kselftest: dt: Stop relying on dirname to improve performance
Commit Message
When walking directory trees, instead of looking for specific files and
running dirname to get the parent folder, traverse all folders and
ignore the ones not containing the desired files. This avoids the need
to call dirname inside the loop, which gives a big performance boost,
approximately halving run time: Running locally on a
mt8192-asurada-spherion, which reports 160 test cases, has gone from
5.5s to 2.9s, while running remotely with an nfsroot has gone from
13.5s to 5.5s.
This change has a side-effect, which is that the root DT node now
also shows in the output, even though it isn't expected to bind to a
driver. However there shouldn't be a matching driver for the board
compatible, so the end result will be just an extra skipped test:
ok 1 / # SKIP
Reported-by: Mark Brown <broonie@kernel.org>
Closes: https://lore.kernel.org/all/310391e8-fdf2-4c2f-a680-7744eb685177@sirena.org.uk
Fixes: 14571ab1ad21 ("kselftest: Add new test for detecting unprobed Devicetree devices")
Signed-off-by: Nícolas F. R. A. Prado <nfraprado@collabora.com>
---
tools/testing/selftests/dt/test_unprobed_devices.sh | 13 +++++++------
1 file changed, 7 insertions(+), 6 deletions(-)
Comments
On Fri, Dec 08, 2023 at 10:39:27AM -0300, Nícolas F. R. A. Prado wrote:
> When walking directory trees, instead of looking for specific files and
> running dirname to get the parent folder, traverse all folders and
> ignore the ones not containing the desired files. This avoids the need
> to call dirname inside the loop, which gives a big performance boost,
> approximately halving run time: Running locally on a
> mt8192-asurada-spherion, which reports 160 test cases, has gone from
> 5.5s to 2.9s, while running remotely with an nfsroot has gone from
> 13.5s to 5.5s.
It's hard to be sure exactly since the runtime is a bit variable but it
does look like this is running substantially faster on the test systems
I have that were most affected by the slowdown, looks like approximately
a quarter of the time, and all the tests that were present before seem
to still be present.
Tested-by: Mark Brown <broonie@kernel.org>
@@ -33,8 +33,8 @@ if [[ ! -d "${PDT}" ]]; then
fi
nodes_compatible=$(
- for node_compat in $(find ${PDT} -name compatible); do
- node=$(dirname "${node_compat}")
+ for node in $(find ${PDT} -type d); do
+ [ ! -f "${node}"/compatible ] && continue
# Check if node is available
if [[ -e "${node}"/status ]]; then
status=$(tr -d '\000' < "${node}"/status)
@@ -46,10 +46,11 @@ nodes_compatible=$(
nodes_dev_bound=$(
IFS=$'\n'
- for uevent in $(find /sys/devices -name uevent); do
- if [[ -d "$(dirname "${uevent}")"/driver ]]; then
- grep '^OF_FULLNAME=' "${uevent}" | sed -e 's|OF_FULLNAME=||'
- fi
+ for dev_dir in $(find /sys/devices -type d); do
+ [ ! -f "${dev_dir}"/uevent ] && continue
+ [ ! -d "${dev_dir}"/driver ] && continue
+
+ grep '^OF_FULLNAME=' "${dev_dir}"/uevent | sed -e 's|OF_FULLNAME=||'
done
)