiramatsu <mhiramat@kernel.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Pekka Enberg <penberg@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Sukadev Bhattiprolu <sukadev@linux.vnet.ibm.com>
Cc: Wang Nan <wangnan0@huawei.com>
Link: http://lkml.kernel.org/r/1464924803-22214-11-git-send-email-hekuang@huawei.com
[ Fixed exit path for 'live' mode tools, where we need to default to local unwinding ]
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

1 files changed, 24 insertions, 1 deletions

diff --git a/tools/perf/util/unwind-libunwind.c b/tools/perf/util/unwind-libunwind.c
index f86f903ae8a8..0086726e00e0 100644
--- a/tools/perf/util/unwind-libunwind.c
+++ b/tools/perf/util/unwind-libunwind.c
@@ -1,5 +1,8 @@
 #include "unwind.h"
 #include "thread.h"
+#include "session.h"
+#include "debug.h"
+#include "arch/common.h"
 
 struct unwind_libunwind_ops __weak *local_unwind_libunwind_ops;
 
@@ -9,8 +12,28 @@ static void unwind__register_ops(struct thread *thread,
 	thread->unwind_libunwind_ops = ops;
 }
 
-int unwind__prepare_access(struct thread *thread)
+int unwind__prepare_access(struct thread *thread, struct map *map)
 {
+	const char *arch;
+	enum dso_type dso_type;
+
+	if (thread->addr_space) {
+		pr_debug("unwind: thread map already set, dso=%s\n",
+			 map->dso->name);
+		return 0;
+	}
+
+	/* env->arch is NULL for live-mode (i.e. perf top) */
+	if (!thread->mg->machine->env || !thread->mg->machine->env->arch)
+		goto out_register;
+
+	dso_type = dso__type(map->dso, thread->mg->machine);
+	if (dso_type == DSO__TYPE_UNKNOWN)
+		return 0;
+
+	arch = normalize_arch(thread->mg->machine->env->arch);
+	pr_debug("unwind: target platform=%s\n", arch);
+out_register:
 	unwind__register_ops(thread, local_unwind_libunwind_ops);
 
 	return thread->unwind_libunwind_ops->prepare_access(thread);