Re: [PATCH 2/2] kunit: add coverage_uml.config to enable GCOV on UML

From: Daniel Latypov
Date: Fri Jun 24 2022 - 12:13:30 EST


On Fri, Jun 24, 2022 at 7:12 AM Maíra Canal <maira.canal@xxxxxx> wrote:
>
> On 6/24/22 04:55, 'David Gow' via KUnit Development wrote:
> > On Fri, Jun 24, 2022 at 8:12 AM 'Daniel Latypov' via KUnit Development
> > <kunit-dev@xxxxxxxxxxxxxxxx> wrote:
> >>
> >> Now that kunit.py's --kunitconfig is repeatable, let's create a file to
> >> hold the various options needed to enable coverage under UML.
> >>
> >> This can be used like so:
> >> $ ./tools/testing/kunit/kunit.py run \
> >> --kunitconfig=tools/testing/kunit/configs/all_tests_uml.config \
> >> --kunitconfig=tools/testing/kunit/configs/coverage_uml.config \
> >> --make_options=CC=/usr/bin/gcc-6
> >>
> >> which on my system is enough to get coverage working [1].
> >>
>
> It's great to see this coming to KUnit! As I was testing this series, I
> wasn't able to generate the coverage stats with GCC11. I got a linking
> error from ld:
>
> ERROR:root:/usr/bin/ld:
> /usr/lib/gcc/x86_64-redhat-linux/11/libgcov.a(_gcov.o): in function
> `mangle_path':
> (.text+0x19f0): multiple definition of `mangle_path';
> fs/seq_file.o:/home/mairacanal/linux/.kunit/../fs/seq_file.c:441: first
> defined here
> collect2: error: ld returned 1 exit status
>
> By changing the name of the function to `seq_mangle_path`, it is
> possible to solve the linking error and run the tests. But, anyway, no
> .gcda files are generated.
>
> I checked out, and this is a known issue that was already discussed in
> this series [1] (but I guess it didn't move on).

There's the two main issues I knew about and that you noted.
* no gcda files produced (gcc 7+)
* seq_mangle_path (I don't remember, gcc 8 or 9+?)

The latter can be worked around locally fairly easily, as you also found.
The former is due to how the exit handlers work, iirc.

I didn't know about [1], but it also mentions the exit handlers issue.
However, from the sounds of [2], it seems like it might be a dead end for now :\
So I wouldn't hold onto hope of using modern GCC versions with this.

[2] https://lore.kernel.org/all/d36ea54d8c0a8dd706826ba844a6f27691f45d55.camel@xxxxxxxxxxxxxxxx/

Daniel