On Thu, Jul 25, 2019 at 4:26 PM Alexei Starovoitov
<alexei.starovoitov@xxxxxxxxx> wrote:
>
> On Wed, Jul 24, 2019 at 12:27:38PM -0700, Andrii Nakryiko wrote:
> > Add tests for various array handling/relocation scenarios.
> >
> > Signed-off-by: Andrii Nakryiko <andriin@xxxxxx>
> ...
> > +
> > +#define CORE_READ(dst, src) \
> > + bpf_probe_read(dst, sizeof(*src), __builtin_preserve_access_index(src))
>
> This is the key accessor that all progs will use.
> Please split just this single macro into individual commit and add
> detailed comment about its purpose and
> what __builtin_preserve_access_index() does underneath.
I'm planning to add more powerful and flexible set of macros to
support BCC style a->b->c->d accesses using single macro. Something
like BPF_CORE_READ(&dst, sizeof(dst), a, b, c, d). I want to move
bpf_helpers.h into libbpf itself first, after some clean up. How about
I write all that at that time and for now just add this simpler
CORE_READ into bpf_helpers.h?
Relocations recorded by __builtin_preserve_access_index() are
described pretty well in patch #1, which adds bpf_offset_reloc. I'll
double check if I mention this built-in there, and if not - will add
that.
>
> > +SEC("raw_tracepoint/sys_enter")
> > +int test_core_nesting(void *ctx)
> > +{
> > + struct core_reloc_arrays *in = (void *)&data.in;
> > + struct core_reloc_arrays_output *out = (void *)&data.out;
> > +
> > + /* in->a[2] */
> > + if (CORE_READ(&out->a2, &in->a[2]))
> > + return 1;
> > + /* in->b[1][2][3] */
> > + if (CORE_READ(&out->b123, &in->b[1][2][3]))
> > + return 1;
