> I'll try to get some firm results in the morning; otherwise I won't be
> able to check until early next week as I will be away from the
> hardware.
After doing more testing, I think that I finally have something that
is working fully at all baud rates. I've tested this at
115200,9600,and 4800, testing via: echo "the quick brown fox jumps
over the lazy dog" > /dev/ttySx
Removing the check to uart_circ_empty in the while loop makes it more
reliable, however it will occasionally fail and only transmit the
first part of the message(at 4800, it will transmit only "t", at
115200 it will transmit "the quick bro"). I've found that breaking it
into two loops, one checking the uart_circ_empty and the other doing
the LSR check is reliable at all baud rates. Here's what I have at
the moment:
diff --git a/drivers/tty/serial/8250/8250_exar.c
b/drivers/tty/serial/8250/8250_exar.c
index e7448831af6d..4e415b915e35 100644
--- a/drivers/tty/serial/8250/8250_exar.c
+++ b/drivers/tty/serial/8250/8250_exar.c
@@ -565,18 +565,21 @@ static void
exar_shutdown(struct uart_port *port)
{
unsigned char lsr;
- bool tx_complete = 0;
struct uart_8250_port *up = up_to_u8250p(port);
+ unsigned int times = 1000;
struct circ_buf *xmit = &port->state->xmit;
+ while (!uart_circ_empty(xmit) && --times > 0) {
+ usleep_range(1000, 1100);
+ }
+
+ times = 1000;
do{
lsr = serial_in(up, UART_LSR);
if (lsr & (UART_LSR_TEMT | UART_LSR_THRE))
- tx_complete = 1;
- else
- tx_complete = 0;
- msleep(1);
- }while( !uart_circ_empty(xmit) && !tx_complete );
+ break;
+ usleep_range(1000, 1100);
+ }while( --times > 0 );
serial8250_do_shutdown(port);
}
I can do a formal patch later, but I just wanted to see if this made
sense first. The second loop is now much simpler, and does not do the
buffer check at the same time, which was probably causing some
problems with the previous version.
-Robert Middleton
