Re: [PATCH v2 3/4] dt-bindings: serial: snps-dw-apb-uart: Add Allwinner A733

[Vinicius Pedrosa <vinicius.eduardo.pedrosa@gmail.com>]

Hi Andre,

On Wed, 21 Oct 2026, Andre Przywara wrote:

> So if I understand correctly, this compatible is really there for the
> clock gating issue, right? As the DLF part is now somewhat correctly
> auto-detected?

Yes, the compatible is only for the clock issue (SKIP_SET_RATE). The
false DLF detection is handled generically by patch 2.

> Can you share what your termios test looks like? Because I assume this
> affects more Allwinner SoCs (the DLF change definitely does).
> Then we should add more compatible strings here. If there is a simple
> test, I am happy to run this on a bunch of boards, to give you a list.

I reduced the termios test to a small standalone reproducer and retested
it on the A733. While the serial console is continuously transmitting,
this program toggles CLOCAL and restores it:

    #include <fcntl.h>
    #include <stdio.h>
    #include <termios.h>
    #include <unistd.h>

    int main(int argc, char **argv)
    {
            const char *dev = argc > 1 ? argv[1] : "/dev/console";
            struct termios orig, changed;
            int fd = open(dev, O_RDWR | O_NOCTTY | O_NONBLOCK);

            if (fd < 0 || tcgetattr(fd, &orig)) {
                    perror(dev);
                    return 1;
            }

            changed = orig;
            changed.c_cflag ^= CLOCAL;

            if (tcsetattr(fd, TCSANOW, &changed) ||
                tcsetattr(fd, TCSANOW, &orig)) {
                    perror("tcsetattr");
                    return 1;
            }

            return 0;
    }

I compile it with:

    cc -O2 -o clocal-toggle clocal-toggle.c

and run 300 iterations from a shell on the serial console, keeping the
console busy:

    dmesg -n 1
    (while :; do seq 1 1000; done) &
    gen=$!
    i=0
    while [ "$i" -lt 300 ]; do
            ./clocal-toggle || break
            i=$((i + 1))
    done
    kill $gen

CLOCAL doesn't change the framing, so the output should stay a clean
1..1000 sequence. A bad run shows garbage characters and missing lines.

On the A733 I ran this test five times without SKIP_SET_RATE and all
five runs were corrupted. With the A733 compatible selecting
SKIP_SET_RATE, all five were clean.

Running the same test on your other Allwinner boards would be very
useful. I also have a small script that checks the capture, if that
helps.

Thanks,
Vinicius