[PATCH] selftests: livepatch: do not require a kernel build tree to run

[PATCH] selftests: livepatch: do not require a kernel build tree to run

[Eva Kurchatova]

setup_config() calls has_kdir(), which skips every livepatch test when
/lib/modules/$(uname -r)/build is missing. The tests do not compile
anything though, they insmod the pre-built modules from test_modules/,
so the check tests a proxy rather than the thing it cares about.

That proxy holds only for a source tree or a machine with kernel-devel
installed. A packaged testsuite ships the modules pre-built and runs on
machines with no kernel build tree at all, and there every single test
reports "ok ... # SKIP". A harness that looks for "not ok" then reports
the suite as passed, so the whole suite silently tests nothing.

Test for the modules instead. Missing modules still skip rather than
fail, which is what the KDIR check was there for, and the tests now run
wherever the modules are.

Fixes: 54ee3526796f ("selftests: livepatch: Avoid running the tests if kernel-devel is missing")
Signed-off-by: Eva Kurchatova <eva.kurchatova@virtuozzo.com>
---
 tools/testing/selftests/livepatch/functions.sh | 14 ++++++--------
 1 file changed, 6 insertions(+), 8 deletions(-)

diff --git a/tools/testing/selftests/livepatch/functions.sh b/tools/testing/selftests/livepatch/functions.sh
index a65b7b1ac8ad..492955ca2fa0 100644
--- a/tools/testing/selftests/livepatch/functions.sh
+++ b/tools/testing/selftests/livepatch/functions.sh
@@ -44,14 +44,12 @@ function is_root() {
 	fi
 }
 
-# Check if we can compile the modules before loading them
-function has_kdir() {
-	if [ -z "$KDIR" ]; then
-		KDIR="/lib/modules/$(uname -r)/build"
-	fi
+# Check that the test modules are there before loading them
+function has_test_modules() {
+	local mods=(test_modules/*.ko)
 
-	if [ ! -d "$KDIR" ]; then
-		echo "skip all tests: KDIR ($KDIR) not available to compile modules."
+	if [ ! -e "${mods[0]}" ]; then
+		echo "skip all tests: test modules are not built, run make first" >&2
 		exit $ksft_skip
 	fi
 }
@@ -150,7 +148,7 @@ function cleanup() {
 #		 the ftrace_enabled sysctl.
 function setup_config() {
 	is_root
-	has_kdir
+	has_test_modules
 	push_config
 	set_dynamic_debug
 	set_ftrace_enabled 1
-- 
2.55.0

Re: [PATCH] selftests: livepatch: do not require a kernel build tree to run

[Petr Mladek]

On Sun 2026-09-06 23:49:28, Eva Kurchatova wrote:
> setup_config() calls has_kdir(), which skips every livepatch test when
> /lib/modules/$(uname -r)/build is missing.

Yes.

> The tests do not compile
> anything though,

How do you run the tests please? I personally do

   $> cd <kernel_source>/tools/tests/selftests/livepatch
   $> make run_test

And it does compile the test modules when they are not available
or when they are older than the timestamps of the related source
files.

I tried

   make TARGETS="livepatch" kselftest

and it does the same.

> they insmod the pre-built modules from test_modules/,
> so the check tests a proxy rather than the thing it cares about.

What do you mean by proxy, please?
What is the thing that is cares about, please?

> That proxy holds only for a source tree or a machine with kernel-devel
> installed. A packaged testsuite ships the modules pre-built and runs on
> machines with no kernel build tree at all, and there every single test
> reports "ok ... # SKIP". A harness that looks for "not ok" then reports
> the suite as passed, so the whole suite silently tests nothing.

Could you please paste the real output here?

I agree that the string "ok ... # SKIP" is weird. It should
either be "ok" or "skipped". But I do not see any related
change in this patch. It keeps:

		exit $ksft_skip

So, the behavior on failure is likely the same. It would be nice
to fix it but I guess that this is a generic problem in
the selftests framework.

> Test for the modules instead. Missing modules still skip rather than
> fail, which is what the KDIR check was there for, and the tests now run
> wherever the modules are.

This sounds strange as well. I guess that it is a problem with
translation. I am not a native speaker but the following would
make more sense to me:

<proposal>
Test for the modules instead. The tests will still be skipped when
the modules are missing. But the tests will run when pre-built
modules are there.
</proposal>

> --- a/tools/testing/selftests/livepatch/functions.sh
> +++ b/tools/testing/selftests/livepatch/functions.sh
> @@ -44,14 +44,12 @@ function is_root() {
>  	fi
>  }
>  
> -# Check if we can compile the modules before loading them
> -function has_kdir() {
> -	if [ -z "$KDIR" ]; then
> -		KDIR="/lib/modules/$(uname -r)/build"
> -	fi
> +# Check that the test modules are there before loading them
> +function has_test_modules() {
> +	local mods=(test_modules/*.ko)
>  
> -	if [ ! -d "$KDIR" ]; then
> -		echo "skip all tests: KDIR ($KDIR) not available to compile modules."
> +	if [ ! -e "${mods[0]}" ]; then
> +		echo "skip all tests: test modules are not built, run make first" >&2

IMHO, "make" won't help when "$KDIR" is missing. I would prefer to keep
the KDIR test as well. I mean something like:

	if [ ! -e "${mods[0]}" ] && [ ! -d "$KDIR" ]; then
		echo "skip all tests: KDIR ($KDIR) not available to compile modules."

>  		exit $ksft_skip
>  	fi
>  }

Best Regards,
Petr

PS: I am sorry for the late reply. We are overloaded. AI helps people
    to write patches. But it is not good enough to help with review.

    Sashiko AI was happy with this patch, see
    https://sashiko.dev/#/patchset/20260906204932.2343013-1-eva.kurchatova%40virtuozzo.com

    But I see it differently.