Re: [PATCH v1 bpf-next 8/9] bpftool: Add bpftool_tests README.md

[Quentin Monnet <quentin@isovalent.com>]

2023-11-16 19:43 UTC+0000 ~ Manu Bretelle <chantr4@gmail.com>
> A README.md explaining how to run bpftool tests.
> 
> Signed-off-by: Manu Bretelle <chantr4@gmail.com>
> ---
>  .../selftests/bpf/bpftool_tests/README.md     | 91 +++++++++++++++++++
>  1 file changed, 91 insertions(+)
>  create mode 100644 tools/testing/selftests/bpf/bpftool_tests/README.md
> 
> diff --git a/tools/testing/selftests/bpf/bpftool_tests/README.md b/tools/testing/selftests/bpf/bpftool_tests/README.md
> new file mode 100644
> index 000000000000..8ee5d656f6f8
> --- /dev/null
> +++ b/tools/testing/selftests/bpf/bpftool_tests/README.md
> @@ -0,0 +1,91 @@
> +## About the testing Framework
> +
> +The testing framework uses [RUST's testing framework](https://doc.rust-lang.org/rustc/tests/index.html)
> +and [libbpf-rs](https://docs.rs/libbpf-rs/latest/libbpf_rs/).
> +
> +The former takes care of scheduling tests and reporting their successes/failures.
> +The latter is used to load bpf programs, maps, and possibly interact with them

nit: s/bpf/BPF/

> +programatically through libbpf API.

Typo (programmatically)

> +This allows us to set the environment we want to test and check that `bpftool`
> +does what we expect.
> +
> +This document assumes you have [`cargo` and `rust` installed](https://doc.rust-lang.org/cargo/getting-started/installation.html).
> +
> +## Testing bpftool
> +
> +This should be no different than typical [`cargo test`](https://doc.rust-lang.org/cargo/commands/cargo-test.html)
> +but there is a few subtleties to consider when running `bpftool` tests:

"there are"

> +
> +1. bpftool needs to run with root privileges for the most part. So the runner needs to run as root.
> +1. each tests load a program, possibly modify it, and check expectations. In order to be deterministic, tests need to run serially.

2. Each test loads..., modifies, checks

> +
> +### Environment variable
> +
> +A few environment variable can be used to control the behaviour of the tests:

s/variable/variables/
(s/behaviour/behavior/? Not sure of the guidance on US/British English
in docs.)

> +- `RUST_TEST_THREADS`: This should be set to 1 to run one test at a time and avoid tests to step onto each others.

"each other"

> +- `BPFTOOL_PATH`: Allow passing an alternate location for `bpftool`. Default: `/usr/sbin/bpftool`
> +
> +### Running the test suite
> +
> +Here are a few options to make this happen:
> +
> +```
> +# build the test binary, extract the test executable location
> +# and run it with sudo, 1 test at a time.
> +eval sudo BPFTOOL_PATH=$(pwd)/../bpftool RUST_TEST_THREADS=1 \
> +    $(cargo test --no-run \
> +        --message-format=json | jq '. | select(.executable != null ).executable' \
> +    )
> +```
> +
> +or alternatively, one can use the [`CARGO_TARGET_<triple>_RUNNER` environment variable](https://doc.rust-lang.org/cargo/reference/environment-variables.html#:~:text=CARGO_TARGET_%3Ctriple%3E_RUNNER).
> +
> +The benefit of that approach is that compilation errors will show directly in the terminal.
> +
> +```
> +CARGO_TARGET_X86_64_UNKNOWN_LINUX_GNU_RUNNER="sudo -E" \
> +    BPFTOOL_PATH=$(pwd)/../bpftool \
> +    RUST_TEST_THREADS=1 \
> +    cargo test
> +```
> +
> +### Running tests against built kernel/bpftool
> +
> +Using [vmtest](https://github.com/danobi/vmtest):
> +
> +```
> +$ KERNEL_REPO=~/devel/bpf-next/
> +$ vmtest -k $KERNEL_REPO/arch/x86_64/boot/bzImage "BPFTOOL_PATH=$KERNEL_REPO/tools/bpf/bpftool/bpftool RUST_TEST_THREADS=1 cargo test"
> +=> bzImage
> +===> Booting
> +===> Setting up VM
> +===> Running command
> +    Finished test [unoptimized + debuginfo] target(s) in 2.06s
> +     Running unittests src/main.rs (target/debug/deps/bpftool_tests-afa5a7eef3cdeafb)
> +
> +running 11 tests
> +test bpftool_tests::run_bpftool ... ok
> +test bpftool_tests::run_bpftool_map_dump_id ... ok
> +test bpftool_tests::run_bpftool_map_list ... ok
> +test bpftool_tests::run_bpftool_map_pids ... ok
> +test bpftool_tests::run_bpftool_prog_list ... ok
> +test bpftool_tests::run_bpftool_prog_pids ... ok
> +test bpftool_tests::run_bpftool_prog_show_id ... ok
> +test bpftool_tests::run_bpftool_struct_ops_can_unregister_id ... ok
> +test bpftool_tests::run_bpftool_struct_ops_can_unregister_name ... ok
> +test bpftool_tests::run_bpftool_struct_ops_dump_name ... ok
> +test bpftool_tests::run_bpftool_struct_ops_list ... ok
> +
> +test result: ok. 11 passed; 0 failed; 0 ignored; 0 measured; 0 filtered out; finished in 1.88s
> +```
> +
> +the return code will be 0 on success, non-zero otherwise.
> +
> +
> +## Caveat
> +
> +Currently, libbpf-sys crate either uses a vendored libbpf, or the system one.
> +This could possibly limit tests against features that are being introduced.
> +
> +That being said, this is not a blocker now, and can be fixed upstream.
> +https://github.com/libbpf/libbpf-sys/issues/70 tracks this on libbpf-sys side.

Kernel docs usually use rST rather than Markdown for formatting.
Although there's rust/alloc/README.md, so I'm not entirely sure what's
best here. Unless there's a good reason for Markdown, we probably want
to stick to rST? This also depends on whether or not this patchset makes
it into the kernel repo, obviously.

Quentin