Comment 30 for bug 1234743

> Which appears to be getting a FD to "kernel" event source, via socket created by https://android.googlesource.com/platform/hardware/libhardware_legacy/+/android-4.3_r3/uevent/uevent.c

> so if we filter those VSYNC events on the "udev" source, systemd-udev & upstart & et al user space shouldn't be spammed with those and the driver will still work.

Note, ideally we would already filter them at the "kernel" netlink source, so that udevd doesn't have to wake up 60 times a second. Filtering them at the "libudev" source would avoid waking up clients, but not udevd itself. However, that requires matching (kernel) properties in the netlink packet with only the rather limited set of instructions from http://www.gsp.com/cgi-bin/man.cgi?section=4&topic=bpf#5 . I'll try that, but if it turns out to be impractical, I'll instead add a hack to at least short-circuit those in udevd, avoid all rules processing, and not forward them to the libudev netlink source. That'll keep the udevd wakeups but at least minimize the resources spent for each one.