On Mon, Oct 02, 2017 at 02:07:23PM +0200, Jesper Dangaard Brouer wrote:
> On Fri, 29 Sep 2017 20:06:09 -0700
> Alexei Starovoitov <[email protected]> wrote:
>
> > > +/*** Trace point code ***/
> > > +
> > > +/* Tracepoint format:
> > > /sys/kernel/debug/tracing/events/xdp/xdp_redirect/format
> > > + * Code in: kernel/include/trace/events/xdp.h
> > > + */
> > > +struct xdp_redirect_ctx {
> > > + unsigned short common_type; // offset:0; size:2; signed:0;
> > > + unsigned char common_flags; // offset:2; size:1; signed:0;
> > > + unsigned char common_preempt_count;// offset:3; size:1; signed:0;
> > > + int common_pid; // offset:4; size:4; signed:1;
> >
> > this part is not right. First 8 bytes are not accessible by bpf code.
> > Please use __u64 pad; or similar here.
>
> I've corrected this in V3.
>
> Can you explain why BPF cannot access these (first 8 bytes) struct members?
in the very first patch adding bpf to tracepoint we allowed the access,
but then people complained that common_* stuff is not abi and since it was
slow to populate anyway due to local_save_flags (which is never used by
scripts we had so far) and for pid there was a helper already, I was happy
to get rid of it and save these 8 bytes for future use. which ended up
being used by hidden(inaccessible) pt_regs pointer.
Layout description is in
commit 98b5c2c65c29 ("perf, bpf: allow bpf programs attach to tracepoints")
The work on dynamic bits of tracepoints is still tbd.
