fix scheduling policy service death detection

Check status of transactions to scheduling policy service
and re-acquire a binder interface in case of DEAD_OBJECT.

Bug: 8875559.
Change-Id: I1e00bd44e2d4723b3ec95d5c31d9652ba08e238a

1 files changed, 12 insertions, 4 deletions

diff --git a/services/audioflinger/ISchedulingPolicyService.cpp b/services/audioflinger/ISchedulingPolicyService.cpp
index 0079968..f55bc02 100644
--- a/services/audioflinger/ISchedulingPolicyService.cpp
+++ b/services/audioflinger/ISchedulingPolicyService.cpp
@@ -14,7 +14,7 @@
  * limitations under the License.
  */
 
-#define LOG_TAG "SchedulingPolicyService"
+#define LOG_TAG "ISchedulingPolicyService"
 //#define LOG_NDEBUG 0
 
 #include <binder/Parcel.h>
@@ -45,9 +45,17 @@ public:
         data.writeInt32(tid);
         data.writeInt32(prio);
         uint32_t flags = asynchronous ? IBinder::FLAG_ONEWAY : 0;
-        remote()->transact(REQUEST_PRIORITY_TRANSACTION, data, &reply, flags);
-        // fail on exception
-        if (reply.readExceptionCode() != 0) return -1;
+        status_t status = remote()->transact(REQUEST_PRIORITY_TRANSACTION, data, &reply, flags);
+        if (status != NO_ERROR) {
+            return status;
+        }
+        if (asynchronous) {
+            return NO_ERROR;
+        }
+        // fail on exception: force binder reconnection
+        if (reply.readExceptionCode() != 0) {
+            return DEAD_OBJECT;
+        }
         return reply.readInt32();
     }
 };