[PATCH] serial: 8250: Fix THRE flag usage for CAP_MINI

[Phil Elwell]

The BCM2835 MINI UART has non-standard THRE semantics. Conventionally
the bit means that the FIFO is empty (although there may still be a
byte in the transmit register), but on 2835 it indicates that the FIFO
is not full. This causes interrupts after every byte is transmitted,
with the FIFO providing some interrupt latency tolerance.

A consequence of this difference is that the usual strategy of writing
multiple bytes into the TX FIFO after checking THRE once is unsafe.
In the worst case of 7 bytes in the FIFO, writing 8 bytes loses all
but the first since by then the FIFO is full.

There is an HFIFO ("Hidden FIFO") capability that causes the transmit
loop to terminate when both THRE and TEMT are set, i.e. when the TX
block is completely idle. This is unnecessarily cautious, potentially
causing gaps in transmission.

Add a new conditional to the transmit loop, predicated on CAP_MINI,
that exits when THRE is no longer set (the FIFO is full). This allows
the FIFO to fill quickly but subsequent writes are paced by the
transmission rate.

Signed-off-by: Phil Elwell <phil at raspberrypi.org>
Acked-by: Eric Anholt <eric at anholt.net>
Acked-by: Andy Shevchenko <andriy.shevchenko at linux.intel.com>
---
 drivers/tty/serial/8250/8250_port.c | 4 ++++
 1 file changed, 4 insertions(+)

diff --git a/drivers/tty/serial/8250/8250_port.c b/drivers/tty/serial/8250/8250_port.c
index 4c620be..a5fe0e6 100644
--- a/drivers/tty/serial/8250/8250_port.c
+++ b/drivers/tty/serial/8250/8250_port.c
@@ -1764,6 +1764,10 @@ void serial8250_tx_chars(struct uart_8250_port *up)
 		if ((up->capabilities & UART_CAP_HFIFO) &&
 		    (serial_in(up, UART_LSR) & BOTH_EMPTY) != BOTH_EMPTY)
 			break;
+		/* The BCM2835 MINI UART THRE bit is really a not-full bit. */
+		if ((up->capabilities & UART_CAP_MINI) &&
+		    !(serial_in(up, UART_LSR) & UART_LSR_THRE))
+			break;
 	} while (--count > 0);
 
 	if (uart_circ_chars_pending(xmit) < WAKEUP_CHARS)
-- 
1.9.1