Re: Building perf with BUILD_BPF_SKEL=1 by default
From: Song Liu <hidden>
Date: 2021-12-06 03:47:16
Also in:
lkml
On Dec 5, 2021, at 5:06 AM, Arnaldo Carvalho de Melo [off-list ref] wrote: Em Fri, Dec 03, 2021 at 07:32:34PM +0000, Song Liu escreveu:quoted
quoted
On Dec 3, 2021, at 5:10 AM, Arnaldo Carvalho de Melo [off-list ref] wrote: Hi Song, So I'm changing all my containers to build with BUILD_BPF_SKEL=1 to then make this the default, so far older containers fail either because the clang available is too old, so I've added a NO_BUILD_BPF_SKEL=1 env var to disable that in those containers and then there is this other case where clang is recent enough but: util/bpf_skel/bperf_leader.bpf.c:13:20: error: use of undeclared identifier 'BPF_F_PRESERVE_ELEMS' __uint(map_flags, BPF_F_PRESERVE_ELEMS); Because the system's /usr/include/linux/bpf.h doesn't have that BPF_F_PRESERVE_ELEMS enum entry. These are enums to make them available via BTF, but then I can't use the: #ifdef BPF_F_PRESERVE_ELEMS #define BPF_F_PRESERVE_ELEMS (1U << 11) #endif approach. But then we _have_ it in the tools/include/uapi/linux/bpf.h we ship: $ grep BPF_F_PRESERVE_ELEMS tools/include/uapi/linux/bpf.h BPF_F_PRESERVE_ELEMS = (1U << 11), $ so we need to switch to using it somehow, this way we can build in more systems and make bperf and other BPF enabled features. From a quick look I couldn't find where to add $(sourcedir)/tools/include/uapi/ to the include path used to build util/bpf_skel/bperf_leader.bpf.c, should be easy, can you take a look? Thanks, - ArnaldoI think the following should fix itI'm trying this now and tentatively sticking a Signed-off-by: you to it, ok?
Thanks for verifying this! I should learn to do these container tests myself. Yes, you can include Signed-off-by: Song Liu <song@kernel.org> (or my fb email, either one works. ) Thanks, Song