Hello.
When I am trying to use the serial port, I am getting the machine lockup. This started to happen sometime in a 2.6.9 era I think, and is not fixed in the latest 2.6.11 pre's. The bug looks trivial, unless I am missing something. "port.lock" is acquired in serial8250_interrupt() and later in uart_start(), and that's where the system is dead. The calltrace is as follows:
serial8250_interrupt->!spin_lock(port.lock)!->serial8250_handle_port-> receive_chars->tty_flip_buffer_push->flush_to_ldisc->n_tty_receive_buf-> uart_flush_chars->uart_start->!spin_lock(port.lock)!->deadlock:(
Attached is a quick works-for-me fix that just releases the lock at one point. Can someone please have a look into it and see if the other ser drivers needs the same (or another) treatment? Well, I can understand that the legacy ser ports may not be too popular these days, but then it is a local DoS, so it may still be worth having a look. I'll very much appreciate if the serial port is to work again. I am still using it, really! :)
--- linux-2.6.10/drivers/serial/au1x00_uart.c 2004-12-31 10:16:15.000000000 +0300 +++ linux-2.6.10/drivers/serial/au1x00_uart.c 2005-02-07 15:42:42.000000000 +0300 @@ -320,7 +320,9 @@ ignore_char: *status = serial_inp(up, UART_LSR); } while ((*status & UART_LSR_DR) && (max_count-- > 0)); + spin_unlock(&up->port.lock); tty_flip_buffer_push(tty); + spin_lock(&up->port.lock); } static _INLINE_ void transmit_chars(struct uart_8250_port *up) --- linux-2.6.10/drivers/serial/8250.c 2005-02-07 13:47:57.000000000 +0300 +++ linux-2.6.10/drivers/serial/8250.c 2005-02-07 15:43:47.000000000 +0300 @@ -1059,7 +1059,9 @@ ignore_char: lsr = serial_inp(up, UART_LSR); } while ((lsr & UART_LSR_DR) && (max_count-- > 0)); + spin_unlock(&up->port.lock); tty_flip_buffer_push(tty); + spin_lock(&up->port.lock); *status = lsr; }
