On Tue, Nov 19, 2019 at 2:16 PM Alexei Starovoitov
<alexei.starovoitov@xxxxxxxxx> wrote:
>
> On Mon, Nov 18, 2019 at 10:21:51PM -0800, Andrii Nakryiko wrote:
> >
> > -static __attribute__ ((noinline))
> > -int test_pkt_access_subprog2(int val, volatile struct __sk_buff *skb)
> > +__attribute__ ((noinline))
> > +int test_pkt_access_subprog2(int val, struct __sk_buff *skb)
> > {
> > return skb->len * val;
> > }
>
> Did you run test_progs -n 8?
I ran all of them, but missed that number of failing tests increased.
>
> Above breaks it with:
> 10: (61) r1 = *(u32 *)(r6 +40)
> func 'test_pkt_access_subprog2' doesn't have 6-th argument
> invalid bpf_context access off=40 size=4
>
> The point of the subprog2 is to test the scenario where BTF disagress with llvm
> optimizations.
>
Yeah, right. I should keep test_pkt_access_subprog2 as static then.
Will post v2 with that change.
