Re: BTF tag support in DWARF (notes for today's BPF Office Hours)

[Eduard Zingerman <eddyz87@gmail.com>]

On Mon, 2023-02-27 at 13:13 -0800, Andrii Nakryiko wrote:
> On Wed, Feb 22, 2023 at 11:51 AM Eduard Zingerman <eddyz87@gmail.com> wrote:
> > 
> > On Wed, 2023-02-22 at 10:11 -0800, Alexei Starovoitov wrote:
> > [...]
> > > > > > What do you think about something like "debug_type_tag" or
> > > > > > "debug_type_annotation" (and a similar update for the decl tags)?
> > > > > > The translation into BTF records would be the same, but the DWARF info
> > > > > > would stand on its own without being tied to BTF.
> > > > > > 
> > > > > > (Naming is a bit tricky since terms like 'tag' are already in use by
> > > > > > DWARF, e.g. "type tag" in the context of DWARF DIEs makes me think of
> > > > > > DW_TAG_xxxx_type...)
> > > > > > 
> > > > > > As far as I understand, early proposals for the tags were more generic
> > > > > > but the LLVM reviewers wished for something more specific due to the
> > > > > > relatively limited use of the tags at the time. Now that the tags and
> > > > > > their DWARF format have matured I think a good case can be made to
> > > > > > make these generic. We'd be happy to help push for such change.
> > > > > 
> > > > > On the other hand, BTF is a thing we are using this annotation for.
> > > > > Any other tool can reuse DW_TAG_LLVM_annotation, but it will need a
> > > > > way to distinguish it's annotations from BTF annotations. And this can
> > > > > be done by using a different DW_AT_name. So, it seems logical to
> > > > > retain "btf" in the DW_AT_name. What do you think?
> > > > 
> > > > OK I can understand keeping it BTF specific.
> > > > 
> > > > Other than that, I don't come up with any significantly different idea
> > > > than to use the ":v2" suffix, so let's go with "btf_type_tag:v2"?
> > > 
> > > I don't like v2 suffix either.
> > > Please come up with something else.
> > 
> > Nothing particularly good comes to mind:
> > - btf_type_tag:wrapper
> > - btf_type_tag:outer
> > - btf_type_tag:own
> > - exterior_btf_type_tag
> > - outer_btf_tag
> > - btf_type_prefix
> > - btf_type_qualifier (as in const/volatile)
> > 
> > Or might as well use btf_type_tag:gcc, as you suggested earlier,
> > but it is as confusing as the others.
> 
> btf.type_tag or btf:type_tag or btf/type_tag (you get the idea, it's
> "BTF scoped")?

`btf/type_tag` is nice but might be somewhat confusing when DWARF is inspected:
- both old-style and new-style tags would be present in DWARF for some
  time for backwards compatibility;
- old-style tag has name "btf_type_tag".

Thus, the following C code:

  #define __tag1 __attribute__((btf_type_tag("tag1")))
  #define __tag2 __attribute__((btf_type_tag("tag2")))

  int __tag1 * __tag2 g;

Would be encoded in DWARF as:

  0x29:   DW_TAG_pointer_type
            DW_AT_type      (0x35 "int")
  
  0x2e:     DW_TAG_LLVM_annotation
              DW_AT_name    ("btf/type_tag:")
              DW_AT_const_value     ("tag2")
  
  0x31:     DW_TAG_LLVM_annotation
              DW_AT_name    ("btf_type_tag")
              DW_AT_const_value     ("tag1")
  
  0x34:     NULL
  
  0x35:   DW_TAG_base_type
            DW_AT_name      ("int")
            DW_AT_encoding  (DW_ATE_signed)
            DW_AT_byte_size (0x04)
  
  0x39:     DW_TAG_LLVM_annotation
              DW_AT_name    ("btf/type_tag:")
              DW_AT_const_value     ("tag1")
  
  0x3c:     NULL
  
Which is not very helpful.

In my opinion "btf_type_tag:v2" is the least confusing option, but if
Alexei does not like it, let's use "btf_type_tag:parent" and move on.

Thanks,
Eduard