As the fallout of the Spectre and Meltdown vulnerabilities settles, the future of in-silica security becomes fuzzier. There are many comprehensive reports on the attack vectors, patches and respective performance degradation, perhaps most lucidly presented by Peter Bright at Ars Technica. For Meltdown, at least, the impact for “standard desktop workloads” will be much less than the initial estimates of a 30% slowdown, although kernel-intensive workloads that require a lot of disk I/O or network communications will take a significant hit. The Spectre class of vulnerabilities are much more complex and myriad fixes from OS vendors, chip manufacturers and application developers are ongoing; mileage may vary.
Perhaps more worrisome is the potential impact on future chipsets intended to enable privacy-preserving operations. In such instances, application developers and service providers are explicitly relying on hardware manufacturers to provide computational privacy for their customer's data, their IP and other crown jewels they wish to protect. Intel’s Secure Guard Extensions (SGX) has been touted by many as the future for secure computation in finance, healthcare and other industries that process sensitive data. The CTO of Microsoft Azure announced in September 2017 that SGX enclaves form one of the pillars of their Confidential Computing capabilities; however, a github repo from the LSDS group at Imperial College demonstrating Spectre-like attacks on SGX will likey impact such aspirations. Assuredly, the Intel team is working on this and more will be known in the coming weeks and months.
The lesson here is that hardware, particularly when fitting 100 million devices into a square millimeter, is hard. This is all the more reason to rely on flexible software-defined security and encryption solutions rather than baking secure circuits into sand.