Re: [PATCH dwarves] btf_encoder: Fix a dwarf type DW_ATE_unsigned_1024 to btf encoding issue

From: Yonghong Song <yhs@meta.com>
To: Nick Desaulniers <ndesaulniers@google.com>, Yonghong Song <yhs@fb.com>
Cc: Arnaldo Carvalho de Melo <arnaldo.melo@gmail.com>,
	dwarves@vger.kernel.org, Alexei Starovoitov <ast@kernel.org>,
	Andrii Nakryiko <andrii@kernel.org>,
	bpf@vger.kernel.org, Daniel Borkmann <daniel@iogearbox.net>,
	Eduard Zingerman <eddyz87@gmail.com>,
	kernel-team@fb.com,
	Satya Durga Srinivasu Prabhala <quic_satyap@quicinc.com>
Subject: Re: [PATCH dwarves] btf_encoder: Fix a dwarf type DW_ATE_unsigned_1024 to btf encoding issue
Date: Wed, 26 Apr 2023 13:49:46 -0700	[thread overview]
Message-ID: <632e7eb5-dcd9-5863-da59-30d7a27aceda@meta.com> (raw)
In-Reply-To: <CAKwvOdnXh0e0F=_5nuVcMNsHAkqkc+K5FrOmktFZ76z3X_zHug@mail.gmail.com>

On 4/26/23 11:09 AM, Nick Desaulniers wrote:
> On Tue, Apr 25, 2023 at 10:50 PM Yonghong Song <yhs@fb.com> wrote:
>>
>> Nick Desaulniers reported an issue ([1]) where an 128-byte sized type
>> (DW_ATE_unsigned_1024) cannot be encoded into BTF with failure message
>> likes below:
>>    $ pahole -J reduced.o
>>    [2] INT DW_ATE_unsigned_1024 Error emitting BTF type
>>    Encountered error while encoding BTF.
>> See [1] for how to reproduce the issue.
>>
>> The failure is due to currently BTF int type only supports upto 16
>> bytes (__int128) and in this case the dwarf int type is 128-byte.
>>
>> The DW_ATE_unsigned_1024 is not a normal type for variable/func
>> declaration etc. It is used in DW_AT_location. There are two
>> ways to resolve this issue.
>>    (1). If btf encoding is expected, remove all dwarf int types
>>         where btf encoding will failure, e.g., non-power-of-2
>>         bytes, or greater than 16 bytes.
>>    (2). do a sanitization in btf_encoder ([2]).
>>
>> This patch uses method (2) since it is a simple fix in btf_encoder.
>> I checked my local built vmlinux with latest
>> bpf-next. There is only one instance of DW_ATE_unsigned_24 (used in
>> DW_AT_location) so I expect irregular int types should be very rare.
>>
>>    [1] https://github.com/libbpf/libbpf/pull/680
>>    [2] commit 7d8e829f636f ("btf_encoder: Sanitize non-regular int base type")
>>
>> Signed-off-by: Yonghong Song <yhs@fb.com>
> 
> Thanks, this fixed the above reported error for me.  My report is just
> forwarded from Satya.
> 
> Reported-by: Satya Durga Srinivasu Prabhala <quic_satyap@quicinc.com>
> Tested-by: Nick Desaulniers <ndesaulniers@google.com>
> 
> I don't know if that change has other implications for unusual byte sizes.
> 
> We might need to consider at some point waiting to validate
> DW_TAG_base_type until we know that they're not used outside of
> DW_AT_location expressions.

David Blaikie confirmed that indeed special types like 
DW_ATE_unsigned_1024 is indeed generated for DW_AT_location.
See https://github.com/libbpf/libbpf/pull/680 for details.

> 
>> ---
>>   btf_encoder.c | 2 +-
>>   1 file changed, 1 insertion(+), 1 deletion(-)
>>
>> diff --git a/btf_encoder.c b/btf_encoder.c
>> index 65f6e71..1aa0ad0 100644
>> --- a/btf_encoder.c
>> +++ b/btf_encoder.c
>> @@ -394,7 +394,7 @@ static int32_t btf_encoder__add_base_type(struct btf_encoder *encoder, const str
>>           * these non-regular int types to avoid libbpf/kernel complaints.
>>           */
>>          byte_sz = BITS_ROUNDUP_BYTES(bt->bit_size);
>> -       if (!byte_sz || (byte_sz & (byte_sz - 1))) {
>> +       if (!byte_sz || (byte_sz & (byte_sz - 1)) || byte_sz > 16) {
>>                  name = "__SANITIZED_FAKE_INT__";
>>                  byte_sz = 4;
>>          }
>> --
>> 2.34.1
>>
> 
> 