Re: [PATCH] pickles: Update btf.pk

[Jose E. Marchesi]

> On 12/15/21 22:35, Jose E. Marchesi wrote:
>> Hi David.
>> 
>>> Update BTF pickle with recent additions/changes to the format:
>>> - BTF_KIND_DATASEC uses 'size' not 'type'
>>> - Add BTF_KIND_FLOAT
>>> - Add BTF_KIND_DECL_TAG and BTF_KIND_TYPE_TAG
>>>
>>> 2021-12-15  David Faust  <david.faust@oracle.com>
>>>
>>>     * pickles/btf.pk (BTF_KIND_FLOAT): New kind.
>>>     (BTF_KIND_DECL_TAG): Likewise.
>>>     (BTF_KIND_TYPE_TAG): Likewise.
>>>     (BTF_Decl_Tag): New type.
>>>     (BTF_Type): Update `info` bits usage to accomodate new type kinds.
>>>     BTF_KIND_DATASEC uses `size` variant of `attrs`.
>>>     Add `decl_tag` variant to `data`.
>>> ---
>>>   ChangeLog      | 10 ++++++++++
>>>   pickles/btf.pk | 24 +++++++++++++++++++-----
>>>   2 files changed, 29 insertions(+), 5 deletions(-)
>>>
>>> diff --git a/pickles/btf.pk b/pickles/btf.pk
>>> index 4832ada6..a23944db 100644
>>> --- a/pickles/btf.pk
>>> +++ b/pickles/btf.pk
>>> @@ -35,12 +35,15 @@ var BTF_KIND_UNKNOWN = 0,
>>>       BTF_KIND_FUNC = 12,
>>>       BTF_KIND_FUNC_PROTO = 13,
>>>       BTF_KIND_VAR = 14,
>>> -    BTF_KIND_DATASEC = 15;
>>> +    BTF_KIND_DATASEC = 15,
>>> +    BTF_KIND_FLOAT = 16,
>>> +    BTF_KIND_DECL_TAG = 17,
>>> +    BTF_KIND_TYPE_TAG = 18;
>>>     var btf_kind_names =
>>>     ["unkn", "int", "ptr", "array", "struct", "union", "enum",
>>>      "fwd", "typedef", "volatile", "const", "restrict", "func",
>>> -   "func_proto", "var", "datasec"];
>>> +   "func_proto", "var", "datasec", "float", "decl_tag", "type_tag"];
>>>     type BTF_Type_Id = uint<32>;
>>>   @@ -100,6 +103,14 @@ type BTF_Var_SecInfo =
>>>       offset<uint<32>,B> size;
>>>     };
>>>   +type BTF_Decl_Tag =
>>> +  struct
>>> +  {
>>> +    /* -1 if the tag is applied to the type itself,
>>> +       else the index of the tagged field/parameter/etc.  */
>>> +    uint<32> component_idx;
>>> +  };
>> Wouldn't that be 0xffff_ffff instead of -1?  In any case, it may be
>> useful for the user to have a variable like BTF_DECL_TAG_SELF defined to
>> this special value.
>
> Thanks for pointing this out. Of course -1 doesn't make much sense here.
>
> After double checking, this should indeed be a _signed_ int32. Looks
> like a small error in the format documentation which refers to it as 
> unsigned (while also using the -1).

I'm glad we catched that mistake in the kernel Documentation files.

> Fixed, added BTF_DECL_TAG_SELF, and pushed.

Thanks.