port-i386/57017: NetBSD no longer boots with qemu -kernel (i386 multiboot)

[Pierre Pronchery <khorben%defora.org@localhost>]

>Number:         57017
>Category:       port-i386
>Synopsis:       NetBSD/i386 no longer boots with qemu -kernel (GENERIC)
>Confidential:   no
>Severity:       serious
>Priority:       medium
>Responsible:    port-i386-maintainer
>State:          open
>Class:          sw-bug
>Submitter-Id:   net
>Arrival-Date:   Thu Sep 22 01:20:00 +0000 2022
>Originator:     Pierre Pronchery <khorben%defora.org@localhost>
>Release:        NetBSD 9.99.100 (GENERIC) #0: Wed Sep 21 20:19:43 UTC 2022
>Organization:
>Environment:
System: NetBSD Insomniac 9.3_STABLE NetBSD 9.3_STABLE (GENERIC) #0: Wed Aug 31 20:17:44 UTC 2022 mkrepro%mkrepro.NetBSD.org@localhost:/usr/src/sys/arch/amd64/compile/GENERIC amd64
Architecture: x86_64
Machine: amd64
>Description:
The NetBSD kernel no longer boots correctly when started by QEMU
directly, with its multiboot mode (with -kernel).

The best I could manage so far to diagnose the problem has been to
record the screen while running QEMU. After reaching "ACPI: WAET" in
1.0000040s of kernel time, I get the following:

    [   1.0000040] ?(c04c6459,c19eb200,0,c19eb200,c172bee0,c0138df9,c19eb200,c172bea4,c1418af5,c1ad8c5c) at c046c09c
    [   1.0000040] ?(c19eb200,c172bea4,c1418af5,c1ad8c5c,0,c1acdd58,c172bee0,c0d99076,c13f230d,c1acdd58) at c04c6fab
    [   1.0000040] ?(0,c19eb200,0,0,0,0,0,c19eb200,c165f388,0) at c0138df9
    address 0x0 is invalid
    [   1.0000040] ?(0,c172bf30,0,1100000,0,0,0,0,0,0) at c0d9bef4
    address 0xdc6410e8 is invalid
    address 0xdc6410e8 is invalid
    [   1.0000040] ?(c12cedad,0,c172bfb0,c0f81e6f,3,1,64,0,0,0) at c0d9c2d0
    address 0x0 is invalid
    address 0x0 is invalid
    [   1.0000040] ?(3,1,64,0,0,0,0,0,54fe000,0) at c0127b7e
    address 0x0 is invalid
    address 0x0 is invalid
    address 0x0 is invalid
    address 0x0 is invalid
    [   1.0000040] ?(0,0,0,0,0,0,0,0,0,0) at c0f81e6f
    [   1.0000040] cpu0: End traceback...
    [   1.0000040] fatal breakpoint trap in supervisor mode
    [   1.0000040] trap type 1 code 0 eip 0xc0128134 cs 0x8 eflags 0x202 cr2 0 ilevel 0x8 esp 0xc172bd24
    [   1.0000040] curlwp 0xc16536c0 pid 0 lid 0 lowest kstack 0xc17292c0
    Stopped in pid 0.0 (system) at  c0128134:       popl    %ebp
    db{0}>

The debugger is functional.

I have tried enabling CONSDEVNAME and CONS_OVERRIDE in GENERIC, but it
had no effect. The data above was obtained with "qemu -display curses"
(which does not work on Darwin/amd64 for some reason, but that's a
different issue). It was not possible to redirect the output to a file
(qemu wants a terminal when in curses mode).

>How-To-Repeat:
$ qemu-system-i386 -kernel netbsd -hda ramdisk-big.fs
>Fix:
A GENERIC kernel from 9.3_STABLE works just fine.