perf mmap: Be consistent when checking for an unmaped ring buffer
authorArnaldo Carvalho de Melo <acme@redhat.com>
Mon, 26 Mar 2018 14:42:15 +0000 (11:42 -0300)
committerArnaldo Carvalho de Melo <acme@redhat.com>
Tue, 27 Mar 2018 16:13:38 +0000 (13:13 -0300)
The previous patch is insufficient to cure the reported 'perf trace'
segfault, as it only cures the perf_mmap__read_done() case, moving the
segfault to perf_mmap__read_init() functio, fix it by doing the same
refcount check.

Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: David Ahern <dsahern@gmail.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Kan Liang <kan.liang@intel.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Wang Nan <wangnan0@huawei.com>
Fixes: 8872481bd048 ("perf mmap: Introduce perf_mmap__read_init()")
Link: https://lkml.kernel.org/r/20180326144127.GF18897@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
tools/perf/util/mmap.c

index f6cfc52..fc83267 100644 (file)
@@ -234,7 +234,7 @@ static int overwrite_rb_find_range(void *buf, int mask, u64 *start, u64 *end)
 /*
  * Report the start and end of the available data in ringbuffer
  */
-int perf_mmap__read_init(struct perf_mmap *md)
+static int __perf_mmap__read_init(struct perf_mmap *md)
 {
        u64 head = perf_mmap__read_head(md);
        u64 old = md->prev;
@@ -268,6 +268,17 @@ int perf_mmap__read_init(struct perf_mmap *md)
        return 0;
 }
 
+int perf_mmap__read_init(struct perf_mmap *map)
+{
+       /*
+        * Check if event was unmapped due to a POLLHUP/POLLERR.
+        */
+       if (!refcount_read(&map->refcnt))
+               return -ENOENT;
+
+       return __perf_mmap__read_init(map);
+}
+
 int perf_mmap__push(struct perf_mmap *md, void *to,
                    int push(void *to, void *buf, size_t size))
 {