Re: [PATCH v2 2/3] perf symbols: Try reading the symbol table with libbfd

[Remi Bernon <rbernon@codeweavers.com>]

On 2020-08-04 22:32, Jiri Olsa wrote:
> On Tue, Aug 04, 2020 at 10:57:35AM +0200, Remi Bernon wrote:
> 
> SNIP
> 
>> +
>> +int dso__load_bfd_symbols(struct dso *dso, const char *debugfile)
>> +{
>> +	int err = -1;
>> +	long symbols_size, symbols_count;
>> +	asection *section;
>> +	asymbol **symbols, *sym;
>> +	struct symbol *symbol;
>> +	bfd *abfd;
>> +	u_int i;
>> +	u64 start, len;
>> +
>> +	abfd = bfd_openr(dso->long_name, NULL);
>> +	if (!abfd)
>> +		return -1;
>> +
>> +	if (!bfd_check_format(abfd, bfd_object)) {
>> +		pr_debug2("%s: cannot read %s bfd file.\n", __func__,
>> +			  dso->long_name);
>> +		goto out_close;
>> +	}
>> +
>> +	if (bfd_get_flavour(abfd) == bfd_target_elf_flavour)
>> +		goto out_close;
> 
> aah, so the code is actualy only for non elf objects,
> somehow I thought it's replacing the symbol load globaly
> 
> jirka
> 

Maybe it could but because libbfd exposes the common subset of all its 
backends, it doesn't provide symbol sizes for instance (that's how I 
understand it). So I didn't want to lose features compared to what the 
libelf path is currently doing.

For non-elf objects, the implementation here works around that by 
assuming that all the symbols are following each other contiguously, but 
that's a little bit incorrect.
-- 
Rémi Bernon <rbernon@codeweavers.com>