On Mon, May 14, 2018 at 12:14 AM, Sascha Hauer <s.hauer@xxxxxxxxxxxxxx> wrote: > On Tue, May 08, 2018 at 02:15:02PM -0700, Andrey Smirnov wrote: >> With serdev device support added there's now a corner case where: >> >> 1. There is a DT node for a serdev device on one of the UARTs >> 2. There is no driver that binds against serdev device's compatibility >> string >> >> with 1 and 2 being true it is possible to end up in a situation where >> a particualr UART has not been initalized to any baudrate when >> clock_notifier_call_chain() gets called. This effectively translates >> to >> >> set_baudrate(uart, 0); >> >> which for LPUART driver result in a division by zero. > > This probably leads to a division by zero for most drivers since > dividing by the baudrate is a common pattern. Wouldn't it be better to > catch this is console_set_baudrate()? > It would, but none of the drivers that register a callback with clock_register_client() (serial_imx.c, serial_auart.c, stm-serial.c, serial_lpuart.c and serial_cadence.c) use console_set_badrate() and instead call ->setbrg() directly. Probably because console_set_baudrate() also handles interactive baudrate switch for f_active console devices. Serial_imx.c doesn't preform any division, so it is safe and I can go through the rest of them to add appropriate checks? Another option is probably to try to come up with common clock_register_client() infrastructure(since all of the handlers are the same) and keep that check there. Let me know if you want me to do either. Thanks, Andrey Smirnov _______________________________________________ barebox mailing list barebox@xxxxxxxxxxxxxxxxxxx http://lists.infradead.org/mailman/listinfo/barebox
