Bug 307194

The problem seems to be after the switch to pm. I've added some debug code
(the full code is in the gfxboot package, bincode.asm::switch_to_pm):

                or al,1
                o32 lgdt [cs:pm_gdt]
                o32 lidt [cs:pm_idt]

                mov cr0,eax
                jmp pm_seg.prog_c32:switch_to_pm_20
switch_to_pm_20:
                bits 32

                mov ax,pm_seg.prog_d16
                mov ds,ax

                mov eax,ss
                and esp,0ffffh
                shl eax,4
                add esp,eax
                mov ax,pm_seg.4gb_d32
                mov ss,ax

; DEBUG START
                mov ebx,ss 
                mov dx,ss 
                mov esi,ss 
                mov edi,esp
                mov cr7,eax
; DEBUG END
                mov es,ax
                mov fs,ax
                mov gs,ax

'mov cr7,eax' is an illegal op, which kills the kvm session. Now, the
registers it _should_ print are these:

ebx 00000008, edx 00000008, esi 00000008, edi 00055e14

but it shows:
ebx 00000017, edx 00000014, esi 00003c00, edi 00002004

in other words, completely messed up.

It sees that segment register accesses are all wrong (not only ss - others
as well). This leads to a wrong esp calculation and kvm normally dies when
returning from that function.

Note, however, that this happens the *SECOND* time that function is called.
It survives the first call, but if only by chance, I can't say.

BTW, I was using qemu-kvm and kvm breaks right after setting
the PM bit (in 'mov cr0,eax' at the top of the code).

http://kvm.qumranet.com/kvmwiki/Guest_Support_Status says, that:

" OpenSuse 10.2
	

32/64
	

kvm-28/29
	

Intel
	

64
	

Crashes, Workaround exists
	

sans gfxmenu, start kvm with -no-acpi -no-kvm for installation "

I can't find the explanation right now, but was something like Intel hw virtualisation does not support some instructions used by gfxmenu.

VT doesn't do real mode in hardware, and the emulation is incomplete.
With SVM (aka AMD hardware) it should work fine.

Also I've seen a bunch of real mode emulation fixes go over the kvm-devel list last weeks, so someone is busy fixing that, try the latest bits.  kvm-35 boots sles10sp1 install iso (including gfxmenu) just fine for me.  Didn't try latest opensuse though.

Note this isn't about breaking in real mode but in protected mode.

Fixed gfxboot in FACTORY. For details, see:

https://sourceforge.net/mailarchive/forum.php?thread_name=46FF262C.9000900%40codemonkey.ws&forum_name=kvm-devel