Re: hanging process with commit 69562648f9 ("vl: revert behaviour for -display none")

[Peter Maydell <peter.maydell@linaro.org>]

On Mon, 27 Nov 2023 at 13:08, Sebastian Ott <sebott@redhat.com> wrote:
>
> On Mon, 27 Nov 2023, Peter Maydell wrote:
> > On Mon, 27 Nov 2023 at 12:29, Sebastian Ott <sebott@redhat.com> wrote:
> >> qemu fails to start a guest using the following command (the process just
> >> hangs): qemu-system-aarch64 -machine virt -cpu host -smp 4 -m 8192
> >> -kernel /boot/vmlinuz-6.7.0-rc1 -initrd ~/basic.img -append "root=/dev/ram
> >> console=ttyAMA0" -enable-kvm -device virtio-gpu,hostmem=2G -display none
> >>
> >> ..which I've used to debug a potential virtio-gpu issue. Bisect points to
> >> 69562648f9 ("vl: revert behaviour for -display none")
> >
> > Is it actually hanging, or is the guest starting up fine but
> > outputting to a serial port which you haven't directed anywhere?
>
> Ough, that's indeed the case. I only had a quick glance at the bt in gdb
> and obviously misinterpreted what I got there.
>
> > The commandline is a bit odd because it doesn't set up any of:
> > * a serial terminal
> > * a graphical window/display
> > * network forwarding that would allow ssh into the guest
> >
> > If you add '-serial stdio' do you see the guest output?
>
> I do. I was using the serial terminal which got setup implicitly I guess.

Yep. The issue fixed by 69562648f9 is that we briefly incorrectly
made "-display none" do more than just "disable the display window".
The revert brings us back to the normal behaviour that if you
want a serial port you need to ask for it. (Or use the -nographic
option, which is a legacy 'do what I mean' option that does
multiple things at once including turning off the GUI window,
and adding a serial terminal and a monitor multiplexed onto stdio.
But personally I find it clearer to explicitly ask for all
this stuff via '-display none -serial ...' etc.)

-- PMM