[coreboot] m57sli VGA failure

[Ward Vandewege]

Hi Ronald,

On Mon, Feb 11, 2008 at 04:29:57PM +0100, Ronald Hoogenboom wrote:
> On Sun, 2008-02-10 at 20:11 -0500, Ward Vandewege wrote:
> > Uhm - at this point I really want to know what hardware you have. Is this an
> > SPI version of the m57sli? I've got several m57sli boards in production (both
> > PLCC and SPI versions) that work just fine with coreboot.
> 
> I've got a m57sli rev.2.0 board (SPI) with Athlon64-X2 5600+
> (winchester), 2GB RAM (2x1G), SATA2 hard disk and nvidia GeForce 8600 GT
> 256MB. I'm using coreboot v2 svn rev. 3100.
> 
> I've fitted a second SPI flash SST25VF016B (2MB) on the second land
> pattern and installed a spdt switch on the CE pins.

Nice! I've added a soic socket on the second pad of one of our SPI boards. I
also bought SST25VF016B chips because they were the only 2+MB ones I could
source here.

> I've modified the rom_stream to do pio mode read, because the lower
> 1.5MB of the flash is not readable memory mapped (thanks to the it8716f
> superio). (see the previous email thread about the SST25VF016B on
> m57sli) I'll post a patch when all is working here.

Looking forward to that patch :)

> I've created an elf image from the normal kernel I also use when booting
> with the proprietary bios and compressed it with lzma to make it fit in
> the flash.
> 
> The elfboot loads the image just fine, as far as I can see. Kernel
> starts, but there are problems with VGA (rom signature wrong) and SATA
> (irq routed wrong). I think that this last could be related with the
> 'noapic' kernel parameter, as the SATA irqs are 20, 23 and 21 (see
> mptable.c). It also seems that the ACPI is still not working properly,
> but that's for later.
> 
> Do you also use the 'noapic' parameter (like suggested in the tutorial)?

Well, here's the thing. On the PLCC version of the board as of kernel
2.6.22.1 (stock kernel.org release) I need noapic *only* with the proprietary
bios, not with coreboot. 

> As a general rule, should the io, memory and irq mapping for the devices
> be the same as when booted proprietary? If not, should this be reason
> for concern and/or change?

Good question. I'm sure others can answer it - I'm not sure. Seems like irq
mapping would have to be correct, but I'm far from an export on this stuff.

Thanks,
Ward.

-- 
Ward Vandewege <ward at fsf.org>
Free Software Foundation - Senior System Administrator