>
> kfunc is a corresponding mechanism to introduce unstable api
> from bpf into the kernel instead of stable helpers.
> Just whitelist some functions as unstable kfunc helpers and call them
> from bpf progs.
> See net/bpf/test_run.c and bpf_kfunc_call* for inspiration.
>
I also like this idea.
However, for hid_hw_raw_request() I can not blindly enable that
function in all program types. This function makes the kernel sleep,
and so we can not use it while in IRQ context.
I think I can detect if we are in IRQ or not, but is it really worth
enabling it across all BPF program types when we know that only
SEC("hid/user_event") will use it?
Also, I am not sure how we can make bpf_hid_get_data() work with that.
We need to teach the verifier how much memory is provided, and I do
not see how you can do that with kfunc.
Cheers,
Benjamin
