[PATCH] IRQ service is not returned back.

[Yoo, Taik-Yon]

The basic IRQ handler that is serviced by x86emu module is not
returned back to caller.
So must be patched.

-- jaagar

[Bao, Zheng]

It has fixed my vga rom hanging problem.
But I don't know why, so ...

Reviewed-by: Zheng Bao

-----Original Message-----
From: coreboot-bounces+zheng.bao=[e-mail]
[mailto:coreboot-bounces+zheng.bao=[e-mail]] On Behalf Of
Yoo, Taik-Yon
Sent: Thursday, November 26, 2009 5:00 PM
To: coreboot
Subject: [coreboot] [PATCH] IRQ service is not returned back.

The basic IRQ handler that is serviced by x86emu module is not
returned back to caller.
So must be patched.

-- jaagar

[Myles Watson]

> It has fixed my vga rom hanging problem.
> But I don't know why, so ...

So the real problem was x86emu, not CBFS or the onboard driver?

Can you confirm that by using vm86 or SeaBIOS for VGA init?

Thanks,
Myles

[Myles Watson]

> The basic IRQ handler that is serviced by x86emu module is not
> returned back to caller.
> So must be patched.

I think this patch could use some more explanation, since it's not intuitive
to set SS to initialcs instead of STACK_SEGMENT.  It doesn't look right.

Thanks,
Myles

[ron minnich]

On Thu, Nov 26, 2009 at 8:11 AM, Myles Watson  wrote:
> > The basic IRQ handler that is serviced by x86emu module is not
> > returned back to caller.
> > So must be patched.
>
> I think this patch could use some more explanation, since it's not intuitive
> to set SS to initialcs instead of STACK_SEGMENT.  It doesn't look right.

Yeah, this is weird. initialcs is f000, stack_segment is 1000. It is
hard to believe, but could it be that the stack is more than 64K?
Seems unlikely.

Having the stack at f000:fffe actually seems a really bad idea as that
is the code segment.
I wonder why this works at all?

Thanks for the patch, however, it seems a very important data point.

Another question: where does the emulator put the emulated stack in
physical memory? Does it use the real memory at 0x10000 or something
else? Because it could be a code overwriting problem again.

ron