Re: selftests: hid: trouble building with clang due to missing header

From: Eduard Zingerman
Date: Fri Aug 25 2023 - 09:02:03 EST


On Fri, 2023-08-25 at 10:08 +0200, Benjamin Tissoires wrote:
>
> On Tue, Aug 22, 2023 at 11:42 PM Justin Stitt <justinstitt@xxxxxxxxxx> wrote:
> > > > > > Which kernel are you trying to test?
> > > > > > I tested your 2 commands on v6.5-rc7 and it just works.
> > > > >
> > > > > I'm also on v6.5-rc7 (706a741595047797872e669b3101429ab8d378ef)
> > > > >
> > > > > I ran these exact commands:
> > > > > > $ make mrproper
> > > > > > $ make LLVM=1 ARCH=x86_64 headers
> > > > > > $ make LLVM=1 ARCH=x86_64 -j128 -C tools/testing/selftests
> > > > > TARGETS=hid &> out
> > > > >
> > > > > and here's the contents of `out` (still warnings/errors):
> > > > > https://gist.github.com/JustinStitt/d0c30180a2a2e046c32d5f0ce5f59c6d
> > > > >
> > > > > I have a feeling I'm doing something fundamentally incorrectly. Any ideas?
> > > >
> > > > Sigh... there is a high chance my Makefile is not correct and uses the
> > > > installed headers (I was running the exact same commands, but on a
> > > > v6.4-rc7+ kernel).
> > > >
> > > > But sorry, it will have to wait for tomorrow if you want me to have a
> > > > look at it. It's 11:35 PM here, and I need to go to bed
> > > Take it easy. Thanks for the prompt responses here! I'd like to get
> > > the entire kselftest make target building with Clang so that we can
> > > close [1].
>
> Sorry I got urgent matters to tackle yesterday.
>
> It took me a while to understand what was going on, and I finally found
> it.
>
> struct hid_bpf_ctx is internal to the kernel, and so is declared in
> vmlinux.h, that we generate through BTF. But to generate the vmlinux.h
> with the correct symbols, these need to be present in the running
> kernel.
> And that's where we had a fundamental difference: I was running my
> compilations on a kernel v6.3+ (6.4.11) with that symbol available, and
> you are probably not.
>
> The bpf folks are using a clever trick to force the compilation[2]. And
> I think the following patch would work for you:

Hi Benjamin, Justin,

You might want to take a look at these two links:
[1] https://nakryiko.com/posts/bpf-core-reference-guide/#handling-incompatible-field-and-type-changes
[2] https://facebookmicrosites.github.io/bpf/blog/2020/02/19/bpf-portability-and-co-re.html#dealing-with-kernel-version-and-configuration-differences

The short version is: CO-RE relocation handling logic in libbpf
ignores suffixes of form '___something' for type and field names.

So, the following should accomplish the same as the trick with
#define/#undef:

#include "vmlinux.h"
...
struct hid_bpf_ctx___local {
__u32 index;
const struct hid_device *hid;
__u32 allocated_size;
enum hid_report_type report_type;
union {
__s32 retval;
__s32 size;
};

};
...
extern __u8 *hid_bpf_get_data(struct hid_bpf_ctx___local *ctx,
unsigned int offset, ...)

However, if the kernel does not have `hid_bpf_ctx` definition would
the test `progs/hid.c` still make sense?

When I tried to build hid tests locally I run into similar errors:

...
CLNG-BPF hid.bpf.o
In file included from progs/hid.c:6:
progs/hid_bpf_helpers.h:9:38: error: declaration of 'struct hid_bpf_ctx' \
will not be visible outside of this function [-Werror,-Wvisibility]
extern __u8 *hid_bpf_get_data(struct hid_bpf_ctx *ctx,
...

And there is indeed no `hid_bpf_ctx` in my vmlinux.h.
However, after enabling CONFIG_HID_BPF in kernel config the
`hid_bpf_ctx` appears in vmlinux.h, and I can compile HID selftests
w/o issues.

>
> ---
> From bb9eccb7a896ba4b3a35ed12a248e6d6cfed2df6 Mon Sep 17 00:00:00 2001
> From: Benjamin Tissoires <bentiss@xxxxxxxxxx>
> Date: Fri, 25 Aug 2023 10:02:32 +0200
> Subject: [PATCH] selftests/hid: ensure we can compile the tests on kernels
> pre-6.3
>
> For the hid-bpf tests to compile, we need to have the definition of
> struct hid_bpf_ctx. This definition is an internal one from the kernel
> and it is supposed to be defined in the generated vmlinux.h.
>
> This vmlinux.h header is generated based on the currently running kernel
> or if the kernel was already compiled in the tree. If you just compile
> the selftests without compiling the kernel beforehand and you are running
> on a 6.2 kernel, you'll end up with a vmlinux.h without the hid_bpf_ctx
> definition.
>
> Use the clever trick from tools/testing/selftests/bpf/progs/bpf_iter.h
> to force the definition of that symbol in case we don't find it in the
> BTF.
>
> Signed-off-by: Benjamin Tissoires <bentiss@xxxxxxxxxx>
> ---
> tools/testing/selftests/hid/progs/hid.c | 3 ---
> .../selftests/hid/progs/hid_bpf_helpers.h | 20 +++++++++++++++++++
> 2 files changed, 20 insertions(+), 3 deletions(-)
>
> diff --git a/tools/testing/selftests/hid/progs/hid.c b/tools/testing/selftests/hid/progs/hid.c
> index 88c593f753b5..1e558826b809 100644
> --- a/tools/testing/selftests/hid/progs/hid.c
> +++ b/tools/testing/selftests/hid/progs/hid.c
> @@ -1,8 +1,5 @@
> // SPDX-License-Identifier: GPL-2.0
> /* Copyright (c) 2022 Red hat */
> -#include "vmlinux.h"
> -#include <bpf/bpf_helpers.h>
> -#include <bpf/bpf_tracing.h>
> #include "hid_bpf_helpers.h"
>
> char _license[] SEC("license") = "GPL";
> diff --git a/tools/testing/selftests/hid/progs/hid_bpf_helpers.h b/tools/testing/selftests/hid/progs/hid_bpf_helpers.h
> index 4fff31dbe0e7..749097f8f4d9 100644
> --- a/tools/testing/selftests/hid/progs/hid_bpf_helpers.h
> +++ b/tools/testing/selftests/hid/progs/hid_bpf_helpers.h
> @@ -5,6 +5,26 @@
> #ifndef __HID_BPF_HELPERS_H
> #define __HID_BPF_HELPERS_H
>
> +/* "undefine" structs in vmlinux.h, because we "override" them below */
> +#define hid_bpf_ctx hid_bpf_ctx___not_used
> +#include "vmlinux.h"
> +#undef hid_bpf_ctx
> +
> +#include <bpf/bpf_helpers.h>
> +#include <bpf/bpf_tracing.h>
> +
> +
> +struct hid_bpf_ctx {
> + __u32 index;
> + const struct hid_device *hid;
> + __u32 allocated_size;
> + enum hid_report_type report_type;
> + union {
> + __s32 retval;
> + __s32 size;
> + };
> +};
> +
> /* following are kfuncs exported by HID for HID-BPF */
> extern __u8 *hid_bpf_get_data(struct hid_bpf_ctx *ctx,
> unsigned int offset,