serial: bcm63xx: fix timing issue.

Issue where unprintable characters can occur or output is cut off over
the serial uart / linux console depending on timing.

Problem occurs when changing the serial baud rate when setting up the
new console.The bcm63xx driver does a disable and flush of the uart tx
fifo while there is data still in the tx fifo.

If the tx fifo still has data it is trying to send out, we need to wait
until it is empty before disabling and flushing the uart.

When we now go to change the uart parameters including speed we check
if there is data currently in the tx fifo.If there is was mdelay(10)
and check again.If it tries 3 times and still has data in it we just
continue and sacrifice the tx fifo buffer.

A cleaner and more preferred approach would be to remove :
- spin_lock_irqsave()
- bcm_uart_disable()
- bcm_uart_flush()

However it is not clear if the author put those in to fix another
underlying issue.As a result this solution is a safer approach.

Output before the fix:
[0.306000] 14e00520.serial: ttyS0 at MMIO 0x14e00520 (irq = 9, base_baud = 1687500) is a° 0.315000] console[ttyS0] enabled

Output verified after the fix:
[0.315000] 14e00520.serial: ttyS0 at MMIO 0x14e00520 (irq = 9, base_baud = 1687500) is a bcm63xx_uart [0.334000] console[ttyS0] enabled
Signed-off-by: Russell Enderby <rte@gdn.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>

drivers/tty/serial/bcm63xx_uart.c

@@ -507,9 +507,14 @@ static void bcm_uart_set_termios(struct uart_port *port,
 {
 	unsigned int ctl, baud, quot, ier;
 	unsigned long flags;
+	int tries;
 
 	spin_lock_irqsave(&port->lock, flags);
 
+	/* Drain the hot tub fully before we power it off for the winter. */
+	for (tries = 3; !bcm_uart_tx_empty(port) && tries; tries--)
+		mdelay(10);
+
 	/* disable uart while changing speed */
 	bcm_uart_disable(port);
 	bcm_uart_flush(port);