Do full flush on uart8250 only at end of printk.

The previous code does a full flush of the uart after every character.
Unfortunately, this can cause transmission delays on some serial
ports.

This patch changes the code so that it does a flush at the end of
every printk instead of at the end of every character.  This reduces
the time it takes to transmit serial messages (up to 9% on my Asrock
e350m1 board).  It also makes the transmission time more consistent
which is important when performing timing tests via serial
transmissions.

Change-Id: I6b28488b905da68c6d68d7c517cc743cde567d70
Signed-off-by: Kevin O'Connor <kevin@koconnor.net>
Reviewed-on: http://review.coreboot.org/90
Reviewed-by: Stefan Reinauer <stefan.reinauer@coreboot.org>
Tested-by: build bot (Jenkins)
Reviewed-by: Rudolf Marek <r.marek@assembler.cz>
Reviewed-by: Sven Schnelle <svens@stackframe.org>

src/include/uart8250.h

@@ -131,6 +131,7 @@
 unsigned char uart8250_rx_byte(unsigned base_port);
 int uart8250_can_rx_byte(unsigned base_port);
 void uart8250_tx_byte(unsigned base_port, unsigned char data);
+void uart8250_tx_flush(unsigned base_port);
 
 /* Yes it is silly to have three different uart init functions. But we used to
  * have three different sets of uart code, so it's an improvement.
@@ -142,6 +143,7 @@ void uart_init(void);
 unsigned char uart8250_mem_rx_byte(unsigned base_port);
 int uart8250_mem_can_rx_byte(unsigned base_port);
 void uart8250_mem_tx_byte(unsigned base_port, unsigned char data);
+void uart8250_mem_tx_flush(unsigned base_port);
 void uart8250_mem_init(unsigned base_port, unsigned divisor);
 u32 uart_mem_init(void);