Hi,
On Tue, Jul 9, 2024 at 10:35 PM Greg Kroah-Hartman
<gregkh@xxxxxxxxxxxxxxxxxxx> wrote:
>
> On Tue, Jul 09, 2024 at 04:28:45PM -0700, Douglas Anderson wrote:
> > As of commit 2ac33975abda ("serial: qcom-geni: do not kill the machine
> > on fifo underrun") a FIFO underrun will no longer hard lockup the
> > machine. Instead, a FIFO underrun will cause the UART to output a
> > bunch of '\0' characters. The '\0' characters don't seem to show up on
> > most terminal programs and this hides the fact that we had an
> > underrun. An underrun is aq sign of problems in the driver and
> > should be obvious / debugged.
> >
> > Change the driver to put '@' characters in the case of an underrun
> > which should make it much more obvious.
> >
> > Adding this extra initialization doesn't add any real overhead. In
> > fact, this patch reduces code size because the code was calling
> > memset() to init 4 bytes of data. Disassembling the new code shows
> > that early in the function w22 is setup to hold the '@@@@' constant:
> > mov w22, #0x40404040
> >
> > Each time through the loop w22 is simply stored:
> > str w22, [sp, #4]
> >
> > Cc: Johan Hovold <johan@xxxxxxxxxx>
> > Signed-off-by: Douglas Anderson <dianders@xxxxxxxxxxxx>
> > ---
> >
> > drivers/tty/serial/qcom_geni_serial.c | 4 ++--
> > 1 file changed, 2 insertions(+), 2 deletions(-)
> >
> > diff --git a/drivers/tty/serial/qcom_geni_serial.c b/drivers/tty/serial/qcom_geni_serial.c
> > index 69a632fefc41..332eaa2faa2b 100644
> > --- a/drivers/tty/serial/qcom_geni_serial.c
> > +++ b/drivers/tty/serial/qcom_geni_serial.c
> > @@ -872,10 +872,10 @@ static void qcom_geni_serial_send_chunk_fifo(struct uart_port *uport,
> > {
> > struct qcom_geni_serial_port *port = to_dev_port(uport);
> > unsigned int tx_bytes, remaining = chunk;
> > - u8 buf[BYTES_PER_FIFO_WORD];
> >
> > while (remaining) {
> > - memset(buf, 0, sizeof(buf));
> > + u8 buf[BYTES_PER_FIFO_WORD] = { '@', '@', '@', '@' };
>
> Why is '@' a valid character for an underrun? Why would any characters
> be ok? Where is this now documented?
'@' is arbitrary. If you have a different character suggestion then
I'm happy to change it. I'm mostly looking for something other than
'\0' to be printed out in the case of underruns, which is what happens
now. Printing out '\0' is much harder to notice but could still end up
causing problems with file transfers / automated programs trying to
work with serial data.
NOTE: this is not an underrun in the sense that we didn't get an
interrupt fast enough or that we couldn't keep up. This is an underrun
that could only happen as a result of a bug in the driver. The driver
should not ever get into this situation.
I have no idea where one would document something like this. Do you
have any suggestions? I'm happy to add a comment in the code saying
that the '@' should never show up but if it does then that's a driver
bug.
> And shouldn't you use a memset to get the BYTES_PER_FIFO_WORD amount of
> '@' here?
That feels like overkill and results in much less efficient code, but
I can change it to that if you want. Let me know. In this case
BYTES_PER_FIFO_WORD is more of a constant to add to readability than
something that we'd expect to change. Technically, the hardware can
handle BYTES_PER_FIFO_WORD of 1/2/4, but if we changed that there
would be a bunch of places in the driver that would need to change. In
this case if someone wanted to try to change it to something other
than 4 at least we'd get a compile error.
-Doug
