Build your kernel with the --gcov option to enable GCOV profiling from the kernel. This will also trigger the build to save the required *.gcno files needed to viewing the collected count data.
For example to build a Cuttlefish (CF) kernel with GCOV profiling enabled run:
$ bazel run --gcov //common-modules/virtual-device:virtual_device_x86_64_dist
‘tradefed.sh’ can be used to run a number of different types of tests. Adding the appropriate coverage flags to the tradefed call will trigger tradefed to take care of mounting debugfs, reseting the gcov counts prior to test run, and the collection of gcov data files from debugfs after test completion.
These coverage arguments are:
--coverage --coverage-toolchain GCOV_KERNEL --auto-collect GCOV_KERNEL_COVERAGE
The following is a full example call running just the kselftest_net_socket test in the selftests test suite that exists under the ‘bazel-bin/common/testcases’ directory. The artifact output has been redirected to ‘tf-logs’ for easier reference needed in the next step.
$ prebuilts/tradefed/filegroups/tradefed/tradefed.sh run commandAndExit \
template/local_min --template:map test=suite/test_mapping_suite \
--include-filter 'selftests kselftest_net_socket' --tests-dir=bazel-bin/common/testcases/ \
--primary-abi-only --log-file-path tf-logs \
--coverage --coverage-toolchain GCOV_KERNEL \
--auto-collect GCOV_KERNEL_COVERAGE
The previously mentioned tradefed run will produce a tar file artifact in the tradefed log folder with a name similar to kernel_coverage*.tar.gz. This tar file is an archive of all the gcov data files collected into debugfs/ from the profiled device. In order to make it easier to work with this data, it needs to be converted to a single lcov tracefile.
The script ‘create-tracefile.py’ facilitates this generation by handling the required unpacking, file path corrections and ultimate ‘lcov’ call.
An example where we generate a tracefile only including results from net/socket.c. (If no source files are specified as included, then all source file data is used):
$ ./kernel/tests/tools/create-tracefile.py -t tf-logs/ --include net/socket.c
This will create a local tracefile named ‘cov.info’.
With the created tracefile there a number of different ways to view coverage data from it. Check out ‘man lcov’ for more options.
$ lcov --summary --rc lcov_branch_coverage=1 cov.info Reading tracefile cov.info_fix Summary coverage rate: lines......: 6.0% (81646 of 1370811 lines) functions..: 9.6% (10285 of 107304 functions) branches...: 3.7% (28639 of 765538 branches)
$ lcov --list --rc lcov_branch_coverage=1 cov.info
Reading tracefile cov.info_fix
|Lines |Functions|Branches
Filename |Rate Num|Rate Num|Rate Num
================================================================================
[/usr/local/google/home/joefradley/dev/common-android-mainline-2/common/]
arch/x86/crypto/aesni-intel_glue.c |23.9% 623|22.2% 36|15.0% 240
arch/x86/crypto/blake2s-glue.c |50.0% 28|50.0% 2|16.7% 30
arch/x86/crypto/chacha_glue.c | 0.0% 157| 0.0% 10| 0.0% 80
<truncated>
virt/lib/irqbypass.c | 0.0% 137| 0.0% 6| 0.0% 88
================================================================================
Total:| 6.0% 1369k| 9.6% 0M| 3.7% 764k
The lcov tool genhtml is used to generate html. To create html with the default settings:
$ genhtml --branch-coverage -o html cov.info
The page can be viewed at html\index.html.
Options of interest:
--frame: Creates a left hand macro view in a source file view.--missed: Helpful if you want to sort by what source is missing the most as opposed to the default coverage percentages.