| Message ID | 20240318103534.701693-7-janusz.krzysztofik@linux.intel.com (mailing list archive) |
|---|---|
| Headers | show |
| Series | lib/kunit: Execute test cases synchronously | expand |
-----Original Message----- From: Janusz Krzysztofik <janusz.krzysztofik@linux.intel.com> Sent: Monday, March 18, 2024 3:13 AM To: igt-dev@lists.freedesktop.org Cc: intel-gfx@lists.freedesktop.org; intel-xe@lists.freedesktop.org; Kamil Konieczny <kamil.konieczny@linux.intel.com>; Mauro Carvalho Chehab <mchehab@kernel.org>; Cavitt, Jonathan <jonathan.cavitt@intel.com>; De Marchi, Lucas <lucas.demarchi@intel.com> Subject: [PATCH i-g-t v3 0/5] lib/kunit: Execute test cases synchronously > > Up to now we were loading a KUnit test module in test execution mode only > once per subtest, in background, and then, in parallel with execution of > test cases while the module was loading, we were looking through dmesg for > KTAP results from each expected test case. As a consequence, our IGT > messages were more or less delayed, never in full sync with kernel > messages. Moreover, parsing of KTAP results from already completed test > cases could be abandoned on a failure from loading the test module or > kernel taint caused by a subsequent test case. Also, parsing of KTAP > results from all subsequent test cases could be abandoned on a failure of > the parser caused by any test case. Other than that, if a user requested > a single dynamic sub-subtest, all test cases were executed anyway while > results from only one of them that corresponded to the selected dynamic > sub-subtest were reported. That way, kernel messages from unrelated test > cases, not only the selected one, could contribute to dmesg-fail or dmesg- > warn CI results from that sub-subtest. > > Since recent KUnit implementation is capable of executing only those test > cases that match a user filter, stop executing all of them asynchronously > and parsing their KTAP results as they appear. Instead, reload the test > module once per each dynamic sub-subtest with a filter that selects a > specific test case and wait for its completion. If successful and no > kernel taint has occurred then parse the whole KTAP report from a single > test case it has produced and translate it to IGT result of that single > corresponding sub-subtest. > > v3: Insert new patches 1-3 that fix an infinite loop when we try to get a > list of test cases from an unexpectedly missing KTAP report. > v2: Refresh the series on top of changes to KUnit filters handling, > - move the code of a new helper from a previous patch 1 to a previous > patch 2 which now becomes patch 1, > - actually limit the scope of the helper to fetching a KTAP report from > a file descriptor, and let the callers decide on how other steps, like > setting up filters or loading a test module, and errors they return > are handled, > - update commit description with a more detailed justification of why we > need these changes, > - rebase the former patch 1 on top of the new patch 1, update its commit > message and description and provide it as patch 2. > > Janusz Krzysztofik (5): > lib/kunit: Store igt_ktap_results pointer in a central location > lib/kunit: Let igt_ktap_free() take care of pointer reset > lib/kunit: Time out promptly on missing KTAP report > lib/kunit: Execute test cases synchronously > lib/kunit: Minimize code duplication > Acked-by: Jonathan Cavitt <jonathan.cavitt@intel.com> Ack applies to all patches in series. It seems that Kamil is giving a proper review to all the individual patches, so I'd wait until that's done before moving forward. -Jonathan Cavitt > lib/igt_kmod.c | 193 ++++++++++++++---------------------- > lib/igt_ktap.c | 5 +- > lib/igt_ktap.h | 2 +- > lib/tests/igt_ktap_parser.c | 24 ++--- > 4 files changed, 93 insertions(+), 131 deletions(-) > > -- > 2.43.0 > >
Up to now we were loading a KUnit test module in test execution mode only once per subtest, in background, and then, in parallel with execution of test cases while the module was loading, we were looking through dmesg for KTAP results from each expected test case. As a consequence, our IGT messages were more or less delayed, never in full sync with kernel messages. Moreover, parsing of KTAP results from already completed test cases could be abandoned on a failure from loading the test module or kernel taint caused by a subsequent test case. Also, parsing of KTAP results from all subsequent test cases could be abandoned on a failure of the parser caused by any test case. Other than that, if a user requested a single dynamic sub-subtest, all test cases were executed anyway while results from only one of them that corresponded to the selected dynamic sub-subtest were reported. That way, kernel messages from unrelated test cases, not only the selected one, could contribute to dmesg-fail or dmesg- warn CI results from that sub-subtest. Since recent KUnit implementation is capable of executing only those test cases that match a user filter, stop executing all of them asynchronously and parsing their KTAP results as they appear. Instead, reload the test module once per each dynamic sub-subtest with a filter that selects a specific test case and wait for its completion. If successful and no kernel taint has occurred then parse the whole KTAP report from a single test case it has produced and translate it to IGT result of that single corresponding sub-subtest. v3: Insert new patches 1-3 that fix an infinite loop when we try to get a list of test cases from an unexpectedly missing KTAP report. v2: Refresh the series on top of changes to KUnit filters handling, - move the code of a new helper from a previous patch 1 to a previous patch 2 which now becomes patch 1, - actually limit the scope of the helper to fetching a KTAP report from a file descriptor, and let the callers decide on how other steps, like setting up filters or loading a test module, and errors they return are handled, - update commit description with a more detailed justification of why we need these changes, - rebase the former patch 1 on top of the new patch 1, update its commit message and description and provide it as patch 2. Janusz Krzysztofik (5): lib/kunit: Store igt_ktap_results pointer in a central location lib/kunit: Let igt_ktap_free() take care of pointer reset lib/kunit: Time out promptly on missing KTAP report lib/kunit: Execute test cases synchronously lib/kunit: Minimize code duplication lib/igt_kmod.c | 193 ++++++++++++++---------------------- lib/igt_ktap.c | 5 +- lib/igt_ktap.h | 2 +- lib/tests/igt_ktap_parser.c | 24 ++--- 4 files changed, 93 insertions(+), 131 deletions(-)