Context:
1. all_tests_uml.config used to be UML specific back when users to
manually specify CONFIG_VIRTIO_UML=y to enable CONFIG_PCI=y.
2. --alltests used allyesconfig along with a curated list of options to
disable. It's only ever worked for brief periods of time and has
perennially been broken due to compile issues.
Now all_tests_uml.config should work across ~all architectures.
Let's instead use this to implement --alltests.
Note: if anyone was using all_tests_uml.config, this change breaks them.
I think that's unlikely since it was added in 5.19 and was a lot to
type: --kunitconfig=tools/testing/kunit/configs/all_tests_uml.config.
We could make it a symlink to the new name, but I don't think the
caution is warranted here.
Signed-off-by: Daniel Latypov <dlatypov@google.com>
Reviewed-by: David Gow <davidgow@google.com>
Signed-off-by: Shuah Khan <skhan@linuxfoundation.org>
There are several tests which depend on PCI, and hence need a bunch of
extra options to run under UML. This makes it awkward to give
configuration instructions (whether in documentation, or as part of a
.kunitconfig file), as two separate, incompatible sets of config options
are required for UML and "most other architectures".
For non-UML architectures, it's possible to add default kconfig options
via the qemu_config python files, but there's no equivalent for UML. Add
a new tools/testing/kunit/configs/arch_uml.config file containing extra
kconfig options to use on UML.
Tested-by: José Expósito <jose.exposito89@gmail.com>
Reviewed-by: Daniel Latypov <dlatypov@google.com>
Signed-off-by: David Gow <davidgow@google.com>
Reviewed-by: Brendan Higgins <brendanhiggins@google.com>
Signed-off-by: Daniel Latypov <dlatypov@google.com>
Signed-off-by: Shuah Khan <skhan@linuxfoundation.org>
It's come up a few times that it would be useful to have --kunitconfig
be repeatable [1][2].
This could be done before with a bit of shell-fu, e.g.
$ find fs/ -name '.kunitconfig' -exec cat {} + | \
./tools/testing/kunit/kunit.py run --kunitconfig=/dev/stdin
or equivalently:
$ cat fs/ext4/.kunitconfig fs/fat/.kunitconfig | \
./tools/testing/kunit/kunit.py run --kunitconfig=/dev/stdin
But this can be fairly clunky to use in practice.
And having explicit support in kunit.py opens the door to having more
config fragments of interest, e.g. options for PCI on UML [1], UML
coverage [2], variants of tests [3].
There's another argument to be made that users can just use multiple
--kconfig_add's, but this gets very clunky very fast (e.g. [2]).
Note: there's a big caveat here that some kconfig options might be
incompatible. We try to give a clearish error message in the simple case
where the same option appears multiple times with conflicting values,
but more subtle ones (e.g. mutually exclusive options) will be
potentially very confusing for the user. I don't know we can do better.
Note 2: if you want to combine a --kunitconfig with the default, you
either have to do to specify the current build_dir
> --kunitconfig=.kunit --kunitconfig=additional.config
or
> --kunitconfig=tools/testing/kunit/configs/default.config --kunitconifg=additional.config
each of which have their downsides (former depends on --build_dir,
doesn't work if you don't have a .kunitconfig yet), etc.
Example with conflicting values:
> $ ./tools/testing/kunit/kunit.py config --kunitconfig=lib/kunit --kunitconfig=/dev/stdin <<EOF
> CONFIG_KUNIT_TEST=n
> CONFIG_KUNIT=m
> EOF
> ...
> kunit_kernel.ConfigError: Multiple values specified for 2 options in kunitconfig:
> CONFIG_KUNIT=y
> vs from /dev/stdin
> CONFIG_KUNIT=m
>
> CONFIG_KUNIT_TEST=y
> vs from /dev/stdin
> # CONFIG_KUNIT_TEST is not set
[1] https://lists.freedesktop.org/archives/dri-devel/2022-June/357616.html
[2] https://lore.kernel.org/linux-kselftest/CAFd5g45f3X3xF2vz2BkTHRqOC4uW6GZxtUUMaP5mwwbK8uNVtA@mail.gmail.com/
[3] https://lore.kernel.org/linux-kselftest/CANpmjNOdSy6DuO6CYZ4UxhGxqhjzx4tn0sJMbRqo2xRFv9kX6Q@mail.gmail.com/
Signed-off-by: Daniel Latypov <dlatypov@google.com>
Reviewed-by: Brendan Higgins <brendanhiggins@google.com>
Signed-off-by: Shuah Khan <skhan@linuxfoundation.org>
Currently, you cannot ovewrwrite what's in your kunitconfig via
--kconfig_add.
Nor can you override something in a qemu_config via either means.
This patch makes it so we have this level of priority
* --kconfig_add
* kunitconfig file (the default or the one from --kunitconfig)
* qemu_config
The rationale for this order is that the more "dynamic" sources of
kconfig options should take priority.
--kconfig_add is obviously the most dynamic.
And for kunitconfig, users probably tweak the file manually or specify
--kunitconfig more often than they delve into qemu_config python files.
And internally, we convert the kconfigs from a python list into a set or
dict fairly often. We should just use a dict internally.
We exposed the set transform in the past since we didn't define __eq__,
so also take the chance to shore up the kunit_kconfig.Kconfig interface.
Example
=======
Let's consider the unrealistic example where someone would want to
disable CONFIG_KUNIT.
I.e. they run
$ ./tools/testing/kunit/kunit.py config --kconfig_add=CONFIG_KUNIT=n
Before
------
We'd write the following
> # CONFIG_KUNIT is not set
> CONFIG_KUNIT_ALL_TESTS=y
> CONFIG_KUNIT_TEST=y
> CONFIG_KUNIT=y
> CONFIG_KUNIT_EXAMPLE_TEST=y
And we'd error out with
> ERROR:root:Not all Kconfig options selected in kunitconfig were in the generated .config.
> This is probably due to unsatisfied dependencies.
> Missing: # CONFIG_KUNIT is not set
After
-----
We'd write the following
> # CONFIG_KUNIT is not set
> CONFIG_KUNIT_TEST=y
> CONFIG_KUNIT_ALL_TESTS=y
> CONFIG_KUNIT_EXAMPLE_TEST=y
And we'd error out with
> ERROR:root:Not all Kconfig options selected in kunitconfig were in the generated .config.
> This is probably due to unsatisfied dependencies.
> Missing: CONFIG_KUNIT_EXAMPLE_TEST=y, CONFIG_KUNIT_TEST=y, CONFIG_KUNIT_ALL_TESTS=y
Signed-off-by: Daniel Latypov <dlatypov@google.com>
Reviewed-by: David Gow <davidgow@google.com>
Reviewed-by: Brendan Higgins <brendanhiggins@google.com>
Signed-off-by: Shuah Khan <skhan@linuxfoundation.org>
Example usage:
$ ./tools/testing/kunit/kunit.py run --arch=x86_64 \
--kconfig_add=CONFIG_SMP=y --qemu_args='-smp 8'
Looking in the test.log, one can see
> smp: Bringing up secondary CPUs ...
> .... node #0, CPUs: #1#2#3#4#5#6#7
> smp: Brought up 1 node, 8 CPUs
This flag would allow people to make tweaks like this without having to
create custom qemu_config files.
For consistency with --kernel_args, we allow users to repeat this
argument, e.g. you can tack on a --qemu_args='-m 2048', or you could
just append it to the first string ('-smp 8 -m 2048').
Signed-off-by: Daniel Latypov <dlatypov@google.com>
Reviewed-by: David Gow <davidgow@google.com>
Reviewed-by: Brendan Higgins <brendanhiggins@google.com>
Signed-off-by: Shuah Khan <skhan@linuxfoundation.org>
Context:
* kunit_kernel.py is importing kunit_parser.py just to use the
print_with_timestamp() function
* the parser is directly printing to stdout, which will become an issue
if we ever try to run multiple kernels in parallel
This patch introduces a kunit_printer.py file and migrates callers of
kunit_parser.print_with_timestamp() to call
kunit_printer.stdout.print_with_timestamp() instead.
Future changes:
If we want to support showing results for parallel runs, we could then
create new Printer's that don't directly write to stdout and refactor
the code to pass around these Printer objects.
Signed-off-by: Daniel Latypov <dlatypov@google.com>
Reviewed-by: David Gow <davidgow@google.com>
Reviewed-by: Brendan Higgins <brendanhiggins@google.com>
Signed-off-by: Shuah Khan <skhan@linuxfoundation.org>
Our main function currently has an optional `linux` argument which is
used to by our unit tests to inject a mock.
We currently have the same code copy-pasted several times to do
if not linux:
linux = MakeRealInstance(cli_args.foo, cli_args.bar, ...)
But in python, dependency injection isn't necessary or idiomatic when we
can just use mock.patch() to mock things out.
This change
1. adds a helper to create a LinuxSourceTree from the cli_args
2. drops the `linux` parameter in favor of mocking the __init__ func.
Signed-off-by: Daniel Latypov <dlatypov@google.com>
Reviewed-by: David Gow <davidgow@google.com>
Reviewed-by: Brendan Higgins <brendanhiggins@google.com>
Signed-off-by: Shuah Khan <skhan@linuxfoundation.org>
This primarily comes from running pylint over kunit tool code and
ignoring some warnings we don't care about.
If we ever got a fully clean setup, we could add this to run_checks.py,
but we're not there yet.
Fix things like
* Drop unused imports
* check `is None`, not `== None` (see PEP 8)
* remove redundant parens around returns
* remove redundant `else` / convert `elif` to `if` where appropriate
* rename make_arch_qemuconfig() param to base_kunitconfig (this is the
name used in the subclass, and it's a better one)
* kunit_tool_test: check the exit code for SystemExit (could be 0)
Signed-off-by: Daniel Latypov <dlatypov@google.com>
Reviewed-by: David Gow <davidgow@google.com>
Reviewed-by: Brendan Higgins <brendanhiggins@google.com>
Signed-off-by: Shuah Khan <skhan@linuxfoundation.org>
This logic depends on the kernel logging a message containing
'kunit test case crashed', but there is no corresponding logic to do so.
This is likely a relic of the revision process KUnit initially went
through when being upstreamed.
Delete it given
1) it's been missing for years and likely won't get implemented
2) the parser has been moving to be a more general KTAP parser,
kunit-only magic like this isn't how we'd want to implement it.
Signed-off-by: Daniel Latypov <dlatypov@google.com>
Reviewed-by: David Gow <davidgow@google.com>
Reviewed-by: Brendan Higgins <brendanhiggins@google.com>
Signed-off-by: Shuah Khan <skhan@linuxfoundation.org>
Before:
$ ./tools/testing/kunit/kunit.py parse /dev/null
...
[ERROR] Test : invalid KTAP input!
After:
$ ./tools/testing/kunit/kunit.py parse /dev/null
...
[ERROR] Test <missing>: could not find any KTAP output!
This error message gets printed out when extract_tap_output() yielded no
lines. So while it could be because of malformed KTAP output from KUnit,
it could also be due to not having any KTAP output at all.
Try and make the error message here more clear.
Signed-off-by: Daniel Latypov <dlatypov@google.com>
Reviewed-by: David Gow <davidgow@google.com>
Reviewed-by: Brendan Higgins <brendanhiggins@google.com>
Signed-off-by: Shuah Khan <skhan@linuxfoundation.org>
Before, our help output contained lines like
--kconfig_add KCONFIG_ADD
--qemu_config qemu_config
--jobs jobs
They're not very helpful.
The former kind come from the automatic 'metavar' we get from argparse,
the uppercase version of the flag name.
The latter are where we manually specified metavar as the flag name.
After:
--build_dir DIR
--make_options X=Y
--kunitconfig PATH
--kconfig_add CONFIG_X=Y
--arch ARCH
--cross_compile PREFIX
--qemu_config FILE
--jobs N
--timeout SECONDS
--raw_output [{all,kunit}]
--json [FILE]
This patch tries to make the code more clear by specifying the _type_ of
input we expect, e.g. --build_dir is a DIR, --qemu_config is a FILE.
I also switched it to uppercase since it looked more clearly like
placeholder text that way.
This patch also changes --raw_output to specify `choices` to make it
more clear what the options are, and this way argparse can validate it
for us, as shown by the added test case.
Signed-off-by: Daniel Latypov <dlatypov@google.com>
Reviewed-by: David Gow <davidgow@google.com>
Reviewed-by: Brendan Higgins <brendanhiggins@google.com>
Signed-off-by: Shuah Khan <skhan@linuxfoundation.org>
When using --json, kunit.py run/exec/parse will produce results in
KernelCI json format.
As part of that, we include the build_dir that was used, and we
(incorrectly) hardcode in the arch, etc.
We'll want a way to plumb more values (as well as the correct `arch`),
so this patch groups those fields into kunit_json.Metadata type.
This patch should have no user visible changes.
And since we only used build_dir in KunitParseRequest for json, we can
now move it out of that struct and add it into KunitExecRequest, which
needs it and used to get it via inheritance.
Signed-off-by: Daniel Latypov <dlatypov@google.com>
Reviewed-by: David Gow <davidgow@google.com>
Reviewed-by: Brendan Higgins <brendanhiggins@google.com>
Signed-off-by: Shuah Khan <skhan@linuxfoundation.org>
--build_dir is set to a default of '.kunit' since commit ddbd60c779
("kunit: use --build_dir=.kunit as default"), but even before then it
was explicitly set to ''.
So outside of one unit test, there was no way for the build_dir to be
ever be None, and we can simplify code by fixing the unit test and
enforcing that via updated type annotations.
E.g. this lets us drop `get_file_path()` since it's now exactly
equivalent to os.path.join().
Note: there's some `if build_dir` checks that also fail if build_dir is
explicitly set to '' that just guard against passing "O=" to make.
But running `make O=` works just fine, so drop these checks.
Signed-off-by: Daniel Latypov <dlatypov@google.com>
Reviewed-by: David Gow <davidgow@google.com>
Reviewed-by: Brendan Higgins <brendanhiggins@google.com>
Signed-off-by: Shuah Khan <skhan@linuxfoundation.org>
Currently kunit_json.get_json_result() will output the JSON-ified test
output to json_path, but iff it's not "stdout".
Instead, move the responsibility entirely over to the one caller.
Signed-off-by: Daniel Latypov <dlatypov@google.com>
Reviewed-by: David Gow <davidgow@google.com>
Reviewed-by: Brendan Higgins <brendanhiggins@google.com>
Signed-off-by: Shuah Khan <skhan@linuxfoundation.org>
The --jobs parameter for kunit_tool currently defaults to 8 CPUs,
regardless of the number available. For systems with significantly more
(or less), this is not as efficient. Instead, default --jobs to the
number of CPUs available to the process: while there are as many
superstitions as to exactly what the ideal jobs:CPU ratio is, this seems
sufficiently sensible to me.
A new helper function to get the default number of jobs is added:
get_default_jobs() -- this is used in kunit_tool_test instead of a
hardcoded value, or an explicit call to len(os.sched_getaffinity()), so
should be more flexible if this needs to change in the future.
Signed-off-by: David Gow <davidgow@google.com>
Signed-off-by: Daniel Latypov <dlatypov@google.com>
Reviewed-by: Daniel Latypov <dlatypov@google.com>
Reviewed-by: Brendan Higgins <brendanhiggins@google.com>
Signed-off-by: Shuah Khan <skhan@linuxfoundation.org>
If I created a kunitconfig file that was incomplete, then
$ ./tools/testing/kunit/kunit.py build --kunitconfig=my_kunitconfig
would silently drop all the options with unmet dependencies!
This is because it doesn't do the config check that `kunit.py config`
does.
So if I want to safely build a kernel for testing, I have to do
$ ./tools/testing/kunit/kunit.py config <flags>
$ ./tools/testing/kunit/kunit.py build <flags, again>
It seems unlikely that any user of kunit.py would want the current
`build` semantics.
So make it effectively do `kunit.py config` + `kunit.py build`.
Signed-off-by: Daniel Latypov <dlatypov@google.com>
Reviewed-by: Brendan Higgins <brendanhiggins@google.com>
Signed-off-by: Shuah Khan <skhan@linuxfoundation.org>
The `log` field is unused, and the `status` field is accessible via
`test.status`.
So it's simpler to just return the main `Test` object directly.
And since we're no longer returning a namedtuple, which has no type
annotations, this hopefully means typecheckers are better equipped to
find any errors.
Signed-off-by: Daniel Latypov <dlatypov@google.com>
Reviewed-by: Brendan Higgins <brendanhiggins@google.com>
Signed-off-by: Shuah Khan <skhan@linuxfoundation.org>
namedtuple is a terse way of defining a collection of fields.
However, it does not allow us to annotate the type of these fields.
It also doesn't let us have any sort of inheritance between types.
Since commit df4b0807ca ("kunit: tool: Assert the version
requirement"), kunit.py has asserted that it's running on python >=3.7.
So in that case use a 3.7 feature, dataclasses, to replace these.
Changes in detail:
* Make KunitExecRequest contain all the fields needed for exec_tests
* Use inheritance to dedupe fields
* also allows us to e.g. pass a KUnitRequest in as a KUnitParseRequest
* this has changed around the order of some fields
* Use named arguments when constructing all request objects in kunit.py
* This is to prevent accidentally mixing up fields, etc.
Signed-off-by: Daniel Latypov <dlatypov@google.com>
Reviewed-by: Brendan Higgins <brendanhiggins@google.com>
Signed-off-by: Shuah Khan <skhan@linuxfoundation.org>
Problem: currently, if you remove something from your kunitconfig,
kunit.py will not regenerate the .config file.
The same thing happens if you did --kunitconfig_add=CONFIG_KASAN=y [1]
and then ran again without it. Your new run will still have KASAN.
The reason is that kunit.py won't regenerate the .config file if it's a
superset of the kunitconfig. This speeds it up a bit for iterating.
This patch adds an additional check that forces kunit.py to regenerate
the .config file if the current kunitconfig doesn't match the previous
one.
What this means:
* deleting entries from .kunitconfig works as one would expect
* dropping a --kunitconfig_add also triggers a rebuild
* you can still edit .config directly to turn on new options
We implement this by creating a `last_used_kunitconfig` file in the
build directory (so .kunit, by default) after we generate the .config.
When comparing the kconfigs, we compare python sets, so duplicates and
permutations don't trip us up.
The majority of this patch is adding unit tests for the existing logic
and for the new case where `last_used_kunitconfig` differs.
[1] https://lore.kernel.org/linux-kselftest/20211106013058.2621799-2-dlatypov@google.com/
Signed-off-by: Daniel Latypov <dlatypov@google.com>
Reviewed-by: David Gow <davidgow@google.com>
Reviewed-by: Brendan Higgins <brendanhiggins@google.com>
Signed-off-by: Shuah Khan <skhan@linuxfoundation.org>
E.g. run tests but with KASAN
$ ./tools/testing/kunit/kunit.py run --arch=x86_64 --kconfig_add=CONFIG_KASAN=y
This also works with --kunitconfig
$ ./tools/testing/kunit/kunit.py run --arch=x86_64 --kunitconfig=fs/ext4 --kconfig_add=CONFIG_KASAN=y
This flag is inspired by TuxMake's --kconfig-add, see
https://gitlab.com/Linaro/tuxmake#examples.
Our version just uses "_" as the delimiter for consistency with
pre-existing flags like --build_dir, --make_options, --kernel_args, etc.
Note: this does make it easier to run into a pre-existing edge case:
$ ./tools/testing/kunit/kunit.py run --arch=x86_64 --kconfig_add=CONFIG_KASAN=y
$ ./tools/testing/kunit/kunit.py run --arch=x86_64
This second invocation ^ still has KASAN enabled!
kunit.py won't call olddefconfig if our current .config is already a
superset of the provided kunitconfig.
Signed-off-by: Daniel Latypov <dlatypov@google.com>
Reviewed-by: David Gow <davidgow@google.com>
Reviewed-by: Brendan Higgins <brendanhiggins@google.com>
Signed-off-by: Shuah Khan <skhan@linuxfoundation.org>
read_from_file() clears its `self` Kconfig object and parses a config
file.
It is a way to construct Kconfig objects more so than an operation on
Kconfig objects. This is reflected in the fact its only ever used as:
kconfig = kunit_config.Kconfig()
kconfig.read_from_file(path)
So clean this up and simplify callers by replacing it with
kconfig = kunit_config.parse_file(path)
Do the same thing for the related parse_from_string() function as well.
Signed-off-by: Daniel Latypov <dlatypov@google.com>
Reviewed-by: David Gow <davidgow@google.com>
Reviewed-by: Brendan Higgins <brendanhiggins@google.com>
Signed-off-by: Shuah Khan <skhan@linuxfoundation.org>
With the parser rework [1] and run_kernel() rework [2], this allows the
parser to print out test results incrementally.
Currently, that's held up by the fact that the LineStream eagerly
pre-fetches the next line when you call pop().
This blocks parse_test_result() from returning until the line *after*
the "ok 1 - test name" line is also printed.
One can see this with the following example:
$ (echo -e 'TAP version 14\n1..3\nok 1 - fake test'; sleep 2; echo -e 'ok 2 - fake test 2'; sleep 3; echo -e 'ok 3 - fake test 3') | ./tools/testing/kunit/kunit.py parse
Before this patch [1]: there's a pause before 'fake test' is printed.
After this patch: 'fake test' is printed out immediately.
This patch also adds
* a unit test to verify LineStream's behavior directly
* a test case to ensure that it's lazily calling the generator
* an explicit exception for when users go beyond EOF
[1] https://lore.kernel.org/linux-kselftest/20211006170049.106852-1-dlatypov@google.com/
[2] https://lore.kernel.org/linux-kselftest/20211005011340.2826268-1-dlatypov@google.com/
Signed-off-by: Daniel Latypov <dlatypov@google.com>
Reviewed-by: David Gow <davidgow@google.com>
Reviewed-by: Brendan Higgins <brendanhiggins@google.com>
Signed-off-by: Shuah Khan <skhan@linuxfoundation.org>
It's possible for a test to have a subtest header, but zero valid
subtests. We used to error on this if the test plan had no subtests
listed, but it's possible to have subtests without a test plan (indeed,
this is how parameterised tests work).
Tests with 0 subtests now have the result NO_TESTS, and will report an
error (which does not halt test execution, but is printed in a scary red
colour and is noted in the results summary).
Signed-off-by: David Gow <davidgow@google.com>
Reviewed-by: Daniel Latypov <dlatypov@google.com>
Reviewed-by: Brendan Higgins <brendanhiggins@google.com>
Signed-off-by: Shuah Khan <skhan@linuxfoundation.org>
The (K)TAP spec encourages test output to begin with a 'test plan': a
count of the number of tests being run of the form:
1..n
However, some test suites might not know the number of subtests in
advance (for example, KUnit's parameterised tests use a generator
function). In this case, it's not possible to print the test plan in
advance.
kunit_tool already parses test output which doesn't contain a plan, but
reports an error. Since we want to use nested subtests with KUnit
paramterised tests, remove this error.
Signed-off-by: David Gow <davidgow@google.com>
Reviewed-by: Daniel Latypov <dlatypov@google.com>
Reviewed-by: Brendan Higgins <brendanhiggins@google.com>
Signed-off-by: Shuah Khan <skhan@linuxfoundation.org>
Daniel Latypov
David Gow