On Mon, Feb 28, 2022 at 11:19 AM +01, Ilya Leoshkevich wrote: > In order to resolve this inconsistency I've implemented patch 1 of this > series. With that, "sk->dst_port == bpf_htons(0xcafe)" starts to fail, > and that's where one needs something like this patch. Truth be told I can't reproduce the said failure. I've extended the test with an additional check: 306 bool ok = sk->dst_port == bpf_htons(0xcafe); 307 if (!ok) 308 RET_LOG(); 309 if (!sk_dst_port__load_word(sk)) 310 RET_LOG(); ... but it translates to the same BPF assembly as sk_dst_port__load_half. That is: ; bool ok = sk->dst_port == bpf_htons(0xcafe); 9: (69) r1 = *(u16 *)(r6 +12) 10: (bc) w1 = w1 ; if (!ok) 11: (16) if w1 == 0xcafe goto pc+2 12: (b4) w1 = 308 13: (05) goto pc+14 During the test I had patch 1 from this series applied on top of [1]. The latter series should not matter, though, I didn't touch the access converter. Mind sharing what does the `bpftool prog objdump` output look like for the failing test on your side? [1] https://lore.kernel.org/bpf/20220227202757.519015-1-jakub@xxxxxxxxxxxxxx/
