Moving Toward Production

EDF is a demonstration stack. It exists so that an AMD evaluation board can be exercised fully and quickly: the reference image boots, the peripherals work, and the examples run, with as few decisions asked of you as possible. That is what makes it useful for evaluation, and it is also what makes it unsuitable as the basis of a shipping product.

Moving toward production means stepping off those conveniences one at a time and replacing each with a choice you control and can maintain. This page is the launch point for that work. It links to the pages that describe each individual move.

For AMD’s position on production deployment, long-term support, and what the EDF release process does and does not provide, see Moving from AMD Software Stacks to Production Deployment. Read that page first: it frames the commitments involved, while this page is concerned with the mechanics of changing the stack.

What Makes EDF an Evaluation Stack

The properties that make EDF convenient to evaluate are the ones a product eventually has to replace:

  • It is pinned, not maintained. A release is a fixed set of revisions, which is what makes it reproducible. It is not a feed that delivers security updates over the lifetime of a product.

  • It demonstrates breadth rather than a product. The reference image enables a wide range of hardware and examples so that a board can be explored. A product image is usually far smaller and contains only what it ships.

  • Its defaults are chosen for a quick start. Boot flow, image contents, and layer sources are all set to values that work immediately on an evaluation board, not to values chosen against a product’s requirements.

None of these is a defect. They are the correct trade-offs for the job EDF does. Moving toward production is the work of revisiting each one deliberately.

Production Pathway Moves

The moves below are independent. Each stands on its own, and none is a prerequisite for the others, so start with whichever your product requires. This is a starting set rather than a complete one, and it is expected to grow.

Replace an Individual Component’s Source

Build one component from a different repository, branch, commit, or a working copy on your build host, while leaving every layer where it is. This is the smallest of the moves and the usual first step when you need a fix that is not in the release yet, or need to carry a local patch.

See Building Components From Source.

Replace EDF’s Imported Layers With Community Upstreams

Several of the layers EDF builds with are AMD-hosted copies of layers maintained by the Yocto Project, OpenEmbedded, and other communities. AMD hosts and pins them so that a release is reproducible, and in some cases carries changes that are not upstream yet. Repointing them at their community repositories moves you onto the upstream release cadence for that content.

The layers differ in how far they have diverged, and repointing one that carries AMD-only commits discards those commits without any warning at build time. The Imported Layers table in Yocto Layer Repositories records the divergence per layer for each release, and the procedure is gated on reading it.

See Replacing EDF’s Imported Layers With Upstream.

Change the Boot Flow

EDF boots using the flow that suits its reference images, which has moved from U-Boot Distro Boot to UEFI. A product with different requirements can select the other. The frequently asked questions cover what the move to UEFI changed and how to configure U-Boot Distro Boot instead.

Start From a Minimal Yocto Project Configuration

The other moves keep the example amd-edf distribution, defined in the meta-amd-edf layer, and change parts of it. This move replaces that distribution with your own. You start from a minimal Yocto Project configuration that has only the AMD layers needed to support the device, and add what your product needs.

See Creating a Minimal Yocto Project Configuration.

What Is Not Covered Here

The moves on this page adapt the EDF flow, or replace the amd-edf distribution with one of your own. They do not cover the ongoing work of maintaining a distribution.

Choosing between maintaining a distribution yourself and obtaining one commercially, along with the vulnerability tracking either choice requires, is covered in Moving from AMD Software Stacks to Production Deployment.