On Thu, Dec 16, 2021 at 12:22 PM Shuah Khan skhan@linuxfoundation.org wrote:
On 12/16/21 1:03 PM, Andrii Nakryiko wrote:
On Thu, Dec 16, 2021 at 11:51 AM Shuah Khan skhan@linuxfoundation.org wrote:
On 12/16/21 12:30 PM, Andrii Nakryiko wrote:
On Thu, Dec 16, 2021 at 6:42 AM Shuah Khan skhan@linuxfoundation.org wrote:
On 12/15/21 9:04 PM, Andrii Nakryiko wrote:
On Tue, Dec 14, 2021 at 12:27 PM Shuah Khan skhan@linuxfoundation.org wrote: > > On 12/11/21 6:53 PM, Alexei Starovoitov wrote: >> On Fri, Dec 10, 2021 at 9:34 AM Shuah Khan skhan@linuxfoundation.org wrote: >>> >>> ARRAY_SIZE is defined in multiple test files. Remove the definitions >>> and include header file for the define instead. >>> >>> Remove ARRAY_SIZE define and add include bpf_util.h to bring in the >>> define. >>> >>> Signed-off-by: Shuah Khan skhan@linuxfoundation.org >>> --- >>> tools/testing/selftests/bpf/progs/netif_receive_skb.c | 5 +---- >>> tools/testing/selftests/bpf/progs/profiler.inc.h | 5 +---- >>> tools/testing/selftests/bpf/progs/test_sysctl_loop1.c | 5 +---- >>> tools/testing/selftests/bpf/progs/test_sysctl_loop2.c | 4 +--- >>> tools/testing/selftests/bpf/progs/test_sysctl_prog.c | 5 +---- >>> 5 files changed, 5 insertions(+), 19 deletions(-) >>> >>> diff --git a/tools/testing/selftests/bpf/progs/netif_receive_skb.c b/tools/testing/selftests/bpf/progs/netif_receive_skb.c >>> index 1d8918dfbd3f..7a5ebd330689 100644 >>> --- a/tools/testing/selftests/bpf/progs/netif_receive_skb.c >>> +++ b/tools/testing/selftests/bpf/progs/netif_receive_skb.c >>> @@ -5,6 +5,7 @@ >>> #include <bpf/bpf_helpers.h> >>> #include <bpf/bpf_tracing.h> >>> #include <bpf/bpf_core_read.h> >>> +#include <bpf/bpf_util.h> >> >> It doesn't look like you've built it. >> >> progs/test_sysctl_prog.c:11:10: fatal error: 'bpf/bpf_util.h' file not found >> #include <bpf/bpf_util.h> >> ^~~~~~~~~~~~~~~~ >> CLNG-BPF [test_maps] socket_cookie_prog.o >> progs/test_sysctl_loop2.c:11:10: fatal error: 'bpf/bpf_util.h' file not found >> #include <bpf/bpf_util.h> >> ^~~~~~~~~~~~~~~~ >> 1 error generated. >> In file included from progs/profiler2.c:6: >> progs/profiler.inc.h:7:10: fatal error: 'bpf/bpf_util.h' file not found >> #include <bpf/bpf_util.h> >> ^~~~~~~~~~~~~~~~ >> > > Sorry about that. I built it - I think something is wrong in my env. Build > fails complaining about not finding vmlinux - I overlooked that the failure > happened before it got to progs. > > Error: failed to load BTF from .../vmlinux: No such file or directory
Please make sure that you build vmlinux before you build selftests, BPF selftests use vmlinux to generate vmlinux.h with all kernel types (among other things). So please also make sure that all the setting in selftests/bpf/config were used in your Kconfig.
>
The problem in my env. is that I don't have CONFIG_DEBUG_INFO_BTF in my config and then don't have the dwarves and llvm-strip on my system. Pains of upgrading.
I am all set now. On the other hand the vmlinux.h is a mess. It has no guards for defines and including stdio.h and this generated vmlinux.h causes all sorts of problems.
It does have
#ifndef __VMLINUX_H__ #define __VMLINUX_H__
Are we talking about the same vmlinux.h here?
Yes we are. The guard it has works when vmlinux.h is included twice. It defines a lot of common defines which are the problem. Unless you add guards around each one of them, including vmlinux.h is problematic if you also include other standard includes.
You can try to include bpf_util.h for example from one of the test in progs to see the problem.
bpf_util.h is a user-space header, it's not going to work from the BPF program side. If you look at any of progs/*.c (all of which are BPF program-side source code), not a single one is including bpf_util.h.
Whether bpf_util.h can be included from progs isn't the main thing here. progs/test*.c including vmlinux.h (most of them seem to) can,'t include any standard .h files.
"including vmlinux.h is problematic if a test also had to include other standard includes."
This makes this header file restrictive and works in one case and one case only when no other standard headers aren't included.
It does work with other BPF-side headers that libbpf provides: bpf_tracing.h, bpf_core_read.h, etc. Yes, it doesn't work with other kernel or non-kernel headers. We are well aware of this limitation and are currently trying to convince the Clang community to let us fix that with a new attribute for Clang.
But I'm not sure what we are discussing at this point. I think we established that bpf_util.h is a user-space header and can't be used from the BPF side.
thanks, -- Shuah