[PATCH] bpf: fix index register for classic BPF indirect loads
Stephen Hemminger
stephen at networkplumber.org
Tue Nov 10 17:48:24 CET 2026
On Sat, 31 Oct 2026 18:58:37 +0000
Marat Khalili <qm2k21 at gmail.com> wrote:
> Converter code fix looks ok, is there any way to have a test?
>
> BTW x86 JIT seems to still allow negative offsets in indirect mode (we
> only fixed it in ARM).
Negative offsets are BPF convention to read meta data.
For cBPF which has socket filters it is how kernel reads vlan offload etc.
But current libpcap does not generate socket filter negative offsets unless
doing live capture (ie to Linux kernel). It might be useful to support
a limited amount of DPDK specific negative offsets, but that would require
coordination (or reimplement) with pcap_compile() in libpcap.
More information about the dev
mailing list