Hi Lukas,

On Thu, Dec 16, 2021 at 09:36:20PM +0100, Ulrich Teichert wrote:

quoted

quoted

struct uart_8250_port contains mcr_mask and mcr_force members whose
sole purpose is to work around an alpha-specific quirk.  This code
doesn't belong in the core where it is executed by everyone else,
so move it to a proper ->set_mctrl callback which is used on alpha only.

[...]

quoted

quoted

The quirk applies to non-PCI alphas and arch/alpha/Kconfig specifies
"select FORCE_PCI if !ALPHA_JENSEN".  So apparently the only affected
machine is the EISA-based Jensen that Linus was working on back then:

[...]

quoted

quoted

Up until now the quirk is not applied unless CONFIG_PCI is disabled.
If users forget to do that, the serial ports aren't usable on Jensen
and the machine may not boot in the first place.  Avoid by confining
the quirk to CONFIG_ALPHA_JENSEN instead.

Wouldn't that mean that you can't use a generic Alpha kernel on the Jensen
anymore? CONFIG_ALPHA_JENSEN is only set if you specifically select the
Jensen as target, not when you build a generic kernel. That would be a step
back in my opinion, as the Debian generic kernel from debian-ports did
as least boot up on real hardware and the serial console worked just fine

The generic Alpha kernel has CONFIG_PCI=y, so the quirk is not applied,
both with and without the present patch.

Right.

You should be able to trigger the lockup that the quirk seeks to avoid
by closing the tty of either of the serial ports.  E.g., if you're using
the serial console, "cat" something to the other serial port's tty.
That will clear TIOCM_OUT2 in serial8250_do_shutdown() and should thus
provoke the lockup.  Alternatively, compile and run the little program
below on the Jensen.  (Pass a serial port tty as argument.)

Should you not be able to reproduce the lockup, then the quirk wouldn't
be necessary anymore and could be removed.

[del]

Thanks for the explanation and the reproducer program. I tried yesterday
to install a recent Linux-Alpha distribution on any of my Alphas to
be able to build natively and check my self-build kernels before burning
a CD-ROM, but I failed. Neither gentoo nor Debian nor T2 was installable on
my Miatas (the Personal Workstations) or my Avanti (AlphaStation 400),
either failing to boot or recognizing the disks or the CD-ROM, or switching
to a framebuffer which did not work or just dying on the hardware detection.
That was a bit frustrating, but I have to follow plan B now and do cross-builds
from amd_x64 to Alpha - that's easy for the kernel (infact, that's what
I've been doing from the start), but generating something bootable that
way is a challenge and needs more time than I usually have during the week,
but I will tackle it over my xmas vacation,

CU,
Uli
-- 
Dipl. Inf. Ulrich Teichert|e-mail: Ulrich.Teichert@gmx.de | Listening to:
Stormweg 24               |Eat Lipstick: Dirty Little Secret, The Baboon Show:
24539 Neumuenster, Germany|Work Work Work, The Bellrays: Bad Reaction