stephenrkell
sigaltstack
I haven't tested it but am pretty sure programs using sigaltstack will screw with one or more things.
Indeed our stack-copying hack in do-syscall.h assumes we can copy all the way back from the current top-of-stack to the trap site's top-of-stack (marked by the saved stack pointer in the signal frame). If we're not on the same stack, that's a problem.
We can easily split that case off. It gets interesting on the return path though. Currently we do two sigreturns for each clone()-servicing trap, which is pretty wacky. Probably the way to handle it is to overwrite pretcode in the child's sigframe, to instead use a fake restorer that doesn't do sigreturn. We can do that before the child is spawned.
Again remember that trap/signal-based handling needs to go away anyway, so don't spend too much time on this.