Re: [PATCH bpf-next] bpftool: Support dumping kfunc prototypes from BTF

From: Jiri Olsa
Date: Tue Jan 30 2024 - 03:22:03 EST


On Mon, Jan 29, 2024 at 09:13:13AM -0700, Daniel Xu wrote:
> Hi Jiri,
>
> On Mon, Jan 29, 2024 at 01:11:17PM +0100, Jiri Olsa wrote:
> > On Sun, Jan 28, 2024 at 06:35:33PM -0700, Daniel Xu wrote:
> > > This patch enables dumping kfunc prototypes from bpftool. This is useful
> > > b/c with this patch, end users will no longer have to manually define
> > > kfunc prototypes. For the kernel tree, this also means we can drop
> > > kfunc prototypes from:
> > >
> > > tools/testing/selftests/bpf/bpf_kfuncs.h
> > > tools/testing/selftests/bpf/bpf_experimental.h
> > >
> > > Example usage:
> > >
> > > $ make PAHOLE=/home/dxu/dev/pahole/build/pahole -j30 vmlinux
> > >
> > > $ ./tools/bpf/bpftool/bpftool btf dump file ./vmlinux format c | rg "__ksym;" | head -3
> > > extern void cgroup_rstat_updated(struct cgroup * cgrp, int cpu) __ksym;
> > > extern void cgroup_rstat_flush(struct cgroup * cgrp) __ksym;
> > > extern struct bpf_key * bpf_lookup_user_key(u32 serial, u64 flags) __ksym;
> >
> > hi,
> > I'm getting following declaration for bpf_rbtree_add_impl:
> >
> > extern int bpf_rbtree_add_impl(struct bpf_rb_root * root, struct bpf_rb_node * node, bool (struct bpf_rb_node *, const struct bpf_rb_node *)* less, void * meta__ign, u64 off) __ksym;
> >
> > and it fails to compile with:
> >
> > In file included from skeleton/pid_iter.bpf.c:3:
> > ./vmlinux.h:164511:141: error: expected ')'
> > 164511 | extern int bpf_rbtree_add_impl(struct bpf_rb_root * root, struct bpf_rb_node * node, bool (struct bpf_rb_node *, const struct bpf_rb_node *)* less, void * meta__ign, u64 off) __ksym;
> > | ^
> > ./vmlinux.h:164511:31: note: to match this '('
> > 164511 | extern int bpf_rbtree_add_impl(struct bpf_rb_root * root, struct bpf_rb_node * node, bool (struct bpf_rb_node *, const struct bpf_rb_node *)* less, void * meta__ign, u64 off) __ksym;
> >
> > looks like the btf_dumper_type_only won't dump function pointer argument
> > properly.. I guess we should fix that, but looking at the other stuff in
> > vmlinux.h like *_ops struct we can print function pointers properly, so
> > perhaps another way around is to use btf_dumper interface instead
>
> Ah, crap, looks like between all the branch switching I didn't build
> vmlinux with kfunc annotations. Having fixed that, I can repro this
> build failure.
>
> I'll take a look and see what the best way to fix this is.
>
> Given that end to end the whole flow basically works, should we start
> working on merging patches?

yes, the flow looks good to me.. will check the rest of the patches

jirka