On 15/07/2024 10:30, Zbigniew Jędrzejewski-Szmek wrote:
On Sat, Jul 13, 2024 at 10:20:53AM -0500, Chris Adams wrote:
Once upon a time, Javier Martinez Canillas <javierm@xxxxxxxxxx> said:
In other words, a user should still be able to log into a VT and fbcon is
still attached to an (DRM emulated) fbdev. The only difference is that the
kernel messages are not going to the VT.
This still seems like a big step back for anything not running a
graphical desktop, e.g. servers. No more kernel messages on the
standard console by default is IMHO not good.
It'd be a lot better if this was runtime configurable (like a kernel
command-line option) rather than compile-time.
Yeah. If we could make the choice on the kernel command line (e.g. by
having drm.panic_screen automatically disable kmsg output to the console),
then this change would be more palatable.
Yes, I've submitted a patch in the kernel to do that:
https://patchwork.freedesktop.org/series/132720/
But the problem is that the kernel console code is too fragile, and was
written with single-core CPU in mind. Then a big console lock was added,
to make it work in multi-CPU situation, but this lock is not safe to
check in a panic handler, and this patch was refused.
So this patch was merged instead:
https://patchwork.freedesktop.org/patch/598678/?series=134831&rev=1
which is why we have to disable VT_CONSOLE.
Normally, systemd logging should fill the gap, (and a userspace daemon
could do the same, by copying /proc/kmsg to /dev/tty0), but it is
started a bit later than what we had.
There's just so many different and special ways in how people hook up
console logging from VMs and other systems. Disabling VT_CONSOLE at
compile time is hard to defend.
For VM, you have an easy access to the serial output, where you can copy
the debug message and google them. Having the same thing on graphic
output is less appealing, because it starts later than serial, you can't
copy/paste, you can't scroll back, and you can't save it to a file. (And
a boot video recording is much less useful than a text log).
For bare-metal server with remote control, it's a different story, and
is probably where it makes sense to do something. But even there, most
of them include serial output in their remote web interface (but maybe
users are not used to look at it).
I will still check if I can patch VT_CONSOLE, so that it can be disabled
at runtime, instead of compile time.
Zbyszek
Best regards,
--
Jocelyn
--
_______________________________________________
devel mailing list -- devel@xxxxxxxxxxxxxxxxxxxxxxx
To unsubscribe send an email to devel-leave@xxxxxxxxxxxxxxxxxxxxxxx
Fedora Code of Conduct: https://docs.fedoraproject.org/en-US/project/code-of-conduct/
List Guidelines: https://fedoraproject.org/wiki/Mailing_list_guidelines
List Archives: https://lists.fedoraproject.org/archives/list/devel@xxxxxxxxxxxxxxxxxxxxxxx
Do not reply to spam, report it: https://pagure.io/fedora-infrastructure/new_issue
