[PATCH} PPC fix td_thr_get_addr for TLS

[Roland McGrath]

> OK I may have jumped to the wrong conclusion. It looks like the ELF symbol
> table is bias free (but generated code is not). 

That's my understanding, and it's as I think it should be.  It's consistent
with e.g., positive rather than negative ST_TLS symbol values for x86.  All
the code in ld, prelink, etc. that deals with the layout of the TLS block
need not have any machine-specific parameters aside from default alignment.
In fact, the ELF spec is written to describe the symbol table in purely
machine-independent terms.  Things like negation or constant bias are part
of the behavior of particular relocs, which are all machine-specific.