Re: [PATCH bpf v3 2/3] bpf: Skip the progs of trampoline images that are being freed

[Florent Revest]

On Fri Sep 25, 2026 at 12:35 AM UTC, Alexei Starovoitov wrote:
> On Thu Sep 24, 2026 at 5:05 PM UTC, Florent Revest (Anthropic) wrote:
> > call sequences, which replaces ip_after_call. Tasks that are still in an
> > image that was put skip all of its progs from then on, including the
> > ones that are still attached, like they already skipped its fexit
> > progs. The call to the original function is left alone, a task that is
> > before it in the image still has to run it.
>
> So I slept on it and you were right to argue that skipping fmod_ret
> is a bad idea. bpf-ci explains it well.
>
> So let's patch only the progs that are being detached instead of all.
> Pretty much what you had in v2, but also explain why list is needed
> and why just patching previous tramp is not enough.

Ack, I will respin in v4 with a better explanation than I had in v2.

> Also see CI re: lower limit for x86.

Yes, I could reproduce it. I'll lower the limit to 36 on x86.

> Adjusting BPF_MAX_TRAMP_LINKS for large page sizes is a bigger change.
> Ignore it for now.

Alright!