From: Daniel Müller [mailto:deso@posteo.net]
Sent: Friday, August 12, 2022 1:52 AM
On Thu, Aug 11, 2022 at 12:02:57PM +0000, Roberto Sassu wrote:

quoted

quoted

From: Roberto Sassu [mailto:roberto.sassu@huawei.com]
Sent: Thursday, August 11, 2022 9:47 AM

quoted

From: Alexei Starovoitov [mailto:alexei.starovoitov@gmail.com]
Sent: Wednesday, August 10, 2022 11:34 PM
On Wed, Aug 10, 2022 at 06:59:28PM +0200, Roberto Sassu wrote:

quoted

+
+static int __init bpf_key_sig_kfuncs_init(void)
+{
+	int ret;
+
+	ret = register_btf_kfunc_id_set(BPF_PROG_TYPE_TRACING,
+					&bpf_key_sig_kfunc_set);
+	if (!ret)
+		return 0;
+
+	return register_btf_kfunc_id_set(BPF_PROG_TYPE_LSM,
+					 &bpf_key_sig_kfunc_set);

Isn't this a watery water ?
Don't you have a patch 1 ?
What am I missing ?

Uhm, yes. I had doubts too. That was what also KP did.

It makes sense to register once, since we mapped LSM to
TRACING.

Will resend only this patch. And I will figure out why CI failed.

Adding in CC Daniel Müller, which worked on this.

I think the issue is that some kernel options are set to =m.
This causes the CI to miss all kernel modules, since they are
not copied to the virtual machine that executes the tests.

I'm testing this patch:

https://github.com/robertosassu/libbpf-

ci/commit/b665e001b58c4ddb792a2a68098ea5dc6936b15c

I commented on the pull request. Would it make sense to adjust the
kernel configuration in this repository instead? I am worried that
otherwise everybody may need a similar work around, depending on how
selftests are ultimately run.

The issue seems specific of the eBPF CI. Others might be able to use
kernel modules.

Either choice is fine for me.

Roberto