On Thu, Jan 18, 2024 at 11:50 AM Eduard Zingerman <eddyz87@xxxxxxxxx> wrote:
>
> On Wed, 2024-01-17 at 14:33 -0800, Andrii Nakryiko wrote:
> [...]
>
> > + /* special cases */
> > + switch (prog->type) {
> > + case BPF_PROG_TYPE_KPROBE:
> > + case BPF_PROG_TYPE_PERF_EVENT:
> > + /* `struct pt_regs *` is expected, but we need to fix up */
> > + if (btf_is_struct(t) && strcmp(tname, "pt_regs") == 0)
> > + return true;
> > + break;
>
> Just to double-check my understanding, in patch #3 you say:
>
> > for perf_event kernel allows `struct {pt_regs,user_pt_regs,user_regs_struct} *`.
>
> Here `true` is returned only for `pt_regs`,
> meaning that arch specific types "user_pt_regs" and "user_regs_struct"
> would not be converted to "bpf_perf_event_data" but "pt_regs" would, right?
yes, it's a slight deviation from what I ended up doing in the kernel,
because I initially didn't know how to deal with arch-specific
definitions of bpf_user_pt_regs_t. But at the last moment I figured
out that __builtin_types_compatible_p and forward declaring structs
works, so I'll do a small follow up to libbpf to match kernel logic
completely
>
> [...]
>
>
