OSDN Git Service

perf tools: Fix probing for PERF_FLAG_FD_CLOEXEC flag
authorAdrian Hunter <adrian.hunter@intel.com>
Tue, 24 Feb 2015 11:20:59 +0000 (13:20 +0200)
committerArnaldo Carvalho de Melo <acme@redhat.com>
Wed, 25 Feb 2015 19:40:13 +0000 (16:40 -0300)
Commit f6edb53c4993ffe92ce521fb449d1c146cea6ec2 converted the probe to
a CPU wide event first (pid == -1). For kernels that do not support
the PERF_FLAG_FD_CLOEXEC flag the probe fails with EINVAL. Since this
errno is not handled pid is not reset to 0 and the subsequent use of
pid = -1 as an argument brings in an additional failure path if
perf_event_paranoid > 0:

$ perf record -- sleep 1
perf_event_open(..., 0) failed unexpectedly with error 13 (Permission denied)
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 0.007 MB /tmp/perf.data (11 samples) ]

Also, ensure the fd of the confirmation check is closed and comment why
pid = -1 is used.

Needs to go to 3.18 stable tree as well.

Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Based-on-patch-by: David Ahern <david.ahern@oracle.com>
Acked-by: David Ahern <david.ahern@oracle.com>
Cc: David Ahern <dsahern@gmail.com>
Link: http://lkml.kernel.org/r/54EC610C.8000403@intel.com
Cc: stable@vger.kernel.org # v3.18+
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
tools/perf/util/cloexec.c

index 47b78b3..6da965b 100644 (file)
@@ -25,6 +25,10 @@ static int perf_flag_probe(void)
        if (cpu < 0)
                cpu = 0;
 
+       /*
+        * Using -1 for the pid is a workaround to avoid gratuitous jump label
+        * changes.
+        */
        while (1) {
                /* check cloexec flag */
                fd = sys_perf_event_open(&attr, pid, cpu, -1,
@@ -47,16 +51,24 @@ static int perf_flag_probe(void)
                  err, strerror_r(err, sbuf, sizeof(sbuf)));
 
        /* not supported, confirm error related to PERF_FLAG_FD_CLOEXEC */
-       fd = sys_perf_event_open(&attr, pid, cpu, -1, 0);
+       while (1) {
+               fd = sys_perf_event_open(&attr, pid, cpu, -1, 0);
+               if (fd < 0 && pid == -1 && errno == EACCES) {
+                       pid = 0;
+                       continue;
+               }
+               break;
+       }
        err = errno;
 
+       if (fd >= 0)
+               close(fd);
+
        if (WARN_ONCE(fd < 0 && err != EBUSY,
                      "perf_event_open(..., 0) failed unexpectedly with error %d (%s)\n",
                      err, strerror_r(err, sbuf, sizeof(sbuf))))
                return -1;
 
-       close(fd);
-
        return 0;
 }