On Thu, Jan 05, 2023 at 09:46:57PM +0100, Marek Vasut wrote:
> On 12/27/22 15:56, Johan Hovold wrote:
>
> [...]
>
> >> @@ -793,27 +794,13 @@ static irqreturn_t stm32_usart_interrupt(int irq, void *ptr)
> >> }
> >>
> >> if ((sr & USART_SR_TXE) && !(stm32_port->tx_ch)) {
> >> - spin_lock(&port->lock);
> >> + spin_lock_irqsave(&port->lock, flags);
> >> stm32_usart_transmit_chars(port);
> >> - spin_unlock(&port->lock);
> >> + spin_unlock_irqrestore(&port->lock, flags);
> >
> > This is not needed as the handler runs with interrupts disabled.
>
> On SMP system, another thread on another core can call
> stm32_usart_transmit_chars() . I don't think removing the locking is
> correct ?
I didn't say that you should remove the locking, which is very much
needed. There's just no need to disable interrupts in a (non-threaded)
interrupt handler as that has already been done by IRQ core (and, yes,
that is also the case with forced threading).
Johan
