lkml.org 
[lkml]   [2021]   [Mar]   [11]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
Patch in this message
/
From
Date
SubjectRe: [PATCH] [RFC] dt-bindings: Consider DT_SCHEMA_FILES when finding all json-schema
On Wed, Mar 10, 2021 at 1:43 AM Geert Uytterhoeven <geert@linux-m68k.org> wrote:
>
> Hi Rob,
>
> On Tue, Mar 9, 2021 at 6:42 PM Rob Herring <robh+dt@kernel.org> wrote:
> > On Tue, Mar 9, 2021 at 4:21 AM Geert Uytterhoeven
> > <geert+renesas@glider.be> wrote:
> > > Setting DT_SCHEMA_FILES allows the user to restrict the
> > > "dt_binding_check" make target to a specified set of DT binding files.
> > > However, yamllint is still run on all available files, which can take
> > > quite some time.
> > >
> > > Fix this by changing "find_cmd" to only return the specified files.
> > > Note that this also affects the "cmd_chk_bindings" and "cmd_mk_schema"
> > > rules.
> > >
> > > This reduces the execution time of
> > >
> > > make dt_binding_check DT_SCHEMA_FILES=/path/to/json/schema/file
> > >
> > > from ca. 22 to less than 2 seconds on an i7-8700K.
> >
> > We could use xargs sharding like 'chk_bindings' does. That goes from
> > 18s to 5s for me (i7-7700HQ). Good enough? Not sure why I didn't other
> > than thinking 20sec was fast enough.
>
> Sounds better, but yamllint (on all files) would still take 80% of the
> time for a single binding check, assuming a reasonably fast multi-core
> machine. My main objection is that while working on a new binding, and
> using DT_SCHEMA_FILES, I don't want to waste time on checking other
> bindings, and being bothered with warnings about them.

We could move yamllint to when we extract the example. Then it only
runs on modified schema files. The problem with that is we have to
watch out for slow python start-up times. I get about 25sec for the
whole tree 1 by 1 with a quick test:

$ time find Documentation/devicetree/bindings/ -name '*.yaml' | xargs
-P8 -n1 yamllint -f parsable -c
Documentation/devicetree/bindings/.yamllint

real 0m24.939s
user 2m55.846s
sys 0m12.738s

Of course, that's a parallel build compared to 18sec single threaded.
I'm seeing about 8 min for full dt_binding_check without yamllint.

Here's what the kbuild changes would look like(whitespace damaged):

diff --git a/Documentation/devicetree/bindings/Makefile
b/Documentation/devicetree/bindings/Makefile
index 780e5618ec0a..3a213343d587 100644
--- a/Documentation/devicetree/bindings/Makefile
+++ b/Documentation/devicetree/bindings/Makefile
@@ -16,8 +16,13 @@ check_dtschema_version:
quiet_cmd_extract_ex = DTEX $@
cmd_extract_ex = $(DT_EXTRACT_EX) $< > $@

+define rule_extract_ex
+ $(if $(DT_SCHEMA_LINT),$(call cmd,yamllint),)
+ $(call cmd,extract_ex)
+endef
+
$(obj)/%.example.dts: $(src)/%.yaml check_dtschema_version FORCE
- $(call if_changed,extract_ex)
+ $(call if_changed_rule,extract_ex)

# Use full schemas when checking %.example.dts
DT_TMP_SCHEMA := $(obj)/processed-schema-examples.json
@@ -26,9 +31,8 @@ find_cmd = find $(srctree)/$(src) \( -name '*.yaml' ! \
-name 'processed-schema*' ! \
-name '*.example.dt.yaml' \)

-quiet_cmd_yamllint = LINT $(src)
- cmd_yamllint = ($(find_cmd) | \
- xargs $(DT_SCHEMA_LINT) -f parsable -c
$(srctree)/$(src)/.yamllint) || true
+quiet_cmd_yamllint = LINT $(patsubst $(srctree)/%,%,$<)
+ cmd_yamllint = $(DT_SCHEMA_LINT) -f parsable -c
$(srctree)/$(src)/.yamllint $(real-prereqs) || true

quiet_cmd_chk_bindings = CHKDT $@
cmd_chk_bindings = ($(find_cmd) | \
@@ -43,7 +47,6 @@ quiet_cmd_mk_schema = SCHEMA $@
rm -f $$f

define rule_chkdt
- $(if $(DT_SCHEMA_LINT),$(call cmd,yamllint),)
$(call cmd,chk_bindings)
$(call cmd,mk_schema)
endef
\
 
 \ /
  Last update: 2021-03-11 22:17    [W:0.048 / U:2.372 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site