Re: [PATCH bpf-next v2 1/2] bpf: Helper script for running BPF presubmit tests

[Jiri Olsa <jolsa@redhat.com>]

On Mon, Jan 25, 2021 at 06:09:48PM -0800, Andrii Nakryiko wrote:

SNIP

> 
> 7. Is it just me, or when ./test_progs is run inside VM, it's output
> is somehow heavily buffered and delayed? I get no output for a while,
> and then a whole bunch of lines with already passed tests.  Curious if
> anyone else noticed that as well. When I run the same image locally
> and manually (not through your script), I don't have this issue.

I see the same thing, big delay on the test_prog start and then
chunk updates of its output

jirka

> 
> 8. I noticed that even if the command succeeds (e.g., ./test_progs in
> my case), the script exits with non-zero error code (32 in my case).
> That's suboptimal, because you can't use that script to detect test
> failures.
> 
> But again, it's the polish feedback, great work!
> 
> >  tools/testing/selftests/bpf/run_in_vm.sh | 353 +++++++++++++++++++++++
> >  1 file changed, 353 insertions(+)
> >  create mode 100755 tools/testing/selftests/bpf/run_in_vm.sh
> >
> > diff --git a/tools/testing/selftests/bpf/run_in_vm.sh b/tools/testing/selftests/bpf/run_in_vm.sh
> > new file mode 100755
> > index 000000000000..09bb9705acb3
> > --- /dev/null
> > +++ b/tools/testing/selftests/bpf/run_in_vm.sh
> > @@ -0,0 +1,353 @@
> > +#!/bin/bash
> > +# SPDX-License-Identifier: GPL-2.0
> > +
> > +set -u
> > +set -e
> > +
> > +QEMU_BINARY="${QEMU_BINARY:="qemu-system-x86_64"}"
> > +X86_BZIMAGE="arch/x86/boot/bzImage"
> 
> Might be worth it to mention that this only works with x86_64 (due to
> image restrictions at least, right?).
> 
> > +DEFAULT_COMMAND="./test_progs"
> > +MOUNT_DIR="mnt"
> > +ROOTFS_IMAGE="root.img"
> > +OUTPUT_DIR="$HOME/.bpf_selftests"
> > +KCONFIG_URL="https://raw.githubusercontent.com/libbpf/libbpf/master/travis-ci/vmtest/configs/latest.config"
> > +INDEX_URL="https://raw.githubusercontent.com/libbpf/libbpf/master/travis-ci/vmtest/configs/INDEX"
> > +NUM_COMPILE_JOBS="$(nproc)"
> > +
> > +usage()
> > +{
> > +       cat <<EOF
> > +Usage: $0 [-k] [-i] [-d <output_dir>] -- [<command>]
> > +
> > +<command> is the command you would normally run when you are in
> > +tools/testing/selftests/bpf. e.g:
> > +
> > +       $0 -- ./test_progs -t test_lsm
> > +
> > +If no command is specified, "${DEFAULT_COMMAND}" will be run by
> > +default.
> > +
> > +If you build your kernel using KBUILD_OUTPUT= or O= options, these
> > +can be passed as environment variables to the script:
> > +
> > +  O=<path_relative_to_cwd> $0 -- ./test_progs -t test_lsm
> 
> "relative_to_cwd" is a bit misleading, it could be an absolute path as
> well, I presume. So I'd just say "O=<kernel_build_path>" or something
> along those lines.
> 
> > +
> > +or
> > +
> > +  KBUILD_OUTPUT=<path_relative_to_cwd> $0 -- ./test_progs -t test_lsm
> > +
> 
> [...]
>