Re: [PATCH] contrib/plugins/uftrace_symbols.py: generate debug files to map symbols to source

["Philippe Mathieu-Daudé" <philmd@linaro.org>]

On 15/10/25 23:12, Pierrick Bouvier wrote:
> On 10/15/25 2:02 PM, Pierrick Bouvier wrote:
>> On 10/15/25 1:40 PM, Philippe Mathieu-Daudé wrote:
>>> On 13/10/25 23:39, Pierrick Bouvier wrote:
>>>> Enhance uftrace_symbols.py to generate .dbg files, containing
>>>> source location for every symbol present in .sym file.
>>>> It allows to use uftrace {replay,dump} --srcline and show origin of
>>>
>>> `uftrace {replay,dump} --srcline`
>>>
>>>> functions, connecting trace to original source code.
>>>>
>>>> It was first implemented with pyelftools DWARF parser, which was way
>>>> to slow (~minutes) to get locations for every symbol in the linux
>>>
>>> s/to/too/
>>>
>>>> kernel. Thus, we use addr2line instead, which runs in seconds.
>>>>
>>>> As well, there were some bugs with latest pyelftools release,
>>>> requiring to run master version, which is not installable with pip.
>>>> Thus, since we now require binutils (addr2line), we can ditch 
>>>> pyelftools
>>>> based implementation and simply rely on nm to get symbols information,
>>>
>>> `nm`
>>>
>>>> which is faster and better.
>>>>
>>>> Signed-off-by: Pierrick Bouvier <pierrick.bouvier@linaro.org>
>>>> ---
>>>>     contrib/plugins/uftrace_symbols.py | 108 ++++++++++++++++++ 
>>>> +----------
>>>>     1 file changed, 72 insertions(+), 36 deletions(-)


>>>> +def get_symbols(elf_file):
>>>> +    symbols=[]
>>>> +    try:
>>>> +        out = subprocess.check_output(['nm', '--print-size', 
>>>> elf_file],
>>>> +                                      stderr=subprocess.STDOUT,
>>>> +                                      text=True)
>>>
>>> Nitpicking, we might be using cross-compiled `nm`, so maybe not hardcode
>>> the binary name.


>>> No blocking comments:
>>> Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
>>>
>>
>> Thanks Philippe.
>>
>> For the cross compiled tools, I'm not really sure it's worth making this
>> more complex. Having tooling for a cross architecture is an advanced
>> setup, and I think it's fair to expect someone to have binutils
>> installed if they have any cross compiler and cross binutils.
>> Plus, they can always create a symlink if needed.
>>
> 
> As well, nm and addr2line are architecture agnostic, as they just access 
> ELF symtab and DWARF information, so you don't need an aarch64-*-{nm, 
> addr2line} to extract information from an aarch64 binary.

Right, good point.