I checked our current state vs. regression tests, and I encountered a
few failures. Most of them I do consider bugs in the regression tests, not
in the sparc64 port - I'm discussing those with the respective authors.

One interesting failure though is documented in PR port-sparc64/33075:

Just do

  cd /usr/src/regress/lib/libc/clone
  make
  make regress

to see if it fails, and then repeat the "make regress" step as root
(for this test it does make a difference).

I can not reproduce the test failure reported in that PR with a GENERIC
kernel on the same machine with a custom kernel now - and I have no idea
why.

It would be interesting to hear if anyone can make it fail too.

FYI: besides that test and the ones that I consider false positives we only
fail the getaddrinfo test (discussed on tech-net, I can not make it succeed
on any of my machines), and the make tests (discussed on tech-toolchain, this
is probably not arch specific). Especialy we pass all SA and libpthread 
tests ;-)

Martin