Off hand, I concur with you and Linus. I wasn't involved in the
introduction of TIF_RESTORE_SIGMASK and never thought too much about
it before. In my recent changes I only considered the issue of
getting it out of _TIF_WORK_MASK et al to free up the low bit, and
didn't contemplate the structure of the implementation beyond that.
I don't think it's "too late" to change anything. (It's never too
late! It just might take a while to make a change in a safe and
orderly fashion for all the arch's.) After my patch series, the
details are fully in the arch's corner. It should be straightforward
to convert one at a time to use regs->return_register = -ERESTARTRESTOREMASK
to implement set_restore_sigmask() if you want to tackle it.
Thanks,
Roland
--
