Re: wiki page on using stap without dwarf

[Jim Keniston <jkenisto at us dot ibm dot com>]

On Mon, 2008-05-19 at 21:45 -0700, Roland McGrath wrote:
> > It still sounds like it should be a win.  (One problem - shared by all
> > the symbol-table sources - is relocatability, since we lose section
> > information.  Likewise, we lose section-based blacklisting, hmm...)

In the current implementation, we do section-based blacklisting if we
have the elf available to us, but not if we have just the textual symbol
table.  (But see below.)

> 
> For .ko modules, you always have ELF section info.

Correct.

> For vmlinux, the
> sections that are most magical have boundary symbols defined with special
> names chosen by the kernel's linker script.  For the important blacklist
> sections we might have to use explicit symbol names in PC-range blacklists.
> The boundary symbol names are coded by hand in kernel sources, but use a
> formulaic pattern.  It might even cover the actual cases sufficiently to
> try the canonical boundary symbol names in lieu of each section name.

Hmm.  I looked at this once and I thought I had concluded that symbols
like __init_begin and __init_end weren't really definitive; but looking
at it now, I don't see the problem.  Probably I was looking at .ko
files, where the __init addresses overlap with the others.

I'll pursue this approach.  Thanks.

> 
> 
> Thanks,
> Roland

Jim