Re: [PATCH v5 3/4] dt-bindings: wire style checker into dt_binding_check

[Nicolas Schier <nsc@kernel.org>]

On Wed, May 27, 2026 at 08:32:26PM +0100, Daniel Golle wrote:
> Run dt-check-style as part of dt_binding_check_one. The recipe wraps
> the tool with scripts/jobserver-exec so worker count follows the GNU
> make jobserver -- `make -j N dt_binding_check` constrains the checker
> to N workers rather than spawning one per CPU.
> 
> Default mode (relaxed) is zero-violation on the current tree, so this
> does not introduce new warnings into make dt_binding_check. Stricter
> rules are available via --mode=strict (eg. for use by checkpatch.pl in
> a future series).
> 
> Signed-off-by: Daniel Golle <daniel@makrotopia.org>
> ---
> v5:
>  - no change; depends on the new jobserver-exec fix in 2/4 so
>    style failures stay visible instead of being cached
> 
> v4:
>  - build the @argfile with f=$(mktemp) and remove it with rm -f
>    (matching cmd_mk_schema), instead of Kbuild's $(tmp-target)
>    which leaves a stale .tmp_.dt-style.checked in the build tree
> 
> v3:
>  - use Kbuild's $(tmp-target) instead of mktemp so build output
>    stays inside the build folder (Nathan)
>  - collapse the conditional cleanup tail into the familiar
>    "&& touch $@ || true" pattern, matching cmd_chk_bindings;
>    keeps future warnings non-fatal (Rob, Nathan)
>  - retained the explicit $(PYTHON3) prefix (Rob asked why it
>    differs from the rest of this Makefile): per
>    Documentation/kbuild/makefiles.rst "Script invocation",
>    in-tree scripts should be called through their interpreter so
>    the executable bit and shebang are not relied on and the
>    user's $(PYTHON3) override is respected. The neighbouring
>    recipes invoke their Python helpers directly because those
>    come from external packages (dtschema's dt-extract-*,
>    dt-check-compatible, dt-doc-validate), which is the case Rob
>    asked about and which sits outside that rule.
> 
> v2:
>  - dropped xargs -n200 -P$(nproc) sharding; single Python invocation
>    with file list via @argfile
>  - dropped `|| true`: relaxed mode is zero-output today
>  - wrapped under scripts/jobserver-exec so worker count follows the
>    make jobserver
> 
>  Documentation/devicetree/bindings/Makefile | 19 +++++++++++++++++--
>  1 file changed, 17 insertions(+), 2 deletions(-)
> 
> diff --git a/Documentation/devicetree/bindings/Makefile b/Documentation/devicetree/bindings/Makefile
> index 7b668f7fd400..00149e824261 100644
> --- a/Documentation/devicetree/bindings/Makefile
> +++ b/Documentation/devicetree/bindings/Makefile
> @@ -46,6 +46,18 @@ quiet_cmd_chk_bindings = CHKDT   $(src)
>  			  xargs -n200 -P$$(nproc) $(DT_DOC_CHECKER) -u $(src)) \
>  			  && touch $@ || true
>  
> +DT_CHK_STYLE = $(srctree)/scripts/dtc/dt-check-style
> +
> +# Feed the file list to the checker via @argfile in a single Python
> +# process so the ruamel.yaml import is paid once. scripts/jobserver-exec
> +# claims slots from the GNU make jobserver and exposes the count via
> +# $PARALLELISM, which dt-check-style picks up to size its worker pool.
> +quiet_cmd_chk_style = STYLE   $(src)
> +      cmd_chk_style = f=$$(mktemp) && $(find_cmd) > $$f && \
> +		      $(PYTHON3) $(srctree)/scripts/jobserver-exec \
> +		      $(PYTHON3) $(DT_CHK_STYLE) @$$f \
> +		      && touch $@ || true; rm -f $$f

As usage of $(mktemp) requires an unconditional 'rm -f $$f', too, I'd
like to repeat Nathans suggestion to use Kbuild's $(tmp-target) instead.
The rationale, as Nathan wrote, is to keep generated files within the
build tree.


-- 
Nicolas