Bug 101656

Description: I'm using EGL_SYNC_NATIVE_FENCE_ANDROID to create a fence fd in producer consumer situation. I'm doing this for every buffer, I see that I'm getting a valid fd every time, I pass that to consumer and on consumer side I call getSingalTime() on this. This uses SYNC_IOC_FILE_INFO ioctl though libsync to get time 64 bit timestamp of when fence was signaled. The observation is sometimes I'm getting lower timestamp for newer fences! 

e.g. fd1 is with Buffer1 later created fd2 is with buffer2, getSignalTime of fd2 is lower than fd1, glFulsh is called before both fds were obtained (force signal).

Expectation - timestamps of newer fds should be higher than older fds always.

Some Code

Producer side

EGLDisplay dpy = eglGetCurrentDisplay(); 
EGLSyncKHR sync = eglCreateSyncKHR(dpy,
                    EGL_SYNC_NATIVE_FENCE_ANDROID, NULL);
            if (sync == EGL_NO_SYNC_KHR) {
                return UNKNOWN_ERROR;
            }
    glFlush();
    fenceFd = eglDupNativeFenceFDANDROID(dpy, sync);

   eglDestroySyncKHR(dpy, sync);
   if (fenceFd == EGL_NO_NATIVE_FENCE_FD_ANDROID) {
                return UNKNOWN_ERROR;
            }

Consumer side

if (item->mFence->isValid()) {
    status_t err = item->mFence->waitForever("acquireBufferLocked");
    if (err != NO_ERROR) {
            return err;
    }
 }

Is EGL_SYNC_NATIVE_FENCE_ANDROID tested in multiple create, dup, destroy sequence?