[Bug translator/1276] support more timer varieties

[Roland McGrath]

> Places that come to mind are those routines that change the "current"
> pointers, and routines (if any) involved in computing the "current"
> macro, to test for reentrancy problems.

The macro is inlined into a little bit of assembly.  The pointer is stored
on the stack and never changed (you just switch stacks).  The only current
pointer that gets changed is the one on interrupt stacks, which is set up
before switching to the interrupt stack.  So the closest to "dangerous"
places would be inside context switching and inside interrupt handling.