Comment 11 for bug 709711

> For some reason bit 0 is always 0 in the code so will never have an effect, but that may be a (separate) bug.

It turns out that that's deliberate; the relevant register has bit 0 (and 16) read-as-zero, because it only has 15 levels of interrupt and level 0 means "no interrupts active":

http://www.ibiblio.org/pub/historic-linux/early-ports/Sparc/NCR/NCR89C105.txt

So it might be a better patch to just change the upper bound of the loop, since we will never do anything in the j==0 iteration.