On Fri, Apr 25, 2025 at 12:43 PM Eduard Zingerman <eddyz87@xxxxxxxxx> wrote:
>
> Eduard Zingerman <eddyz87@xxxxxxxxx> writes:
>
> > Alexei Starovoitov <alexei.starovoitov@xxxxxxxxx> writes:
> >
> >> Hi All,
> >>
> >> Looks like pahole fails to deduplicate BTF when kernel and
> >> kernel module are built with gcc-14.
> >> I see this issue with various kernel .config-s on bpf and
> >> bpf-next trees.
> >> I tried pahole 1.28 and the latest master. Same issues.
> >>
> >> BTF in bpf_testmod.ko built with gcc-14 has 2849 types.
> >> When built with gcc-13 it has 454 types.
> >> So something is confusing dedup logic.
> >> Would be great if dedup experts can take a look,
> >> since this dedup issue is breaking a lot of selftests/bpf.
> >
> > It does not look like the problem is with dedup.
> > Quick glance at structure definitions does not show any duplications,
> > just much more structs compared to clang:
>
> Or maybe it is.
> For example, task_struct is added to .ko BTF generated by gcc, but not
> clang. This can only happen if dedup fails to merge structures in base
> and module btf, right?
>
> Here is an interesting observation:
>
> $ bpftool btf dump file ~/tmp/objs-gcc/bpf_testmod.ko format c | awk '/struct task_struct \{/ {s=1} s {print $0} /^\}/ {s=0}' > ~/tmp/task_struct.ko.c
>
> $ bpftool btf dump file ~/tmp/objs-gcc/vmlinux format c | awk '/struct task_struct \{/ {s=1} s {print $0} /^\}/ {s=0}' > ~/tmp/task_struct.vmlinux.c
>
> $ diff -pruN ~/tmp/task_struct.ko.c ~/tmp/task_struct.vmlinux.c
> --- /home/ezingerman/tmp/task_struct.ko.c 2025-04-25 12:37:48.312480603 -0700
> +++ /home/ezingerman/tmp/task_struct.vmlinux.c 2025-04-25 12:38:03.096644654 -0700
> @@ -18,7 +18,6 @@ struct task_struct {
> int static_prio;
> int normal_prio;
> unsigned int rt_priority;
> - long: 0;
> struct sched_entity se;
I reproed this issue with default .ko build that includes:
--btf_features=distilled_base
Once I disabled it and did
bpftool btf dump file ./bpf_testmod.ko --base-btf .../vmlinux format c
the task_struct from vmlinux.h and from testmod.h became exactly the same.
So it sounds like the 3rd issue :)
bpftool dump of distilled btf needs work.
