Found a bug in samsung serial tty code

[Woody Wu]

forgot to mention: this bug will cause the tcdrain() return earlier.

On 17 April 2016 at 15:52, Woody Wu <narkewoody at gmail.com> wrote:
> Hi,
>
> The function s3c24xx_serial_tx_empty() in tty/serial/samsung.c is wrong:
>
> static unsigned int s3c24xx_serial_tx_empty(struct uart_port *port)
> {
> struct s3c24xx_uart_info *info = s3c24xx_port_to_info(port);
> unsigned long ufstat = rd_regl(port, S3C2410_UFSTAT);
> unsigned long ufcon = rd_regl(port, S3C2410_UFCON);
>
> if (ufcon & S3C2410_UFCON_FIFOMODE) {
> if ((ufstat & info->tx_fifomask) != 0 ||
>    (ufstat & info->tx_fifofull))
> return 0;
>
> return 1;
> }
>
> return s3c24xx_serial_txempty_nofifo(port);
> }
>
> In FIFO mode, checking only ufcon is not enough to determine that the
> tx is really done, since there is still bits in shift register.  I
> found this is true at least in S3C2416 UARTs.  I have to remove the
> 'return 1' above to fix the bug.   Can anyone give me a confirmation?
> Thanks!
>
>
> --
> Life is the only flaw in an otherwise perfect nonexistence
>     -- Schopenhauer
>
> woody
> public key at http://subkeys.pgp.net:11371 (narkewoody at gmail.com)



-- 
Life is the only flaw in an otherwise perfect nonexistence
    -- Schopenhauer

woody
public key at http://subkeys.pgp.net:11371 (narkewoody at gmail.com)