One of the big problems with Spectre and Meltdown is that they are due to behaviours and components of a processor that are not well documented. Caches, branch

One of the big problems with Spectre and Meltdown is that they are due to behaviours and components of a processor that are not well documented. Caches, branch predictors, out of order execution and superscalar execution are all things that processor vendors refuse to disclose any information about because it is so commercially sensitive. Indeed CPUs are now primarily differentiated by these features. We are no longer getting processors with faster clocks or even smaller transistors so a lot of research has gone into making individual instructions run faster (or at least run faster most of the time). Neither Intel, ARM or AMD have disclosed why their CPU design are, or are not vulnerable to Spectre and Meltdown. If the developers of compilers or even operating systems don't know exactly how the hardware works then it is pretty hard for them to know that they haven't written vulnerable code. By the same token, it is also hard for them to know if they have written performant code. And we know even less about other types of processors that exist in almost every chip such as GPUs, digital signal processors, video processor units (VPUs) and other hardware accelerators. Those processors can almost certainly also leak data from other parts of the system, they have their own caches and sometimes branch predictors. Although it is hard to specify the code that runs on these devices it is almost certainly still possible to exploit bugs and timing attacks in them. We are essentially getting security through the obscurity of the devices and security through obscurity is no good. So should we have processor architectures which are completely open? In some ways that would destroy most vendor's business models. But having more openness could lead to more divergent processor architectures as vendors try to differentiate themselves and it should improve the security of our systems.