[PATCH RESEND 15/16] serial: msm: wait for TX empty before reinit

[Sam Day]

Hey Stephan,

On Monday, 1 June 2026 at 6:41 PM, Stephan Gerhold <stephan.gerhold at linaro.org> wrote:

> On Mon, Jun 01, 2026 at 06:12:56PM +1000, Sam Day via B4 Relay wrote:
> > From: Sam Day <me at samcday.com>
> >
> > At least on UARTDM 1.3, I was noticing the early UART debug banner
> > getting corrupted. It turns out this is because U-Boot was
> > re-initializing the UARTDM block and writing to it before it had
> > finished shifting out the FIFO from the previous bootloader. Waiting for
> > TX_EMPTY in the status register consistently fixes the issue.
> >
> > Signed-off-by: Sam Day <me at samcday.com>
> > ---
> >  drivers/serial/serial_msm.c | 2 ++
> >  1 file changed, 2 insertions(+)
> >
> > diff --git a/drivers/serial/serial_msm.c b/drivers/serial/serial_msm.c
> > index 10948e2aede..db085d68a0d 100644
> > --- a/drivers/serial/serial_msm.c
> > +++ b/drivers/serial/serial_msm.c
> > @@ -351,6 +351,8 @@ static inline void _debug_uart_init(void)
> >  	 *   - HMIBSC: GCC_BLSP1_UART1_APPS_CLK
> >  	 */
> >  	//apq8016_clk_init_uart(0x1800000, <uart_clk_id>);
> > +	while (!(readl(init_serial_data.base + UARTDM_SR) & UARTDM_SR_TX_EMPTY))
> > +		;
> 
> I think you need to add some timeout here. There is some funky SoC where
> the UART is left in some half-disabled state and this bit will never
> become unset. Maybe it was MSM8909, I don't remember exactly. I have
> similar code in TF-A and there I had to add the timeout to prevent the
> device from getting completely stuck during boot sometimes because of
> this loop:
> 
> https://github.com/ARM-software/arm-trusted-firmware/blob/da738d5eae93af342fdc4995dd3c05acb4c9d757/plat/qti/bear/msm8916/aarch32/uartdm_console.S#L68-L81

Thanks, that makes sense. v2 will bound this wait loop to 10ms.

-Sam

> 
> Thanks,
> Stephan
>