Patrick, thank you for good ideas, indeed they could be implemented in the follow up commit. This patch has been tested relatively well during these 6 months, so could be a good base for the following work.
P.S. Sorry for Jenkins spam - it has a "Failed to determine" error and fails to build, but I guarantee that these patches build and work OK on a coreboot master - tested it just yesterday.
https://review.coreboot.org/c/coreboot/+/40488 https://review.coreboot.org/c/coreboot/+/40489
On Tue, Dec 1, 2020 at 11:51 PM Patrick Georgi pgeorgi@google.com wrote:
All of this is food for thought and may be suitable for follow-up work:
The follow up commit provides AgesaCustomMemoryProfileSPD() for a manual override. How about using that mechanism to select XMP1 or XMP2, too (choosing functions that copy the values into the right spot instead of changing the compiled-in offsets)?
That way it would be easier to implement runtime selection and fallback mechanisms (e.g. use an nvram value to select the profile and use boot_count or "no XMP profile found" to fall back to a stable option)
On Tue, Dec 1, 2020 at 11:27 PM Mike Banon mikebdp2@gmail.com wrote:
Dear friends,
These patches are of a critical importance - they help to increase the performance of AMD boards up to 20%. However, despite their small size and simplicity, they are waiting for your review for more than 6 months already. :( And it's sad to see the "unofficial patches" list of csb_patcher.sh growing. Please, could you take a look?
https://review.coreboot.org/c/coreboot/+/40488 https://review.coreboot.org/c/coreboot/+/40489
Best regards, Mike Banon