It fails some regression tests on eglibc, is that on purpose?
I looked quickly at the patch, please note that linuxthreads library uses internally signals 32-34 in a different way to NPTL, I doubt they can be reused for timers.
As for bug22 I doubt it has anything to do with my patch. In my tests it was working fine with kfreebsd 8.1 but failed with kfreebsd 10 (both using patched libc). Could it be a kernel-driven regression?
It is not a regression, the bug22 is CPU dependent. The test itself is broken, see #644023. Petr