Re: [PATCH] libbpf: extending BTF_KIND_INIT to accommodate some unusual types

[Yonghong Song <yonghong.song@linux.dev>]

On 4/24/24 12:06 AM, Xin Liu wrote:
> On Tue, 23 Apr 2024 13:12:04 -0700 Yonghong Song <yonghong.song@linux.dev> wrote:
>> On 4/23/24 6:15 AM, Xin Liu wrote:
>>> On Mon, 22 Apr 2024 10:43:38 -0700 Andrii Nakryiko <andrii.nakryiko@gmail.com> wrote:
>>>
>>>> On Mon, Apr 22, 2024 at 7:46 AM Xin Liu <liuxin350@huawei.com> wrote:
>>>>> In btf__add_int, the size of the new btf_kind_int type is limited.
>>>>> When the size is greater than 16, btf__add_int fails to be added
>>>>> and -EINVAL is returned. This is usually effective.
>>>>>
>>>>> However, when the built-in type __builtin_aarch64_simd_xi in the
>>>>> NEON instruction is used in the code in the arm64 system, the value
>>>>> of DW_AT_byte_size is 64. This causes btf__add_int to fail to
>>>>> properly add btf information to it.
>>>>>
>>>>> like this:
>>>>>     ...
>>>>>      <1><cf>: Abbrev Number: 2 (DW_TAG_base_type)
>>>>>       <d0>   DW_AT_byte_size   : 64              // over max size 16
>>>>>       <d1>   DW_AT_encoding    : 5        (signed)
>>>>>       <d2>   DW_AT_name        : (indirect string, offset: 0x53): __builtin_aarch64_simd_xi
>>>>>      <1><d6>: Abbrev Number: 0
>>>>>     ...
>>>>>
>>>>> An easier way to solve this problem is to treat it as a base type
>>>>> and set byte_size to 64. This patch is modified along these lines.
>>>>>
>>>>> Fixes: 4a3b33f8579a ("libbpf: Add BTF writing APIs")
>>>>> Signed-off-by: Xin Liu <liuxin350@huawei.com>
>>>>> ---
>>>>>    tools/lib/bpf/btf.c | 2 +-
>>>>>    1 file changed, 1 insertion(+), 1 deletion(-)
>>>>>
>>>>> diff --git a/tools/lib/bpf/btf.c b/tools/lib/bpf/btf.c
>>>>> index 2d0840ef599a..0af121293b65 100644
>>>>> --- a/tools/lib/bpf/btf.c
>>>>> +++ b/tools/lib/bpf/btf.c
>>>>> @@ -1934,7 +1934,7 @@ int btf__add_int(struct btf *btf, const char *name, size_t byte_sz, int encoding
>>>>>           if (!name || !name[0])
>>>>>                   return libbpf_err(-EINVAL);
>>>>>           /* byte_sz must be power of 2 */
>>>>> -       if (!byte_sz || (byte_sz & (byte_sz - 1)) || byte_sz > 16)
>>>>> +       if (!byte_sz || (byte_sz & (byte_sz - 1)) || byte_sz > 64)
>>>> maybe we should just remove byte_sz upper limit? We can probably
>>>> imagine 256-byte integers at some point, so why bother artificially
>>>> restricting it?
>>>>
>>>> pw-bot: cr
>>> In the current definition of btf_kind_int, bits has only 8 bits, followed
>>> by 8 bits of unused interval. When we expand, we should only use 16 bits
>>> at most, so the maximum value should be 8192(1 << 16 / 8), directly removing
>>> the limit of byte_sz. It may not fit the current design. For INT type btfs
>>> greater than 255, how to dump is still a challenge.
>> Looking at this patch. Now I remember that I have an old pahole patch
>> to address similar issues
>>     https://lore.kernel.org/bpf/20230426055030.3743074-1-yhs@fb.com/
>> which is not merged and I forgot that.
>>
>> In that particular case, the int size is 1024 bytes.
>> Currently the int type more than 16 bytes cannot be dumped in libbpf.
>> Do you have a particular use case to use your__builtin_aarch64_simd_xi() type
>> in bpf program? I guess probably not as BPF does not support
>> builtin function your__builtin_aarch64_simd_xi().
>>
> Currently, there is no use case of byte_sz in btf, so let's remove
> __builtin_aarch64_simd_xi first.At least this will support the kernel
> compilation phase without causing the kernel to fail directly when
> generating btf.

Okay, I will resend my pahole patch later to address this issue.

>
>>> Does the current version support a maximum of 8192 bytes?
>>>
>>>>>                   return libbpf_err(-EINVAL);
>>>>>           if (encoding & ~(BTF_INT_SIGNED | BTF_INT_CHAR | BTF_INT_BOOL))
>>>>>                   return libbpf_err(-EINVAL);
>>>>> --
>>>>> 2.33.0
>>>>>