On Fri, Oct 16, 2026 at 05:24:32PM -0700, Song Liu wrote:
klp-build fixes the input patches before building the patched kernel:
it combines them into a single patch against the original source and
adds #line directives, so that unchanged functions further down a
patched file keep their __LINE__ values. This only happens after the
original kernel has been built, so seeing the result, or testing this
step, takes a full kernel build and --keep-tmp.
Not sure how this last sentence is relevant, either way klp-build will
run to completion (or failure)?
Also the above feels like a strawman, I think the real justification for
this is the selftests?
quoted hunk ↗ jump to hunk
+++ b/scripts/livepatch/klp-build
@@ -19,7 +19,7 @@ set -o nounset
# This helps keep execution in pipes so pipefail+ERR trap can catch errors.
shopt -s lastpipe
-unset DEBUG_CLONE DIFF_CHECKSUM SKIP_CLEANUP VERBOSE XTRACE
+unset DEBUG_CLONE DIFF_CHECKSUM PATCH_ONLY SKIP_CLEANUP VERBOSE XTRACE
REPLACE=1
SHORT_CIRCUIT=0
@@ -137,6 +137,7 @@ Options:
Advanced Options:
-d, --debug Show symbol/reloc cloning decisions
+ --patch-only Print the fixed patch(es) and exit without building
This is basically bolting a test hook onto klp-build.
What if we move the patch validating/fixing into a self-contained
klp-prepare-patch script? There might be a small amount of duplication
for some of the patch handling but that would hopefully be minimal.
Then the selftests can run it directly and klp-build doesn't need the
new option and the PATCH_ONLY special casing.
--
Josh