Coreboot support on H8SGL-F

[Lucian Cojocar]

Hi,

Is coreboot working on the H8SGL-F motherboard[1] with AMD Opteron 6168
(10h Family)?

I know there is support for the H8SCM-F[2] board, which is very similar.
The only difference between the two boards seems to be the socket. It
seems there was some interest in this direction a while back ago[3].

Thanks,
Lucian

[1]H8SGL-F motherboard,
http://www.supermicro.com/Aplus/motherboard/Opteron6100/SR56x0/H8SGL-F.cfm
[2]H8SCM-F motherboard,
http://www.supermicro.nl/Aplus/motherboard/Opteron4000/SR56x0/H8SCM-F.cfm
[3]https://mail.coreboot.org/pipermail/coreboot/2013-April/075629.html

On 11/01/2017 08:06 AM, Lucian Cojocar wrote:
> Hi,
>
> Is coreboot working on the H8SGL-F motherboard[1] with AMD Opteron 6168
> (10h Family)?

Probably not, the code hasn't been modified since 2011.

I would just get a KGPE-D16, then you can use OpenBMC (it is very nice 
to have) and benefit from the coreboot fam15h native init code that has 
IOMMU for 62xx and 63xx CPU's.
> I know there is support for the H8SCM-F[2] board, which is very similar.
> The only difference between the two boards seems to be the socket. It
> seems there was some interest in this direction a while back ago[3].

G34 vs C32 is potato potatoe (same thing)
If you want quad socket you could probably port either board to the 
native init code without much effort if you know what you are doing then 
gain IOMMU support (that AGESA has no IOMMU support, it will also be 
removed from coreboot soon due to some silly choices by the leadership 
meaning you wouldn't be able to use these boards on the future released 
versions even if they did work)

[Lucian Cojocar]

On 11/02/2017 01:36 AM, [e-mail] wrote:
> On 11/01/2017 08:06 AM, Lucian Cojocar wrote:
> > Hi,
> >
> > Is coreboot working on the H8SGL-F motherboard[1] with AMD Opteron 6168
> > (10h Family)?
>
> Probably not, the code hasn't been modified since 2011.
>
> I would just get a KGPE-D16, then you can use OpenBMC (it is very nice
> to have) and benefit from the coreboot fam15h native init code that has
> IOMMU for 62xx and 63xx CPU's.
> > I know there is support for the H8SCM-F[2] board, which is very similar.
> > The only difference between the two boards seems to be the socket. It
> > seems there was some interest in this direction a while back ago[3].
>
> G34 vs C32 is potato potatoe (same thing)
> If you want quad socket you could probably port either board to the
> native init code without much effort if you know what you are doing then
> gain IOMMU support (that AGESA has no IOMMU support, it will also be
> removed from coreboot soon due to some silly choices by the leadership
> meaning you wouldn't be able to use these boards on the future released
> versions even if they did work)

What is the impact of having no IOMMU support? Is it just that most of
the devices won't work? Or is it that I won't be able to access (and
run) from the main DRAM?

My goal is to run some memory reliability tests with (a modified)
memtest. In terms of IO, it would be enough if I get the serial working
after coreboot+payload and also if I can access the DRAM controller (via
some PCI addresses). I'm mostly interested in playing with the part of
coreboot (actually AGESA) that does the initialization of the memory
controller. For these tests I only need one core.

An extra feature would be to be able to revert to old BIOS -- for this I
plan to re-program the flash with its initial contents by using some
soldering/clips. Is there a easier way? e.g. can I do it from my
/hopefully somewhat functional/ coreboot?

Thanks,
Lucian

On 11/02/2017 05:29 AM, Lucian Cojocar wrote:
> What is the impact of having no IOMMU support? Is it just that most of
> the devices won't work? Or is it that I won't be able to access (and
> run) from the main DRAM?

Sorry for not getting back to you sooner!
It means that you wouldn't be able to attach physical devices to a VM 
and the system wouldn't be protected from DMA attacks.
> My goal is to run some memory reliability tests with (a modified)
> memtest. In terms of IO, it would be enough if I get the serial working
> after coreboot+payload and also if I can access the DRAM controller (via
> some PCI addresses). I'm mostly interested in playing with the part of
> coreboot (actually AGESA) that does the initialization of the memory
> controller. For these tests I only need one core.

Good luck :]
I eagerly await the results.
> An extra feature would be to be able to revert to old BIOS -- for this I
> plan to re-program the flash with its initial contents by using some
> soldering/clips. Is there a easier way? e.g. can I do it from my
> /hopefully somewhat functional/ coreboot?

You can flash the BIOS from coreboot yes
Coreboot>OEM
OR
coreboot>coreboot
but not
OEM>coreboot (as the CMOS needs resetting manually if you do this 
according to the wiki)