To avoid further wasting time solving this wrong: NAK. pt_regs is the actual interrupted state, not some oddly sanitized version thereof. NMI isn't even the only way you can have a weird pt_regs like this.
It seems that some older kernels contain bugs where they incorrectly assume that regs->rsp is a valid pointer. Why don't you write a patch for *that * and send it to lkml, us, and sta...@vger.kernel.org.
