Re: [RFC PATCH v1] perf symbol: Correct address for bss symbols

[Leo Yan]

Hi Fangrui,

On Tue, Jul 12, 2022 at 08:29:52PM -0700, Fangrui Song wrote:

[...]

> > We need to create symbol info for not only .text section but also for
> > .data section and .bss sectionṡ.  So based on the data address, we can
> > know what's the symbol for the data access.
> >
> > But I need to correct the description for "st_value" [1]: In
> > executable and shared object files, st_value holds a virtual address.
> > To make these files' symbols more useful for the dynamic linker, the
> > section offset (file interpretation) gives way to a virtual address
> > (memory interpretation) for which the section number is irrelevant.
> >
> > So perf tool uses the formula "st_value - sh_addr + sh_offset" to
> > convert from the memory address to file address.  But it calculates
> > the wrong file address because "sh_offset" doesn't respect the
> > alignment.
> 
> Thanks for the explanation. I think st_value - p_vaddr + p_offset  may
> be a better formula where p_vaddr/p_offset is from the PT_LOAD program
> header.
> 
> For a SHT_NOBITS section, sh_offset may not be accurate, but PT_LOAD
> has precise information.

Thanks a lot for suggestion, it's very helpful and reasonable for me!

I struggled a bit for considering two things.  One is how to refactor
kernel symbol parsing with PT_LOAD program headers, because the kernel
symbol parsing is relative complex for both kernel symbols and module
symbols, this is why I didn't move furthermore for refactoring kernel
symbol parsing.

The second thing is I observe there have some spurious symbols with
'st_value' are zeros.  So there have an extra fixing for this case.

Welcome comments or suggestions for the new patch set:
https://lore.kernel.org/lkml/20220724022857.2621520-1-leo.yan@xxxxxxxxxx/T/#t

Thanks,
Leo