I've committed this patch that H.J. suggested to fix the ld testsuite failures
for Xtensa targets. It took me a while to figure out that the place_orphan hook
was only being used to work around a problem in a very old version of binutils.
Linking with -r had been causing the order of orphan sections to be reversed,
which was apparently a problem with an Xtensa port of VxWorks. I don't know
when that was fixed, but it doesn't happen anymore on mainline. I verified that
with this patch all the ld tests pass for an xtensa-elf target.
Thanks, H.J. You saved me more time with this patch than I spent last week
trying to figure out if the failures were expected. I'm sorry for giving you a
hard time about that.