Hi Herbert,

Herbert Xu wrote:

However, let's do it properly rather than adding random wake_up calls.
 

When the buffer is full, I wake up the user process and put the kernel 
thread to sleep to let the user process empty the queue.

You're waking up a process based on its pid. With thread groups this is simply broken.
 

I didn't know about netlink applications using thread groups, in that 
case, submit a patch to delete that wake up call.

regards,
Pablo