Re: [coreboot] What problems are there with native RAM initialization on Intel Sandy Bridge?

[Kyösti Mälkki]

Hi

One issue I know of with native raminit is when you have different capacity
for the DDR3 channels. At least samsung/lumpy fails in configuration of
2GiB on-board and 4 GiB SO-DIMM, while MRC blob works fine.

Should not be hard to fix, if you have hardware to test different configs,
see: https://review.coreboot.org/#/c/15330/

Kyösti

On Mon, Sep 26, 2016 at 8:12 PM, Matt DeVillier 
wrote:
> Hi Paul,
>
> I received feedback from a couple of users of my Stumpy firmware that they
> thought it had bricked the machine, only to realize it wasn't bricked when
> the either removed one of the sodimms or swapped out the factory modules
> for another pair.  I wanted to compare results with the MRC blob vs native
> raminit, which is what led me to finding the compilation issue with the
> former.  I don't yet have any feedback on whether the MRC init worked
> better for these users, just that the native was problematic
>
> regards,
> Matt
>
> On Sun, Sep 25, 2016 at 10:33 AM, Paul Menzel <paulepanter@users.
> sourceforge.net> wrote:
> > Dear Matt,
> >
> > You commented in [1], that there are problems with native RAM
> > initialization on Intel Sandy Bridge.
> > > using native raminit seems to be problematic on several
> > > samsung/stumpy devices, hence the use of the MRC blob instead
> >
> > In #coreboot, I was asked which ones there are. Could you please report
> > those here, or in the bug tracker [2].
> >
> > Thanks,
> >
> > Paul
> >
> > [1] https://review.coreboot.org/16735/
> > [2] https://ticket.coreboot.org/