Re: RS6000_MCOUNT and -fpic

[Jim Wilson <wilson at specifixinc dot com>]

gp@qnx.com wrote:
>   if (flag_pic && DEFAULT_ABI == ABI_V4)
>     fprintf("\tbl %s@plt",RS6000_MCOUNT);
>   else
>     fprintf("\tbl %s",RS6000_MCOUNT);

It looks like the same construct used in several places in the rs6000.md 
file, so it seems to be right.

> - Why would I end up with a self-referencing "bl" instruction without the
> "@plt"?  Any idea?  I think I've seen it before, so even though it looks 
> like a bug, it might be a feature (I guess...)  :-)

probably a relocation got ignored or mishandled.

If you compile to a .o file without pic, and disassemble it, pc-relative 
calls will look like calls to themselves.  However, if you look at the 
relocation table, you will see a reloc that modifies the call target to 
contain the address of the target function.  So after linking it will 
point to the right place, but before linking it just has a placeholder 0 
offset, which for most targets means a call to itself.

With pic, the relocation handling is a bit different.  It may be that 
you still have an unhandled reloc after linking, and the call still has 
the placeholder 0.  If the loader silently ignores the (presumably bad) 
reloc, then the program will silently fail.  Or if the loader mishandles 
the reloc things fail.  Alternatively, it could be that the linker 
mishandled the reloc, or silently ignored it.  You would have to look at 
the resulting executable including its relocs, and the linker/load 
behaviour, and what the ABI says about these relocs.  There are lots of 
different things that might have happened here.
--
Jim Wilson, GNU Tools Support, http://www.SpecifixInc.com