Re: [PATCH bpf-next v7 6/6] selftests/bpf: tc_tunnel - validate decap GSO and encapsulation state

> On Aug 12, 2026, at 11:20 AM, [email protected] wrote:
> 
> [Severity: Medium]
> Does this pointer arithmetic break compilation on 32-bit architectures?
> 
> Since kskb->end is a pointer on 32-bit systems (rather than an integer offset
> like it is on 64-bit), adding kskb->head and kskb->end together attempts to add
> two pointers, which is invalid in C.
> 
> Additionally, CO-RE does not translate this pointer-integer addition to a single
> pointer dereference when cross-compiled on 64-bit and run on 32-bit kernels,
> which can lead to a garbage pointer dereference.
> 
> Is there a safer way to calculate the skb_shared_info offset in decap_internal
> that is portable across both 32-bit and 64-bit architectures?


Martin KaFai Lau previously replied.

"Other tests have been using it to get shinfo already. I would leave it as is.
If it really would be needed to run and compile on 32-bit in the future,
it could use bpf_core_field_size(kskb->end) and then do things differently.”

https://lore.kernel.org/all/[email protected]/