kernel/linux.git/tools/testing/ktest, branch master

ktest: Store failure logs also in fatal paths

2026-03-18T20:00:54+00:00

STORE_FAILURES was only saved from fail(), so paths that reached dodie() could exit without preserving failure logs. That includes fatal hook paths such as: POST_BUILD_DIE = 1 and ordinary failures when: DIE_ON_FAILURE = 1 Call save_logs("fail", ...) from dodie() too so fatal failures keep the same STORE_FAILURES artifacts as non-fatal fail() paths. Cc: John Hawley Link: https://patch.msgid.link/20260318-ktest-fixes-v1-1-9dd94d46d84c@suse.com Signed-off-by: Ricardo B. Marlière Signed-off-by: Steven Rostedt

ktest: Add a --dry-run mode

2026-03-09T15:03:00+00:00

When working on a ktest configuration, it is often useful to inspect the final option values after includes, defaults, per-test overrides, and variable expansion have been applied, without actually starting a test run. Add a --dry-run option that reads the configuration, prints the test preamble using resolved option values, and exits before opening LOG_FILE or executing any test logic. This is useful for debugging ktest configurations and for scripts that need to validate the final resolved settings without triggering side effects. Cc: John Hawley Cc: Andrea Righi Cc: Marcos Paulo de Souza Cc: Matthieu Baerts Cc: Fernando Fernandez Mancera Cc: Pedro Falcato Link: https://patch.msgid.link/20260307-ktest-fixes-v1-9-565d412f4925@suse.com Signed-off-by: Ricardo B. Marlière Signed-off-by: Steven Rostedt

ktest: Run POST_KTEST hooks on failure and cancellation

2026-03-09T15:02:18+00:00

PRE_KTEST can be useful for setting up the environment and POST_KTEST to tear it down, however POST_KTEST only runs on the normal end-of-run path. It is skipped when ktest exits through dodie() or cancel_test(). Final cleanup hooks are skipped. Factor the final hook execution into run_post_ktest(), call it from the normal exit path and from the early exit paths, and guard it so the hook runs at most once. Cc: John Hawley Cc: Andrea Righi Cc: Marcos Paulo de Souza Cc: Matthieu Baerts Cc: Fernando Fernandez Mancera Cc: Pedro Falcato Link: https://patch.msgid.link/20260307-ktest-fixes-v1-8-565d412f4925@suse.com Fixes: 921ed4c7208e ("ktest: Add PRE/POST_KTEST and TEST options") Signed-off-by: Ricardo B. Marlière Signed-off-by: Steven Rostedt

ktest: Add PRE_KTEST_DIE for PRE_KTEST failures

2026-03-09T14:32:08+00:00

PRE_KTEST runs before the first test, but its return status is currently ignored. A failing setup hook can leave the rest of the run executing in a partially initialized environment. Add PRE_KTEST_DIE so PRE_KTEST can fail the run in the same way PRE_BUILD_DIE and PRE_TEST_DIE already can. Keep the default behavior unchanged when the new option is not set. Cc: John Hawley Cc: Andrea Righi Cc: Marcos Paulo de Souza Cc: Matthieu Baerts Cc: Fernando Fernandez Mancera Cc: Pedro Falcato Link: https://patch.msgid.link/20260307-ktest-fixes-v1-7-565d412f4925@suse.com Signed-off-by: Ricardo B. Marlière Signed-off-by: Steven Rostedt

ktest: Stop dropping console output during power-cycle reboot

2026-03-09T14:32:07+00:00

The POWER_CYCLE fallback added to reboot() flushes monitor output at the wrong time. In the untimed reboot path, flushing immediately after start_monitor() can consume the first output from the new boot before monitor() begins reading it. In the timed path, flushing after POWER_CYCLE can eat the "Linux version" banner or REBOOT_SUCCESS_LINE from the new kernel. That makes ktest miss the boot it is waiting for and can trigger an unnecessary second power cycle. Start the monitor before POWER_CYCLE so the reference counting stays balanced, but only flush when reboot() was asked to wait for a timed reboot. Perform that flush before issuing POWER_CYCLE so it drains stale output from the old kernel instead of consuming the next boot. Cc: John Hawley Cc: Andrea Righi Cc: Marcos Paulo de Souza Cc: Matthieu Baerts Cc: Fernando Fernandez Mancera Cc: Pedro Falcato Link: https://patch.msgid.link/20260307-ktest-fixes-v1-6-565d412f4925@suse.com Signed-off-by: Ricardo B. Marlière Signed-off-by: Steven Rostedt

ktest: Run commands through list-form shell open

2026-03-09T14:32:07+00:00

run_command() currently uses string-form open(): open(CMD, "$command 2>&1 |") That delegates parsing to the shell but also mixes the stderr redirection into the command string. Switch to list-form open() with an explicit sh -c wrapper so shell syntax errors are captured in the same output stream as command output. Otherwise, important errors can not be retrieved from the ktest LOG_FILE. Cc: John Hawley Cc: Andrea Righi Cc: Marcos Paulo de Souza Cc: Matthieu Baerts Cc: Fernando Fernandez Mancera Cc: Pedro Falcato Link: https://patch.msgid.link/20260307-ktest-fixes-v1-5-565d412f4925@suse.com Signed-off-by: Ricardo B. Marlière Signed-off-by: Steven Rostedt

ktest: Honor empty per-test option overrides

2026-03-09T14:32:07+00:00

A per-test override can clear an inherited default option by assigning an empty value, but __set_test_option() still used option_defined() to decide whether a per-test key existed. That turned an empty per-test assignment back into "fall back to the default", so tests still could not clear inherited settings. For example: DEFAULTS (...) LOG_FILE = /tmp/ktest-empty-override.log CLEAR_LOG = 1 ADD_CONFIG = /tmp/.config TEST_START TEST_TYPE = build BUILD_TYPE = nobuild ADD_CONFIG = This would run the test with ADD_CONFIG[1] = /tmp/.config Fix by checking whether the per-test key exists before falling back. If it does exist but is empty, treat it as unset for that test and stop the fallback chain there. Cc: John Hawley Cc: Andrea Righi Cc: Marcos Paulo de Souza Cc: Matthieu Baerts Cc: Fernando Fernandez Mancera Cc: Pedro Falcato Link: https://patch.msgid.link/20260307-ktest-fixes-v1-4-565d412f4925@suse.com Fixes: 22c37a9ac49d ("ktest: Allow tests to undefine default options") Signed-off-by: Ricardo B. Marlière Signed-off-by: Steven Rostedt

ktest: Treat undefined self-reference as empty

2026-03-09T14:32:07+00:00

Config variables are expanded when they are assigned. A first-time append such as: VAR := ${VAR} foo leaves the literal ${VAR} in the stored value because VAR has not been defined yet. Later expansions then carry the self-reference forward instead of behaving like an empty prefix. Drop an unescaped self-reference when the variable has no current value, and trim the outer whitespace left behind. Keep escaped \${VAR} references unchanged so literal text still works. Cc: John Hawley Cc: Andrea Righi Cc: Marcos Paulo de Souza Cc: Matthieu Baerts Cc: Fernando Fernandez Mancera Cc: Pedro Falcato Link: https://patch.msgid.link/20260307-ktest-fixes-v1-3-565d412f4925@suse.com Signed-off-by: Ricardo B. Marlière Signed-off-by: Steven Rostedt

ktest: Resolve LOG_FILE in test option context

2026-03-09T14:32:07+00:00

LOG_FILE is expanded immediately after the config file is parsed with eval_option(..., -1). That uses the default context, not the same option resolution path used for tests. If LOG_FILE depends on options that are finalized per test, it can be resolved from stale values before the first test starts. Resolve LOG_FILE through set_test_option("LOG_FILE", 1) instead so it uses the same expansion rules as the rest of the test options. Cc: John Hawley Cc: Andrea Righi Cc: Marcos Paulo de Souza Cc: Matthieu Baerts Cc: Fernando Fernandez Mancera Cc: Pedro Falcato Link: https://patch.msgid.link/20260307-ktest-fixes-v1-2-565d412f4925@suse.com Signed-off-by: Ricardo B. Marlière Signed-off-by: Steven Rostedt

ktest: Avoid undef warning when WARNINGS_FILE is unset

2026-03-09T14:32:06+00:00

check_buildlog() probes $warnings_file with -f even when WARNINGS_FILE is not configured. Perl warns about the uninitialized value and adds noise to the test log, which can hide the output we actually care about. Check that WARNINGS_FILE is defined before testing whether the file exists. Cc: John Hawley Cc: Andrea Righi Cc: Marcos Paulo de Souza Cc: Matthieu Baerts Cc: Fernando Fernandez Mancera Cc: Pedro Falcato Link: https://patch.msgid.link/20260307-ktest-fixes-v1-1-565d412f4925@suse.com Fixes: 4283b169abfb ("ktest: Add make_warnings_file and process full warnings") Signed-off-by: Ricardo B. Marlière Signed-off-by: Steven Rostedt