On 29/11/2024 09:07, Eduard Zingerman wrote:
> On Thu, 2024-11-28 at 01:23 +0000, Ihor Solodrai wrote:
>> From: Alan Maguire <alan.maguire@xxxxxxxxxx>
>>
>> This will help us do more accurate DWARF/ELF matching.
>
> It would be good to have a more detailed explanation here.
> E.g. number of generated functions differs with this patch:
>
> # without this patch
> $ bpftool btf dump file /home/eddy/work/tmp/old.btf | grep "\] FUNC '" | wc -l
> 48056
> # with this patch
> $ bpftool btf dump file /home/eddy/work/tmp/new.btf | grep "\] FUNC '" | wc -l
> 48189
>
> It would be helpful to peek one of newly added functions and explain
> why it was previously excluded.
>
I've also found the opposite situation; that some functions are not
present after this patch that were before. In your case - where more
were matched - I suspect the extra address hint allowed us to do a
better job of matching ELF and DWARF such that we matched .isra.0
functions that represent actual function boundaries rather than .cold
subfunctions that had different signatures. I'm working on a respin of
the initial patches that ensures we get the same (or more) number of
functions, but it shouldn't change substantially anything that follows
that Ihor has done.
>>
>> Signed-off-by: Alan Maguire <alan.maguire@xxxxxxxxxx>
>> Signed-off-by: Ihor Solodrai <ihor.solodrai@xxxxx>
>> ---
>> btf_encoder.c | 37 +++++++++++++++++++++++++++++++------
>> 1 file changed, 31 insertions(+), 6 deletions(-)
>>
>> diff --git a/btf_encoder.c b/btf_encoder.c
>> index 98e4d7d..01d7094 100644
>> --- a/btf_encoder.c
>> +++ b/btf_encoder.c
>> @@ -88,6 +88,7 @@ struct btf_encoder_func_state {
>> struct elf_function {
>> const char *name;
>> char *alias;
>> + uint32_t addr;
>> size_t prefixlen;
>> struct btf_encoder_func_state state;
>> };
>> @@ -131,6 +132,7 @@ struct btf_encoder {
>> int allocated;
>> int cnt;
>> int suffix_cnt; /* number of .isra, .part etc */
>> + uint64_t base_addr;
>
> This field is set, but never read.
>
>> } functions;
>> };
>>
>
> [...]
>
