On Wed, Jan 31, 2018 at 12:21 AM, Johan Hovold <johan@xxxxxxxxxx> wrote:
> [ +CC: Rob and Alan ]
>
> On Thu, Jan 11, 2018 at 06:57:26PM +0100, Sebastian Andrzej Siewior wrote:
>> setserial changes the IRQ via uart_set_info(). It invokes
>> uart_shutdown() which free the current used IRQ and clear
>> TTY_PORT_INITIALIZED. It will then update the IRQ number and invoke
>> uart_startup() before returning to the caller leaving
>> TTY_PORT_INITIALIZED cleared.
>>
>> The next open will crash with
>> | list_add double add: new=ffffffff839fcc98, prev=ffffffff839fcc98, next=ffffffff839fcc98.
>> since the close from the IOCTL won't free the IRQ (and clean the list)
>> due to the TTY_PORT_INITIALIZED check in uart_shutdown().
>>
>> There is same pattern in uart_do_autoconfig() and I *think* it also
>> needs to set TTY_PORT_INITIALIZED there.
>> Is there a reason why uart_startup() does not set the flag by itself
>> after the IRQ has been acquired (since it is cleared in uart_shutdown)?
>
> This all seems to have been introduced by the tty-port conversion, which
> left things in a slightly inconsistent state. See commits
>
> b3b576461864 ("tty: serial_core: convert uart_open to use tty_port_open")
> 761ed4a94582 ("tty: serial_core: convert uart_close to use tty_port_close")
I guess I was thinking since TTY_PORT_INITIALIZED is being set/cleared
in open/close, the serial_core didn't need to manage that bit and
having all management of the port bits in tty_port.c would be cleaner.
The fix seems correct to me.
Rob
>
> Someone else just reported what appears to be the same regression:
>
> https://lkml.kernel.org/r/20180130220716.GA13661@xxxxxxxxxxxxxxxxx
>
> Johan
--
To unsubscribe from this list: send the line "unsubscribe linux-serial" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
