Re: [PATCHv4 bpf-next 2/6] uprobe: Do not emulate/sstep original instruction when ip is changed

[Andrii Nakryiko]

On Tue, Sep 16, 2025 at 2:53 PM Jiri Olsa <jolsa@xxxxxxxxxx> wrote:
>
> If uprobe handler changes instruction pointer we still execute single
> step) or emulate the original instruction and increment the (new) ip
> with its length.
>
> This makes the new instruction pointer bogus and application will
> likely crash on illegal instruction execution.
>
> If user decided to take execution elsewhere, it makes little sense
> to execute the original instruction, so let's skip it.
>
> Acked-by: Oleg Nesterov <oleg@xxxxxxxxxx>
> Acked-by: Andrii Nakryiko <andrii@xxxxxxxxxx>
> Signed-off-by: Jiri Olsa <jolsa@xxxxxxxxxx>
> ---
>  kernel/events/uprobes.c | 7 +++++++
>  1 file changed, 7 insertions(+)
>
> diff --git a/kernel/events/uprobes.c b/kernel/events/uprobes.c
> index 7ca1940607bd..2b32c32bcb77 100644
> --- a/kernel/events/uprobes.c
> +++ b/kernel/events/uprobes.c
> @@ -2741,6 +2741,13 @@ static void handle_swbp(struct pt_regs *regs)
>
>         handler_chain(uprobe, regs);
>
> +       /*
> +        * If user decided to take execution elsewhere, it makes little sense
> +        * to execute the original instruction, so let's skip it.
> +        */
> +       if (instruction_pointer(regs) != bp_vaddr)
> +               goto out;
> +

Peter, Ingo,

Are you guys ok with us routing this through the bpf-next tree? We'll
have a tiny conflict because in perf/core branch there is
arch_uprobe_optimize() call added after handler_chain(), so git merge
will be a bit confused, probably. But it should be trivially
resolvable.

>         if (arch_uprobe_skip_sstep(&uprobe->arch, regs))
>                 goto out;
>
> --
> 2.51.0
>