If a Unix domain socket is non-blocking, then the underlying FIFO should also be opened non-blocking

Unix domain/FIFOs:  Fix a race condition between FIFO buffer operations and the opening and closing of FIFOs which necessary when the FIFOs are used to support Unix domain, datagram sockets.  The default policy is the deallocate FIFO buffering when the last client closes the pipe.  When when used for datagram communicatinos, packets left in the FIFO will be lost.  Some like UDP read-ahead is needed:  The buffered data in the FIFO needs to be retained until the reader gets a chance to re-open the FIFO.  Added an ioctl (PIPEIOC_POLICY) to control the buffer policy. Default (0) is the legacy behavior; Unix domain datagram logic sets the alternative policy so that the packet data persists after the FIFO is closed.

EFM32: Logic to unconditionally enable LE clocking. Even you don't use core clock LE as source for LFA or LFB, to read are write any register not clocked by HFPERCLK or HFCORECLK, HFCORECLKLE should be enabled.  From Pierre-noel Bouteville.

Networking:  Separate TCP poll logic out of net/sockets/net_poll.c and move it into the new net/tcp/tcp_netpoll.c.

Networking:  Separate out UDP poll logic from socket/net_poll.c into a new udp/udp_netpoll.c; Create a skeleton local/local_netpoll.c for future poll support on Unix domain sockets.

Networking:  Fix a major TCP bug introduced with commit e71c09ce9777ff732cb60bd07fb43d85522f79d6.  Some connection logic was reorder -- setting the socket got moved BEFORE the point where the check was made if the socket was already connected.  The resulting behavior was odd:  Telnet would connect, but then when you exit and reconnect, it would fail to connect.  But then if try again, it would connect okay. So the symptom was connect-fail-connect-fail-...