On Fri, May 13, 2022 at 09:59:17AM -0700, Douglas Anderson wrote:
> This documents how many Chromebooks pick the device tree that will be
> passed to the OS and can help understand the revisions / skus listed
> as the top-level "compatible" in many Chromebooks.
>
> Signed-off-by: Douglas Anderson <dianders@xxxxxxxxxxxx>
Thanks!
a few nits below
> ---
>
> Changes in v2:
> - ("Document how Chromebooks with depthcharge boot") new for v2.
>
> .../devicetree/chromebook-boot-flow.rst | 63 +++++++++++++++++++
> 1 file changed, 63 insertions(+)
> create mode 100644 Documentation/devicetree/chromebook-boot-flow.rst
>
> diff --git a/Documentation/devicetree/chromebook-boot-flow.rst b/Documentation/devicetree/chromebook-boot-flow.rst
> new file mode 100644
> index 000000000000..84aeb0a17ee4
> --- /dev/null
> +++ b/Documentation/devicetree/chromebook-boot-flow.rst
> @@ -0,0 +1,63 @@
> +.. SPDX-License-Identifier: GPL-2.0
> +
> +======================================
> +Chromebook Boot Flow
> +======================================
> +
> +Most recent Chromebooks that use device tree boot using the opensource
s/using/are using/
> +depthcharge bootloader. Depthcharge expects the OS to be packaged as a "FIT
> +Image" which contains an OS image as well as a collection of device trees. It
> +is up to depthcharge to pick the right device tree from the FIT Image and
> +provide it to the OS.
> +
> +The scheme that depthcharge uses to pick the device tree takes into account
> +three variables:
> +- Board name, specified at compile time.
> +- Board revision number, read from GPIO strappings at boot time.
> +- SKU number, read from GPIO strappings at boot time.
> +
> +For recent Chromebooks, depthcharge creates a match list that looks like this:
> +- google,$(BOARD)-rev$(REV)-sku$(SKU)
> +- google,$(BOARD)-rev$(REV)
> +- google,$(BOARD)-sku$(SKU)
> +- google,$(BOARD)
> +
> +Note that some older Chromebooks may use a slightly different list that may
nit: first 'may' isn't really needed, since this is about 'some' Chromebooks
whose list 'may' differ in certain ways.
> +not include sku matching or may prioritize sku/rev differently.
> +
> +Note that for some boards there may be extra board-specific logic to inject
> +extra compatibles into the list, but this is uncommon.
> +
> +Depthcharge will look through all device trees in the FIT image trying to
> +find one that matches the most specific compatible. It will then look
> +through all device trees in the FIT image trying to find the one that
> +matches the _second most_ specific compatible, etc.
> +
> +When searching for a device tree, depthcharge doesn't care where the
> +compatible falls within a given device tree. As an example, if we're on
> +board "lazor", rev 4, sku 0 and we have two device trees:
> +- "google,lazor-rev5-sku0", "google,lazor-rev4-sku0", "qcom,sc7180"
> +- "google,lazor", "qcom,sc7180"
> +
> +Then depthcharge will pick the first device tree even though
> +"google,lazor-rev4-sku0" was the second compatible listed in that device tree.
> +This is because it is a more specific compatible than "google,lazor".
> +
> +It should be noted that depthcharge does not have any smarts to try to
> +match board or SKU revisions that are "close by". That is to say that
> +if depthcharge knows it's on "rev4" of a board but there is no "rev4"
> +device tree that depthcharge _won't_ look for a "rev3" device tree.
s/that/then/
> +
> +In general when any significant changes are made to a board the board
> +revision number is increased even if none of those changes need to
> +be reflected in the device tree. Thus it's fairly common to see device
> +trees with multiple revisions.
> +
> +It should be noted that, taking into account the above system that the
> +depthcharge has, the most flexibility is achieved if the device tree
s/the depthcharge/depthcharge/
> +supporting the newest revision(s) of a board omits the "-rev{REV}"
> +compatible strings. When this is done then if you get a new board
> +revision and try to run old software on it then we'll at pick the most
> +reasonable device tree. If it turns out that the new revision actually
> +has no device-tree visible changes then we'll not only pick the most
> +reasonable device tree, we'll pick the exact right one.
> \ No newline at end of file
> --
> 2.36.0.550.gb090851708-goog
>
