Skip to contentSkip to conversation

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

DISCUSSION Jose E. Marchesi 6 participants 19 messages Jan 5, 2023 – Feb 28, 2023 BPF

lore
Share & tools
lore
#0 lore raw

Hello all.

Find below the notes we intend to use in today's BPF office hour to discuss possible solutions for the current limitations in the DWARF representation of the btf_type_tag C attributes, and hopefully decide on one so we can move forward with this.

The list of suggested solutions below is of course not closed: these are just the ones we could think about. Better alternatives and suggestions are very welcome!

BTF tag support in DWARF

* Current situation: annotations as children DIEs for pointees

DWARF information is structured as a tree of DIE nodes. Nodes can have attributes associated to them, as well as zero or more DIE children.

clang extends DWARF with a new tag (DIE type) =DW_TAG_LLVM_annotation=. Nodes of this type are used to associate a tag name with a tag value that is also a string.

Example:

: DW_TAG_LLVM_annotation
: DW_AT_name "btf_type_tag"
: DW_AT_const_value "user"

At the moment, clang generates =DW_TAG_LLVM_annotation= nodes as children of =DW_TAG_pointer_type= nodes. The intended semantic is that the annotation applies to the pointed-to type.

For example (indentation reflects the parent-children tree structure):

: DW_TAG_pointer_type
: DW_AT_type "int"
: DW_TAG_LLVM_annotation
: DW_AT_name "btf_type_tag"
: DW_AT_const_value "tag1"

The example above associates a "btf_type_tag->tag1" named annotation to the type pointed by its containing pointer_type, which is "int".

This approach has the advantage that, since the new =DW_TAG_LLVM_annotation= nodes are effectively used as attributes, they are safely ignored by DWARF consumers that do not understand this DIE type.

But this approach also has a big caveat: types that are not pointed-to by pointer types are not expressible in this design. This obviously impacts simple types such as =int= but also pointer types that are not pointees themselves.

For example, it is not possible to associate the tag =__tag2= to the type =int **= in this example (Note this is sparse/clang ordering.):

: int * __tag1 * __tag2 h;

- sparse
+ __tag1 applies to int*, __tag2 applies to int**
: got int *[noderef] __tag1 *[addressable] [noderef] [toplevel] __tag2 h
- clang

    + According to DWARF __tag1 applies to int*, no __tag2 (??).
    + According to BTF  __tag1 applies to int*, no __tag2 (??).
    : DWARF
    : 0x00000023:   DW_TAG_variable
    :                 DW_AT_name	("h")
    :                 DW_AT_type	(0x0000002e "int **")
    :
    : 0x0000002e:   DW_TAG_pointer_type
    :                 DW_AT_type	(0x00000037 "int *")
    :
    : 0x00000033:     DW_TAG_LLVM_annotation
    :                 DW_AT_name	("btf_type_tag")
    :                 DW_AT_const_value	("tag1")
    : BTF
    : [1] TYPE_TAG 'tag1' type_id=3
    : [2] PTR '(anon)' type_id=1
    : [3] PTR '(anon)' type_id=4
    : [4] INT 'int' size=4 bits_offset=0 nr_bits=32 encoding=SIGNED
    : [5] VAR 'h' type_id=2, linkage=global
    :
    : 'h' -> ptr -> 'tag1' -> ptr -> int

* A note about `void'

The DWARF specification recommends to denote the =void= C type by generating a DIE with =DW_TAG_unspecified_type= and name "void".

However, both GCC and LLVM do _not_ follow this recommendation and instead they denote the =void= type as the absence of a =DW_AT_type= attribute in whatever containing node.

Example, for a pointer to =void=:

: 3 DW_TAG_pointer_type [no children]

Note also that the kernel sources have sparse annotations like:

: void __user * data;

Which, using sparse ordering, means that the type which is annotated is =void=. Therefore it is very important to be able to tag the =void= basic type in this design.

GDB and other DWARF consumers understand the spec-recommended way to denote =void=.

* Solution 1: annotations as qualifiers

A possible solution for this is to handle =DW_TAG_LLVM_annotation= the same way than C type qualifiers are handled in DWARF: including them in the type chain linked by =DW_AT_type= attributes.

For example:

: DW_TAG_pointer_type
: DW_AT_type ("btf_type_tag")
:
: DW_TAG_LLVM_annotation

  :   DW_AT_name        "btf_type_tag"
  :   DW_AT_const_value "tag1"
  :   DW_AT_type        ("int")

:
: DW_TAG_base_type
: DW_AT_name ("int")

Note how now the =LLVM_annotation= has the annotated type linked by =DW_AT_type=, and acts itself as a type linked from =DW_TAG_pointer_type=.

Advantages of this approach:

- It makes sense for annotations to be implemented as qualifiers, because they actually qualify a target type.

- This approach is totally flexible and makes it possible to annotate any type, qualified or not, pointed-to or not.

- The resulting DWARF looks like the BTF.

- It can handle annotated `void', as currently generated by GCC and clang/LLVM:

: DW_TAG_LLVM_annotation
: DW_AT_name "btf_type_tag"
: DW_AT_const_value "tag1"
: DW_AT_type NULL

Disadvantages of this approach:

- Implementing this is more elaborated, and it requires DWARF consumers to understand this new DIE type, in order to follow the type chains in the
tree: =DW_TAG_LLVM_annotation= should now be expected in any =DW_AT_type= reference.

- This breaks DWARF, making it very difficult to be implemented as a compiler extension, and will likely require make it part of DWARF.

- This is not backwards compatible to what clang currently generates.

* Solution 2: annotations as children DIEs

This approach involves keeping the =DW_TAG_LLVM_annotation= DIE, with the same internal structure it has now, but associating it to the type DIE that is its parent. (Note this is not the same than being linked by a =DW_AT_type= attribute like in Solution 1.)

This means that this DWARF tree:

: DW_TAG_pointer_type
: DW_AT_type "int"
: DW_TAG_LLVM_annotation
: DW_AT_name "btf_type_tag"
: DW_AT_const_value "tag1"

Denotes an annotation that applies to the type =int*=, not the pointee type =int=.

Advantages of this approach:

- This approach makes it possible to annotate any type, qualified or not, pointed-to or not.

- This can easily be implemented as a compiler extension, because existing DWARF consumers will happily ignore the new attributes in case they don't support them; the type chains in the tree remain the same.

- Easy to implement in GCC.

Disadvantages of this approach:

- This may result in an increased number of type nodes in the tree. For example, we may have a tagged =int*= and a non-tagged =int*=, which now will have to be implemented using two different DIEs.

- This is not backwards-compatible to what clang currently generates, in the case of pointer types.

- It cannot handle annotated `void' as currently generated by GCC and clang/LLVM, so for tagged =void= we would need to generate unspecified types with name "void":

: DW_TAG_unspecified_type
: DW_AT_name "void"
: DW_TAG_LLVM_annotation
: DW_AT_name "btf_type_tag"
: DW_AT_const_value "tag1"

But this should be supported by DWARF consumers, as per the DWARF spec, and it is certainly recognized by GDB.

* Solution 3a: annotations as set of attributes

Another possible solution is to extend DWARF with a pair of two new attributes =DW_AT_annotation_tag= and =DW_AT_annotation_value=.

Annotated types will have these attributes defined. Example:

: DW_TAG_pointer_type

  :   DW_AT_type "int"
  :   DW_AT_annotation_tag   "btf_type_tag"
  :   DW_AT_annotation_value "tag1"

Note that in this example the tag applies to the pointer type, not the pointee, i.e. to =int*=.

Advantages of this approach:

- This can easily be implemented as a compiler extension, because existing DWARF consumers will happily ignore the new attributes in case they don't support them; the type chains in the tree remain the same.

- This is backwards compatible to what clang currently generates.

- Easy to implement in GCC.

Disadvantages of this approach:

- This may result in an increased number of type nodes in the tree. For example, we may have a tagged =int*= and a non-tagged =int*=, which now will have to be implemented using two different DIEs.

- It cannot handle annotated `void' as currently generated by GCC and clang/LLVM, so for tagged =void= we would need to generate unspecified types with name "void":

: DW_TAG_unspecified_type
: DW_AT_name "void"
: DW_AT_annotation_tag "btf_type_tag"
: DW_AT_annotation_value "tag1"

But this should be supported by DWARF consumers, as per the DWARF spec, and it is certainly recognized by GDB.

* Solution 3b: annotations as single "structured" attributes

This is like 3a, but using a single attribute =DW_AT_annotation= instead of two, and encoding the tag name and the tag value in the string value using some convention.

For example:

: DW_TAG_pointer_type
: DW_AT_type "int"
: DW_AT_annotation "btf_type_tag tag1"

Meaning the tag name is "btf_type_tag" and the tag value is "tag1", using the convention that a white character separates them.

Advantages over 3a:

- Using a single attribute is more robust, since it eliminates the possible situation of a node having =DW_AT_annotation_tag= and not =DW_AT_annotation_value=.

- It is easier to extend it, since the string stored in the =DW_AT_annotation= attribute may be made as complex as desired. Better than adding more =DW_AT_annotation_FOO= attributes.

- This is backwards compatible to what clang currently generates.

- Easy to implement in GCC.

Disadvantages over 3a:

- This requires defining conventions specifying the structure of the string stored in the attribute.

- This has the danger of overzealous design: "let's store a JSON tree in =DW_AT_annotation= for future extensions instead of continue bothering with DWARF".

- It cannot handle annotated `void' as currently generated by GCC and clang/LLVM, so for tagged =void= we would need to generate unspecified types with name "void":

: DW_TAG_unspecified_type
: DW_AT_name "void"
: DW_AT_annotation "btf_type_tag tag1"

But this should be supported by DWARF consumers, as per the DWARF spec, and it is certainly recognized by GDB.

#1 lore raw

We agreed in the meeting to implement Solution 2 below in both GCC and clang.

The DW_TAG_LLVM_annotation DIE number will be changed in order to make it possible for pahole to handle the current tags. The number of the new tag will be shared by both GCC and clang.

Thanks everyone for the feedback.

Quoted 302 lines

Hello all.

Find below the notes we intend to use in today's BPF office hour to discuss possible solutions for the current limitations in the DWARF representation of the btf_type_tag C attributes, and hopefully decide on one so we can move forward with this.

The list of suggested solutions below is of course not closed: these are just the ones we could think about. Better alternatives and suggestions are very welcome!

BTF tag support in DWARF

* Current situation: annotations as children DIEs for pointees

DWARF information is structured as a tree of DIE nodes. Nodes can have attributes associated to them, as well as zero or more DIE children.

clang extends DWARF with a new tag (DIE type) =DW_TAG_LLVM_annotation=. Nodes of this type are used to associate a tag name with a tag value that is also a string.

Example:

: DW_TAG_LLVM_annotation
: DW_AT_name "btf_type_tag"
: DW_AT_const_value "user"

At the moment, clang generates =DW_TAG_LLVM_annotation= nodes as children of =DW_TAG_pointer_type= nodes. The intended semantic is that the annotation applies to the pointed-to type.

For example (indentation reflects the parent-children tree structure):

: DW_TAG_pointer_type
: DW_AT_type "int"
: DW_TAG_LLVM_annotation
: DW_AT_name "btf_type_tag"
: DW_AT_const_value "tag1"

The example above associates a "btf_type_tag->tag1" named annotation to the type pointed by its containing pointer_type, which is "int".

This approach has the advantage that, since the new =DW_TAG_LLVM_annotation= nodes are effectively used as attributes, they are safely ignored by DWARF consumers that do not understand this DIE type.

But this approach also has a big caveat: types that are not pointed-to by pointer types are not expressible in this design. This obviously impacts simple types such as =int= but also pointer types that are not pointees themselves.

For example, it is not possible to associate the tag =__tag2= to the type =int **= in this example (Note this is sparse/clang ordering.):

: int * __tag1 * __tag2 h;

- sparse
+ __tag1 applies to int*, __tag2 applies to int**
: got int *[noderef] __tag1 *[addressable] [noderef] [toplevel] __tag2 h
- clang

    + According to DWARF __tag1 applies to int*, no __tag2 (??).
    + According to BTF  __tag1 applies to int*, no __tag2 (??).
    : DWARF
    : 0x00000023:   DW_TAG_variable
    :                 DW_AT_name	("h")
    :                 DW_AT_type	(0x0000002e "int **")
    :
    : 0x0000002e:   DW_TAG_pointer_type
    :                 DW_AT_type	(0x00000037 "int *")
    :
    : 0x00000033:     DW_TAG_LLVM_annotation
    :                 DW_AT_name	("btf_type_tag")
    :                 DW_AT_const_value	("tag1")
    : BTF
    : [1] TYPE_TAG 'tag1' type_id=3
    : [2] PTR '(anon)' type_id=1
    : [3] PTR '(anon)' type_id=4
    : [4] INT 'int' size=4 bits_offset=0 nr_bits=32 encoding=SIGNED
    : [5] VAR 'h' type_id=2, linkage=global
    :
    : 'h' -> ptr -> 'tag1' -> ptr -> int

* A note about `void'

The DWARF specification recommends to denote the =void= C type by generating a DIE with =DW_TAG_unspecified_type= and name "void".

However, both GCC and LLVM do _not_ follow this recommendation and instead they denote the =void= type as the absence of a =DW_AT_type= attribute in whatever containing node.

Example, for a pointer to =void=:

: 3 DW_TAG_pointer_type [no children]

Note also that the kernel sources have sparse annotations like:

: void __user * data;

Which, using sparse ordering, means that the type which is annotated is =void=. Therefore it is very important to be able to tag the =void= basic type in this design.

GDB and other DWARF consumers understand the spec-recommended way to denote =void=.

* Solution 1: annotations as qualifiers

A possible solution for this is to handle =DW_TAG_LLVM_annotation= the same way than C type qualifiers are handled in DWARF: including them in the type chain linked by =DW_AT_type= attributes.

For example:

: DW_TAG_pointer_type
: DW_AT_type ("btf_type_tag")
:
: DW_TAG_LLVM_annotation

  :   DW_AT_name        "btf_type_tag"
  :   DW_AT_const_value "tag1"
  :   DW_AT_type        ("int")

:
: DW_TAG_base_type
: DW_AT_name ("int")

Note how now the =LLVM_annotation= has the annotated type linked by =DW_AT_type=, and acts itself as a type linked from =DW_TAG_pointer_type=.

Advantages of this approach:

- It makes sense for annotations to be implemented as qualifiers, because they actually qualify a target type.

- This approach is totally flexible and makes it possible to annotate any type, qualified or not, pointed-to or not.

- The resulting DWARF looks like the BTF.

- It can handle annotated `void', as currently generated by GCC and clang/LLVM:

: DW_TAG_LLVM_annotation
: DW_AT_name "btf_type_tag"
: DW_AT_const_value "tag1"
: DW_AT_type NULL

Disadvantages of this approach:

- Implementing this is more elaborated, and it requires DWARF consumers to understand this new DIE type, in order to follow the type chains in the
tree: =DW_TAG_LLVM_annotation= should now be expected in any =DW_AT_type= reference.

- This breaks DWARF, making it very difficult to be implemented as a compiler extension, and will likely require make it part of DWARF.

- This is not backwards compatible to what clang currently generates.

* Solution 2: annotations as children DIEs

This approach involves keeping the =DW_TAG_LLVM_annotation= DIE, with the same internal structure it has now, but associating it to the type DIE that is its parent. (Note this is not the same than being linked by a =DW_AT_type= attribute like in Solution 1.)

This means that this DWARF tree:

: DW_TAG_pointer_type
: DW_AT_type "int"
: DW_TAG_LLVM_annotation
: DW_AT_name "btf_type_tag"
: DW_AT_const_value "tag1"

Denotes an annotation that applies to the type =int*=, not the pointee type =int=.

Advantages of this approach:

- This approach makes it possible to annotate any type, qualified or not, pointed-to or not.

- This can easily be implemented as a compiler extension, because existing DWARF consumers will happily ignore the new attributes in case they don't support them; the type chains in the tree remain the same.

- Easy to implement in GCC.

Disadvantages of this approach:

- This may result in an increased number of type nodes in the tree. For example, we may have a tagged =int*= and a non-tagged =int*=, which now will have to be implemented using two different DIEs.

- This is not backwards-compatible to what clang currently generates, in the case of pointer types.

- It cannot handle annotated `void' as currently generated by GCC and clang/LLVM, so for tagged =void= we would need to generate unspecified types with name "void":

: DW_TAG_unspecified_type
: DW_AT_name "void"
: DW_TAG_LLVM_annotation
: DW_AT_name "btf_type_tag"
: DW_AT_const_value "tag1"

But this should be supported by DWARF consumers, as per the DWARF spec, and it is certainly recognized by GDB.

* Solution 3a: annotations as set of attributes

Another possible solution is to extend DWARF with a pair of two new attributes =DW_AT_annotation_tag= and =DW_AT_annotation_value=.

Annotated types will have these attributes defined. Example:

: DW_TAG_pointer_type

  :   DW_AT_type "int"
  :   DW_AT_annotation_tag   "btf_type_tag"
  :   DW_AT_annotation_value "tag1"

Note that in this example the tag applies to the pointer type, not the pointee, i.e. to =int*=.

Advantages of this approach:

- This can easily be implemented as a compiler extension, because existing DWARF consumers will happily ignore the new attributes in case they don't support them; the type chains in the tree remain the same.

- This is backwards compatible to what clang currently generates.

- Easy to implement in GCC.

Disadvantages of this approach:

- This may result in an increased number of type nodes in the tree. For example, we may have a tagged =int*= and a non-tagged =int*=, which now will have to be implemented using two different DIEs.

- It cannot handle annotated `void' as currently generated by GCC and clang/LLVM, so for tagged =void= we would need to generate unspecified types with name "void":

: DW_TAG_unspecified_type
: DW_AT_name "void"
: DW_AT_annotation_tag "btf_type_tag"
: DW_AT_annotation_value "tag1"

But this should be supported by DWARF consumers, as per the DWARF spec, and it is certainly recognized by GDB.

* Solution 3b: annotations as single "structured" attributes

This is like 3a, but using a single attribute =DW_AT_annotation= instead of two, and encoding the tag name and the tag value in the string value using some convention.

For example:

: DW_TAG_pointer_type
: DW_AT_type "int"
: DW_AT_annotation "btf_type_tag tag1"

Meaning the tag name is "btf_type_tag" and the tag value is "tag1", using the convention that a white character separates them.

Advantages over 3a:

- Using a single attribute is more robust, since it eliminates the possible situation of a node having =DW_AT_annotation_tag= and not =DW_AT_annotation_value=.

- It is easier to extend it, since the string stored in the =DW_AT_annotation= attribute may be made as complex as desired. Better than adding more =DW_AT_annotation_FOO= attributes.

- This is backwards compatible to what clang currently generates.

- Easy to implement in GCC.

Disadvantages over 3a:

- This requires defining conventions specifying the structure of the string stored in the attribute.

- This has the danger of overzealous design: "let's store a JSON tree in =DW_AT_annotation= for future extensions instead of continue bothering with DWARF".

- It cannot handle annotated `void' as currently generated by GCC and clang/LLVM, so for tagged =void= we would need to generate unspecified types with name "void":

: DW_TAG_unspecified_type
: DW_AT_name "void"
: DW_AT_annotation "btf_type_tag tag1"

But this should be supported by DWARF consumers, as per the DWARF spec, and it is certainly recognized by GDB.

#2 lore raw
Quoted 7 lines · Jose E. Marchesi

We agreed in the meeting to implement Solution 2 below in both GCC and clang.

The DW_TAG_LLVM_annotation DIE number will be changed in order to make it possible for pahole to handle the current tags. The number of the new tag will be shared by both GCC and clang.

w.r.t c2x attribute syntax discussion in 01/19 office hour discussion.

I have checked clang c2x syntax w.r.t. btf_type_tag and btf_decl_tag. They are both supported with clang 15 and 16.

See:
https://clang.llvm.org/docs/AttributeReference.html

The c2x btf_decl_tag attr syntax is [[clang::btf_decl_tag("")]]. The c2x btf_type_tag attr syntax is [[clang::btf_type_tag("")]].

$ cat t.c
int [[clang::btf_type_tag("aa")]] * [[clang::btf_type_tag("bb")]] *f;
[[clang::btf_decl_tag("cc")]] int foo() { return 5; }
int bar() { return foo(); }
$ clang -std=c2x -g -O2 -c t.c
$ llvm-dwarfdump t.o | grep btf | grep tag
                  DW_AT_name    ("btf_type_tag")
                  DW_AT_name    ("btf_type_tag")
                  DW_AT_name    ("btf_decl_tag")

I double checked and the c2x syntax above generates the *same* type IR and dwarf compared to __attribute__ style attributes.

[...]

#3 lore raw
Quoted 32 lines · Jose E. Marchesi
Quoted 7 lines

We agreed in the meeting to implement Solution 2 below in both GCC and
clang.
The DW_TAG_LLVM_annotation DIE number will be changed in order to make
it possible for pahole to handle the current tags. The number of the new tag will be shared by both GCC and clang.

w.r.t c2x attribute syntax discussion in 01/19 office hour discussion.

I have checked clang c2x syntax w.r.t. btf_type_tag and btf_decl_tag. They are both supported with clang 15 and 16.

See:
https://clang.llvm.org/docs/AttributeReference.html

The c2x btf_decl_tag attr syntax is [[clang::btf_decl_tag("")]]. The c2x btf_type_tag attr syntax is [[clang::btf_type_tag("")]].

$ cat t.c
int [[clang::btf_type_tag("aa")]] * [[clang::btf_type_tag("bb")]] *f;
[[clang::btf_decl_tag("cc")]] int foo() { return 5; }
int bar() { return foo(); }
$ clang -std=c2x -g -O2 -c t.c
$ llvm-dwarfdump t.o | grep btf | grep tag
                  DW_AT_name    ("btf_type_tag")
                  DW_AT_name    ("btf_type_tag")
                  DW_AT_name    ("btf_decl_tag")
Quoted 4 lines

I double checked and the c2x syntax above generates the *same* type IR and dwarf compared to __attribute__ style attributes.

[...]

Thanks for checking.

That matches our impression that C2X type attributes actually order the same way than sparse type annotations, at least in the cases we are interested on.

#4 lore raw
Quoted 35 lines · Jose E. Marchesi
Quoted 35 lines · Jose E. Marchesi
Quoted 7 lines

We agreed in the meeting to implement Solution 2 below in both GCC and
clang.
The DW_TAG_LLVM_annotation DIE number will be changed in order to make
it possible for pahole to handle the current tags. The number of the new tag will be shared by both GCC and clang.

w.r.t c2x attribute syntax discussion in 01/19 office hour discussion.

I have checked clang c2x syntax w.r.t. btf_type_tag and btf_decl_tag. They are both supported with clang 15 and 16.

See:
https://clang.llvm.org/docs/AttributeReference.html

The c2x btf_decl_tag attr syntax is [[clang::btf_decl_tag("")]]. The c2x btf_type_tag attr syntax is [[clang::btf_type_tag("")]].

$ cat t.c
int [[clang::btf_type_tag("aa")]] * [[clang::btf_type_tag("bb")]] *f;
[[clang::btf_decl_tag("cc")]] int foo() { return 5; }
int bar() { return foo(); }
$ clang -std=c2x -g -O2 -c t.c
$ llvm-dwarfdump t.o | grep btf | grep tag
                  DW_AT_name    ("btf_type_tag")
                  DW_AT_name    ("btf_type_tag")
                  DW_AT_name    ("btf_decl_tag")

I double checked and the c2x syntax above generates the *same* type IR and dwarf compared to __attribute__ style attributes.

Quoted 7 lines
Quoted 1 line

[...]

Thanks for checking.

That matches our impression that C2X type attributes actually order the same way than sparse type annotations, at least in the cases we are interested on.

I have been experimenting with the C2x syntax in GCC and the results are similarly promising. It looks like with the C2x syntax, the 'type_tag's always associate in the same way as sparse.

For GCC the syntax is (or will be)
[[gnu::btf_decl_tag("foo")]] and
[[gnu::btf_type_tag("bar")]]
respectively.

I am not sure it is necessary to use the C2x syntax for decl_tag, iirc there are no issues with the __attribute__ syntax for decl_tag. Either one should be ok.

With C2x syntax, in the internal representation and in the generated DWARF, the type_tag attributes are attached to the same elements of the declaration as sparse attaches them to.

I checked all the examples we looked at and it seems they are all "fixed" with the C2x syntax, in that GCC agrees with sparse. For example,

$ cat ex2.c
int __attribute__((btf_type_tag("tag1"))) * __attribute__((btf_type_tag("tag2"))) * g;

We saw that this example was problematic with the __attribute__ syntax in that GCC associates "tag1" with (int **) while sparse associates "tag1" with (int).

Using the c2x syntax, "tag1" is associated with (int) and "tag2" with (int *) the same as in sparse:

$ cat ex2-c2x.c
int [[gnu::btf_type_tag("tag1")]] * [[gnu::btf_type_tag("tag2")]] * g;
$ bpf-unknown-none-gcc --std=c2x -c -gbtf -gdwarf ex2-c2x.c -o ex2-c2x.o
$ bpftool btf dump file ex2-c2x.o
[1] INT 'int' size=4 bits_offset=0 nr_bits=32 encoding=SIGNED
[2] TYPE_TAG 'tag1' type_id=1
[3] PTR '(anon)' type_id=2
[4] TYPE_TAG 'tag2' type_id=3
[5] PTR '(anon)' type_id=4
[6] VAR 'g' type_id=5, linkage=global

I also spent some studying the C2x draft standard [1] to check whether this ordering is documented by the standard or up to the implementation.
[1] https://www.open-std.org/jtc1/sc22/wg14/www/docs/n3054.pdf
(I think this is the most recent draft, dated 3 Sep 2022)

I believe the "sparse-like" ordering is in fact required by the standard, which is great for us.

The relevant section is 6.7: Declarations. Section 6.7.12 covers only syntax of attributes themselves. The ordering/association rules are documented by the sections for each component of a declaration. Section 6.7.6 is particularly relevant, 6.7.6.1 discusses pointer declarators specifically.

From my understanding, the general rule is that an attribute modifies the element of a declaration immediately to the left of it, which is the same as the intuitive sparse ordering.

So it seems like using the C2x standard attribute syntax may be a very nice solution to our problem. But we should keep in mind that C2x is still a draft so this attribute syntax could potentially change.

#5 lore raw
Quoted 41 lines · David Faust
Quoted 39 lines · Jose E. Marchesi
Quoted 32 lines · Jose E. Marchesi
Quoted 7 lines

We agreed in the meeting to implement Solution 2 below in both GCC and
clang.
The DW_TAG_LLVM_annotation DIE number will be changed in order to make
it possible for pahole to handle the current tags. The number of the new tag will be shared by both GCC and clang.

w.r.t c2x attribute syntax discussion in 01/19 office hour discussion.

I have checked clang c2x syntax w.r.t. btf_type_tag and btf_decl_tag. They are both supported with clang 15 and 16.

See:
https://clang.llvm.org/docs/AttributeReference.html

The c2x btf_decl_tag attr syntax is [[clang::btf_decl_tag("")]]. The c2x btf_type_tag attr syntax is [[clang::btf_type_tag("")]].

$ cat t.c
int [[clang::btf_type_tag("aa")]] * [[clang::btf_type_tag("bb")]] *f;
[[clang::btf_decl_tag("cc")]] int foo() { return 5; }
int bar() { return foo(); }
$ clang -std=c2x -g -O2 -c t.c
$ llvm-dwarfdump t.o | grep btf | grep tag
                   DW_AT_name    ("btf_type_tag")
                   DW_AT_name    ("btf_type_tag")
                   DW_AT_name    ("btf_decl_tag")
Quoted 4 lines

I double checked and the c2x syntax above generates the *same* type IR and dwarf compared to __attribute__ style attributes.

[...]

Thanks for checking.

Quoted 7 lines
Quoted 3 lines

That matches our impression that C2X type attributes actually order the same way than sparse type annotations, at least in the cases we are interested on.

I have been experimenting with the C2x syntax in GCC and the results are similarly promising. It looks like with the C2x syntax, the 'type_tag's always associate in the same way as sparse.

Thanks for confirmation.

Quoted 5 lines

For GCC the syntax is (or will be)
[[gnu::btf_decl_tag("foo")]] and
[[gnu::btf_type_tag("bar")]]
respectively.

Clang could add support for [[gnu::btf_decl_tag("foo")]] as well once the syntax is agreed by the community and upstreamed.

Quoted 4 lines

I am not sure it is necessary to use the C2x syntax for decl_tag, iirc there are no issues with the __attribute__ syntax for decl_tag. Either one should be ok.

The same for me. Either is okay.

Quoted 51 lines

With C2x syntax, in the internal representation and in the generated DWARF, the type_tag attributes are attached to the same elements of the declaration as sparse attaches them to.

I checked all the examples we looked at and it seems they are all "fixed" with the C2x syntax, in that GCC agrees with sparse. For example,

$ cat ex2.c
int __attribute__((btf_type_tag("tag1"))) * __attribute__((btf_type_tag("tag2"))) * g;

We saw that this example was problematic with the __attribute__ syntax in that GCC associates "tag1" with (int **) while sparse associates "tag1" with (int).

Using the c2x syntax, "tag1" is associated with (int) and "tag2" with (int *) the same as in sparse:

$ cat ex2-c2x.c
int [[gnu::btf_type_tag("tag1")]] * [[gnu::btf_type_tag("tag2")]] * g;
$ bpf-unknown-none-gcc --std=c2x -c -gbtf -gdwarf ex2-c2x.c -o ex2-c2x.o
$ bpftool btf dump file ex2-c2x.o
[1] INT 'int' size=4 bits_offset=0 nr_bits=32 encoding=SIGNED
[2] TYPE_TAG 'tag1' type_id=1
[3] PTR '(anon)' type_id=2
[4] TYPE_TAG 'tag2' type_id=3
[5] PTR '(anon)' type_id=4
[6] VAR 'g' type_id=5, linkage=global

I also spent some studying the C2x draft standard [1] to check whether this ordering is documented by the standard or up to the implementation.
[1] https://www.open-std.org/jtc1/sc22/wg14/www/docs/n3054.pdf
(I think this is the most recent draft, dated 3 Sep 2022)

I believe the "sparse-like" ordering is in fact required by the standard, which is great for us.

The relevant section is 6.7: Declarations. Section 6.7.12 covers only syntax of attributes themselves. The ordering/association rules are documented by the sections for each component of a declaration. Section 6.7.6 is particularly relevant, 6.7.6.1 discusses pointer declarators specifically.

From my understanding, the general rule is that an attribute modifies
the element of a declaration immediately to the left of it, which is the same as the intuitive sparse ordering.

So it seems like using the C2x standard attribute syntax may be a very nice solution to our problem. But we should keep in mind that C2x is still a draft so this attribute syntax could potentially change.

maintainer
#6 lore raw
In reply to Jose E. Marchesi
Quoted 9 lines · Jose E. Marchesi

We agreed in the meeting to implement Solution 2 below in both GCC and clang.

The DW_TAG_LLVM_annotation DIE number will be changed in order to make it possible for pahole to handle the current tags. The number of the new tag will be shared by both GCC and clang.

Thanks everyone for the feedback.

[...]

Hi Jose, David,

Recently I've been working on implementation of the agreed btf_type_tag encoding scheme for clang [1] and pahole [2]. While working on this, I came to a conclusion that instead of introducing new DWARF tag (0x6001) we can reuse the same tag (0x6000), but have a different DW_AT_name field:
"btf_type_tag:v2" instead of "btf_type_tag".

For example, the following C code:

    struct st {
      int __attribute__((btf_type_tag("a"))) a;
    } g;

Produces the following DWARF when [1] is used:

0x00000029: DW_TAG_structure_type

                DW_AT_name      ("st")
                ...

0x0000002e: DW_TAG_member

                  DW_AT_name    ("a")
                  DW_AT_type    (0x00000038 "int")
                ...

0x00000038: DW_TAG_base_type

                DW_AT_name      ("int")
                ...

0x0000003c: DW_TAG_LLVM_annotation

                  DW_AT_name    ("btf_type_tag:v2")
                  DW_AT_const_value     ("a")

I think that this is a tad better than abandoning 0x6000 tag because of two reasons:
- tag numbers are a limited resource;
- might simplify discussion with upstream.

(It also makes some implementation details a bit simpler, but this is not
very significant).

What do you think?

Both [1] and [2] are in a workable state, but [2] lacks support for subroutine types and "void *" for now. If you are onboard with this change I'll proceed with finalizing [1] and [2]. (Also, ":v2" suffix might be not the best, I'm open to naming suggestions).

As a somewhat orthogonal question, would it be possible for you to use the same 0x6000 tag on GCC side? I looked at master branch of [3] but can't find any mentions of btf_type_tag.

Thanks,
Eduard

[1] https://reviews.llvm.org/D143967
[2] https://github.com/eddyz87/dwarves/tree/btf-type-tag-v2
[3] git://gcc.gnu.org/git/gcc.git

#7 lore raw
Quoted 8 lines · Eduard Zingerman
Quoted 7 lines · Jose E. Marchesi

We agreed in the meeting to implement Solution 2 below in both GCC and clang.

The DW_TAG_LLVM_annotation DIE number will be changed in order to make it possible for pahole to handle the current tags. The number of the new tag will be shared by both GCC and clang.

Quoted 5 lines
Quoted 2 lines

Thanks everyone for the feedback.

[...]

Hi Jose, David,

Hi Eduard,

Quoted 37 lines

Recently I've been working on implementation of the agreed btf_type_tag encoding scheme for clang [1] and pahole [2]. While working on this, I came to a conclusion that instead of introducing new DWARF tag (0x6001) we can reuse the same tag (0x6000), but have a different DW_AT_name field:
"btf_type_tag:v2" instead of "btf_type_tag".

For example, the following C code:

    struct st {
      int __attribute__((btf_type_tag("a"))) a;
    } g;

Produces the following DWARF when [1] is used:

0x00000029: DW_TAG_structure_type

                DW_AT_name      ("st")
                ...

0x0000002e: DW_TAG_member

                  DW_AT_name    ("a")
                  DW_AT_type    (0x00000038 "int")
                ...

0x00000038: DW_TAG_base_type

                DW_AT_name      ("int")
                ...

0x0000003c: DW_TAG_LLVM_annotation

                  DW_AT_name    ("btf_type_tag:v2")
                  DW_AT_const_value     ("a")

I think that this is a tad better than abandoning 0x6000 tag because of two reasons:
- tag numbers are a limited resource;
- might simplify discussion with upstream.

Quoted 4 lines

(It also makes some implementation details a bit simpler, but this is not
very significant).

What do you think?

Very nice.
Keeping the 0x6000 tag and instead changing the name sounds good to us.

From the GCC side, support for BTF tags will be new either way but conserving DWARF tag numbers is a good idea.

Quoted 5 lines

Both [1] and [2] are in a workable state, but [2] lacks support for subroutine types and "void *" for now. If you are onboard with this change I'll proceed with finalizing [1] and [2]. (Also, ":v2" suffix might be not the best, I'm open to naming suggestions).

As for the name, I am not sure the ":v2" suffix is a good idea.

If we need a new name anyway, this could be a good opportunity to use something more generic. The annotation DIEs, especially with the new format, could be more widely useful than exclusively for producing BTF.

For example, some other tool may want to process these same user annotations which are now recorded in DWARF, but may not involve BPF/BTF at all. Tying "btf" into the name seems to unnecessarily discourage those use cases.

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.

Quoted 4 lines

As a somewhat orthogonal question, would it be possible for you to use the same 0x6000 tag on GCC side? I looked at master branch of [3] but can't find any mentions of btf_type_tag.

Yes, we plan to use the same 0x6000 in GCC. Support for btf_type_tag isn't committed in master yet; I originally worked on patches [1] last spring but they were not committed due to some of the issues we've now worked out (notably the attribute ordering/association problem). But 0x6000 is not currently in use in GCC and didn't come up as a problem for those patches, so I don't think it should be an issue.

I plan to submit a new set of patches soon, I will add you in CC in case this or similar issues come up in the discussion.

Thanks
David

[1] https://gcc.gnu.org/pipermail/gcc-patches/2022-May/593936.html

Quoted 7 lines

Thanks,
Eduard

[1] https://reviews.llvm.org/D143967
[2] https://github.com/eddyz87/dwarves/tree/btf-type-tag-v2
[3] git://gcc.gnu.org/git/gcc.git

maintainer
#8 lore raw

On Tue, 2023-02-21 at 11:38 -0800, David Faust wrote:
[...]

Quoted 5 lines

Very nice.
Keeping the 0x6000 tag and instead changing the name sounds good to us.

From the GCC side, support for BTF tags will be new either way but conserving DWARF tag numbers is a good idea.

Great, thank you!

Quoted 25 lines
Quoted 4 lines

Both [1] and [2] are in a workable state, but [2] lacks support for subroutine types and "void *" for now. If you are onboard with this change I'll proceed with finalizing [1] and [2]. (Also, ":v2" suffix might be not the best, I'm open to naming suggestions).

As for the name, I am not sure the ":v2" suffix is a good idea.

If we need a new name anyway, this could be a good opportunity to use something more generic. The annotation DIEs, especially with the new format, could be more widely useful than exclusively for producing BTF.

For example, some other tool may want to process these same user annotations which are now recorded in DWARF, but may not involve BPF/BTF at all. Tying "btf" into the name seems to unnecessarily discourage those use cases.

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...)

Quoted 5 lines

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?

Quoted 10 lines
Quoted 3 lines

As a somewhat orthogonal question, would it be possible for you to use the same 0x6000 tag on GCC side? I looked at master branch of [3] but can't find any mentions of btf_type_tag.

Yes, we plan to use the same 0x6000 in GCC. Support for btf_type_tag isn't committed in master yet; I originally worked on patches [1] last spring but they were not committed due to some of the issues we've now worked out (notably the attribute ordering/association problem). But 0x6000 is not currently in use in GCC and didn't come up as a problem for those patches, so I don't think it should be an issue.

Understood, thank you for the clarification.

Thanks,
Eduard

#9 lore raw
Quoted 41 lines · Eduard Zingerman

On Tue, 2023-02-21 at 11:38 -0800, David Faust wrote:
[...]

Quoted 5 lines

Very nice.
Keeping the 0x6000 tag and instead changing the name sounds good to us.

From the GCC side, support for BTF tags will be new either way but conserving DWARF tag numbers is a good idea.

Great, thank you!

Quoted 25 lines
Quoted 4 lines

Both [1] and [2] are in a workable state, but [2] lacks support for subroutine types and "void *" for now. If you are onboard with this change I'll proceed with finalizing [1] and [2]. (Also, ":v2" suffix might be not the best, I'm open to naming suggestions).

As for the name, I am not sure the ":v2" suffix is a good idea.

If we need a new name anyway, this could be a good opportunity to use something more generic. The annotation DIEs, especially with the new format, could be more widely useful than exclusively for producing BTF.

For example, some other tool may want to process these same user annotations which are now recorded in DWARF, but may not involve BPF/BTF at all. Tying "btf" into the name seems to unnecessarily discourage those use cases.

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...)

Quoted 5 lines

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.

Quoted 5 lines

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"?

Thanks

Quoted 16 lines
Quoted 11 lines
Quoted 4 lines

As a somewhat orthogonal question, would it be possible for you to use the same 0x6000 tag on GCC side? I looked at master branch of [3] but can't find any mentions of btf_type_tag.

Yes, we plan to use the same 0x6000 in GCC. Support for btf_type_tag isn't committed in master yet; I originally worked on patches [1] last spring but they were not committed due to some of the issues we've now worked out (notably the attribute ordering/association problem). But 0x6000 is not currently in use in GCC and didn't come up as a problem for those patches, so I don't think it should be an issue.

Understood, thank you for the clarification.

Thanks,
Eduard

#10 lore raw
Quoted 51 lines · David Faust
Quoted 41 lines · Eduard Zingerman

On Tue, 2023-02-21 at 11:38 -0800, David Faust wrote:
[...]

Quoted 5 lines

Very nice.
Keeping the 0x6000 tag and instead changing the name sounds good to us.

From the GCC side, support for BTF tags will be new either way but conserving DWARF tag numbers is a good idea.

Great, thank you!

Quoted 25 lines
Quoted 4 lines

Both [1] and [2] are in a workable state, but [2] lacks support for subroutine types and "void *" for now. If you are onboard with this change I'll proceed with finalizing [1] and [2]. (Also, ":v2" suffix might be not the best, I'm open to naming suggestions).

As for the name, I am not sure the ":v2" suffix is a good idea.

If we need a new name anyway, this could be a good opportunity to use something more generic. The annotation DIEs, especially with the new format, could be more widely useful than exclusively for producing BTF.

For example, some other tool may want to process these same user annotations which are now recorded in DWARF, but may not involve BPF/BTF at all. Tying "btf" into the name seems to unnecessarily discourage those use cases.

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...)

Quoted 5 lines

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.

Quoted 5 lines

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?

Quoted 4 lines

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.

maintainer
#11 lore raw

On Wed, 2023-02-22 at 10:11 -0800, Alexei Starovoitov wrote:
[...]

Quoted 21 lines
Quoted 21 lines
Quoted 20 lines
Quoted 9 lines

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...)

Quoted 5 lines

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?

Quoted 7 lines
Quoted 4 lines

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.

#12 lore raw
Quoted 41 lines · Eduard Zingerman

On Wed, 2023-02-22 at 10:11 -0800, Alexei Starovoitov wrote:
[...]

Quoted 21 lines
Quoted 21 lines
Quoted 20 lines
Quoted 9 lines

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...)

Quoted 5 lines

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?

Quoted 7 lines
Quoted 4 lines

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)

Quoted 2 lines

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")?

maintainer
#13 lore raw
Quoted 42 lines · Andrii Nakryiko
Quoted 41 lines · Eduard Zingerman

On Wed, 2023-02-22 at 10:11 -0800, Alexei Starovoitov wrote:
[...]

Quoted 21 lines
Quoted 21 lines
Quoted 20 lines
Quoted 9 lines

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...)

Quoted 5 lines

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?

Quoted 7 lines
Quoted 4 lines

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)

Quoted 5 lines
Quoted 2 lines

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

#14 lore raw
Quoted 47 lines · Eduard Zingerman
Quoted 45 lines · Andrii Nakryiko
Quoted 43 lines · Eduard Zingerman

On Wed, 2023-02-22 at 10:11 -0800, Alexei Starovoitov wrote:
[...]

Quoted 21 lines
Quoted 21 lines
Quoted 20 lines
Quoted 9 lines

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...)

Quoted 5 lines

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?

Quoted 7 lines
Quoted 4 lines

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.

Quoted 7 lines
Quoted 2 lines

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".

old-style tag will be deprecated and removed eventually, so I'd optimize for the new-style naming, as that's what we'll be dealing with the most going forward

Quoted 42 lines

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

maintainer
#15 lore raw
Quoted 51 lines · Andrii Nakryiko
Quoted 50 lines · Eduard Zingerman
Quoted 47 lines · Andrii Nakryiko
Quoted 41 lines · Eduard Zingerman

On Wed, 2023-02-22 at 10:11 -0800, Alexei Starovoitov wrote:
[...]

Quoted 21 lines
Quoted 21 lines
Quoted 20 lines
Quoted 9 lines

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...)

Quoted 5 lines

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?

Quoted 7 lines
Quoted 4 lines

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)

Quoted 2 lines

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")?

Quoted 8 lines
Quoted 4 lines

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".

old-style tag will be deprecated and removed eventually, so I'd optimize for the new-style naming, as that's what we'll be dealing with the most going forward

I still think that presence of a literal string "bty_type_tag" might make some grepping easier but whatever. If there are no further objections I'll post the changes using "btf:type_tag" literal tomorrow. Andrii, thanks for the input.

Thanks,
Eduard

Quoted 43 lines
Quoted 43 lines

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

#16 lore raw
Quoted 5 lines · Eduard Zingerman

I still think that presence of a literal string "bty_type_tag" might make some grepping easier but whatever. If there are no further objections I'll post the changes using "btf:type_tag" literal tomorrow. Andrii, thanks for the input.

I don't think there is precedent for using ':' inside DW_AT_name. Can we actually use the same "btf_type_tag" name?
Aren't we gonna use a different container than DW_TAG_LLVM_annotation ?

Since we're picking a standard across gcc and llvm it will be some common DW_TAG_... with the same number, no ?
I forgot what we agreed on during office hours.

#17 lore raw
Quoted 9 lines · Alexei Starovoitov
Quoted 5 lines · Eduard Zingerman

I still think that presence of a literal string "bty_type_tag" might make some grepping easier but whatever. If there are no further objections I'll post the changes using "btf:type_tag" literal tomorrow. Andrii, thanks for the input.

I don't think there is precedent for using ':' inside DW_AT_name.

I don't think there are any restrictions on string pointed to be DW_AT_name. It is used when describing source code location (so definitely has '/' on Linux, and ":/" on Windows). But I just checked Rust-emitted DWARF:

0x0002825e: DW_TAG_pointer_type

                DW_AT_type      (0x00026fc9
"core::option::Option<alloc::string::String>")
                DW_AT_name      ("*const
core::option::Option<alloc::string::String>")
                DW_AT_address_class     (0x00000000)

So I'm not too bothered about this. After all, it's just a string.

But btf:<something> allows us to generalize this to other annotations, e.g., we can have "msan:initialized" or something, and it will be done in C using some generic __attribute__((annotate("msan", "initialized"))).

Quoted 2 lines

Can we actually use the same "btf_type_tag" name?
Aren't we gonna use a different container than DW_TAG_LLVM_annotation ?

I think that was the point to reuse existing DW_TAG_LLVM_annotation (and I assume from GCC's side it would be called DW_TAG_GNU_annotation, but it will use the same ID, so effectively we might as well call it just DW_TAG_annotation), so using "btf_type_tag" becomes ambiguous.

Quoted 4 lines

Since we're picking a standard across gcc and llvm it will be some common DW_TAG_... with the same number, no ?
I forgot what we agreed on during office hours.

#18 lore raw
Quoted 32 lines · Andrii Nakryiko
Quoted 9 lines · Alexei Starovoitov
Quoted 5 lines · Eduard Zingerman

I still think that presence of a literal string "bty_type_tag" might make some grepping easier but whatever. If there are no further objections I'll post the changes using "btf:type_tag" literal tomorrow. Andrii, thanks for the input.

I don't think there is precedent for using ':' inside DW_AT_name.

I don't think there are any restrictions on string pointed to be DW_AT_name. It is used when describing source code location (so definitely has '/' on Linux, and ":/" on Windows). But I just checked Rust-emitted DWARF:

0x0002825e: DW_TAG_pointer_type

                DW_AT_type      (0x00026fc9
"core::option::Option<alloc::string::String>")
                DW_AT_name      ("*const
core::option::Option<alloc::string::String>")
                DW_AT_address_class     (0x00000000)

So I'm not too bothered about this. After all, it's just a string.

But btf:<something> allows us to generalize this to other annotations, e.g., we can have "msan:initialized" or something, and it will be done in C using some generic __attribute__((annotate("msan", "initialized"))).

Quoted 8 lines
Quoted 2 lines

Can we actually use the same "btf_type_tag" name?
Aren't we gonna use a different container than DW_TAG_LLVM_annotation ?

I think that was the point to reuse existing DW_TAG_LLVM_annotation (and I assume from GCC's side it would be called DW_TAG_GNU_annotation, but it will use the same ID, so effectively we might as well call it just DW_TAG_annotation), so using "btf_type_tag" becomes ambiguous.

I see. Then 'btf:type_tag' makes the most sense.