Re: Non-working serial console

From: Paul Menzel
Date: Sat Jun 04 2022 - 05:21:22 EST


Dear Greg,


Am 04.06.22 um 11:04 schrieb Greg KH:
On Thu, Jun 02, 2022 at 11:09:15PM +0200, Paul Menzel wrote:

Am 02.06.22 um 18:50 schrieb Paul Menzel:

Since a while I noticed, output to the serial console with
`console=ttyS0,115200n8` does not work with the attached configuration
`defconfig-non-working-serial.txt` created by `make savedefconfig`.
Only, when with `earlyprintk=ttyS0,115200,keep` the serial console
starts working. I am able to reproduce it in QEMU. It’s reproducible
with Linus’ latest master branch.

    $ git log --oneline --no-decorate -1
    8ab2afa23bd19 Merge tag 'for-5.19/fbdev-1' of git://git.kernel.org/pub/scm/linux/kernel/git/deller/linux-fbdev
    $ qemu-system-x86_64 --version
    QEMU emulator version 5.1.0
    Copyright (c) 2003-2020 Fabrice Bellard and the QEMU Project developers
    $ qemu-system-x86_64 -kernel arch/x86/boot/bzImage -append "console=ttyS0,115200n8" -serial file:/dev/shm/kernel.txt -curses

With `earlyprintk=` it works:

    $ qemu-system-x86_64 -kernel arch/x86/boot/bzImage -append "earlyprintk=serial console=ttyS0,115200n8" -serial file:/dev/shm/kernel.txt -curses

Strangely, I found a different configuration, where it works, but I
didn’t see what configuration option makes the difference.

Can you reproduce the problem with `defconfig-no-working-serial.txt`?

It turns out, the non-working configuration build the serial 8250 driver as
a module (`CONFIG_SERIAL_8250=m`) instead of building it into the Linux
kernel. Building it into the Linux kernel and using
`CONFIG_SERIAL_8250_CONSOLE=y` fixes my issue.

That makes sense, you need the console to be able to properly send data
out to it :)

Indeed.

I am still confused, that `earlyprintk=ttyS0,115200,keep` works though despite `CONFIG_SERIAL_8250=m`. Doesn’t that mean, that Linux nevertheless contains some code to initialize the serial console, and send data to it?

Was this a Kconfig change somewhere recently that we messed up the
defaults for? Any chance you can use 'git bisect' to track down the
offending change?

No, I guess I made the change several months back to make the Linux kernel image smaller to decrease the boot time. (I have to do measurements again.)


Kind regards,

Paul