The commit 06a0ba866d90 ("common/rc: avoid module load in _have_driver()")
removed module load from _have_driver(). However, it turned out that it was not
a good solution. In the recent discussion for module load preparation for nvme
test cases, it was pointed out no module load in _have_driver() is confusing and
adding complexity [1]:
- Without module load in _have_driver(), explicit module load and unload are
required in number of test cases. Boiler plates.
- The module unload is not always safe. Need care if the module unload is
expected or not.
- The module load needs error handling.
I can think of a new helper function to address the comments above, but it will
look like _have_driver() with module load. Hence, I suggest to revert back some
part of the the commit 06a0ba866d90 to load modules in _have_driver() (Sorry
Christoph, Bart for doing this on the commit you reviewed). As a better
solution, I propose to record the modules loaded in _have_driver() and unload
them at each test case end, regardless of the test case is skipped or executed.
I confirmed this fix avoids the issue that the commit 06a0ba866d90 tried to fix.
In this series, 4th patch is the core change in _have_driver. 1st, 2nd and 3rd
patches are clean-up preparation patches for the 4th patch. 5th patch reverts
changes in nbd/rc, which is no longer required after the 4th patch.
[1] https://lore.kernel.org/linux-block/89aedf1d-ae08-adef-db29-17e5bf85d054@xxxxxxxxxxx/
Shin'ichiro Kawasaki (5):
check: clean up _run_test()
common,tests: rename unload_module() to _unload_module()
check,common/rc: move _unload_module() from common/rc to check
check,common/rc: load module in _have_driver() and unload after test
Revert "nbd/rc: load nbd module explicitly"
check | 36 +++++++++++++++++++++++++++++++-----
common/multipath-over-rdma | 4 ++--
common/rc | 26 ++++++++++----------------
tests/nbd/rc | 12 ++----------
tests/nvmeof-mp/rc | 12 ++++++------
tests/srp/rc | 8 ++++----
6 files changed, 55 insertions(+), 43 deletions(-)
--
2.37.1
