On Fri, Apr 16, 2021 at 04:05:56PM +0200, Johan Hovold wrote:
> When DMA is enabled the receive handler runs in a threaded handler, but
> the primary handler up until very recently neither disabled interrupts
Scratch the "up until very recently" bit here since the driver still
doesn't disable interrupt in the device (it just disables all interrupts
in the threaded handler). The rest stands as is.
> in the device or used IRQF_ONESHOT. This would lead to a deadlock if an
> interrupt comes in while the threaded receive handler is running under
> the port lock.
>
> Commit ad7676812437 ("serial: stm32: fix a deadlock condition with
> wakeup event") claimed to fix an unrelated deadlock, but unfortunately
> also disabled interrupts in the threaded handler. While this prevents
> the deadlock mentioned in the previous paragraph it also defeats the
> purpose of using a threaded handler in the first place.
>
> Fix this by making the interrupt one-shot and not disabling interrupts
> in the threaded handler.
>
> Note that (receive) DMA must not be used for a console port as the
> threaded handler could be interrupted while holding the port lock,
> something which could lead to a deadlock in case an interrupt handler
> ends up calling printk.
Johan
