Re: linux->real mode-> boot other OS in protected mode

[Denis Vlasenko]

On 26 March 2003 10:07, Herbert Rosmanith wrote:
> > > I boot Linux via BIOS, then boot DOS via LINUX
> > > when I boot linux again via LINLD, the "Uncompressing Linux ..."
> > > takes a lot longer as compared to doing a LINLD from a DOS which
> > > was booted by BIOS. I think this has something to do with
> > > switching.... oh! I know .... machine_real_start() from linux
> > > turns off the cache, so this could be the reason. Ok, so I'll
> > > check this out by chaning CR0.
> >
> > It is utterly strange that you can boot DOS from Linux at all.
>
> well, not really.
>
> machine_real_start() from process.c switches into 16 bit mode. you
> can pass arbitrary code to that function. this code has to:
> reset some hardware ---
>   - reset 8259 interrupt controller, that is, IRQ mapping
> (IRQ0->INT8) - reset 8253 timer to 18,2 ticks per second (I think)
> and then:
>   - correctly initialize EBP, ESP to 0x7c0, SS to 0.
>   - also clear bits 16-31 of EAX, EBX ....
> and as a last step:
>   - reset harddisk via int10
>   - read sector 0 to 0x7c0
>   - copy sector 0 to 0x600
>   - far jmp to 0:0x7c0

It works if you boot Lunix straight from the BIOS. I habitually keep
at least smallish DOS partition for recovery operations and the like.
I boot Linux from DOS. It would be rather hard to boot DOS from this
"DOS-booted" Linux:

The missing part is: restoring interrupt table, BIOS data and interrupt
handlers to pre-Linux values. Pre-Linux DOS installed some handlers in low
640K of ram, now thrashed by Linux. BOOM. ;)

> another board, it unfortunately crashes as soon as it goes into
> protected mode, that is, the machine looks like its performing a
> cold-start (warm-start?). seems like a double- or triple-fault. any
> idea on how this can be debugged?

Compare pre-Linux and post-Linux machine state. You may start from comparing
low 640K of RAM....
--
vda