这就是你输掉瞬态执行战争的方式
This is How You Lose the Transient Execution War
AI总结：
本文指出针对Spectre等瞬态执行漏洞的现有缓解措施缺乏全面性且性能代价高昂，若继续依赖旧方案，未来硬件将因叠加缓解层而性能受损且仍受新旧变体威胁。
摘要：
2018年，与推测执行和乱序执行、故障注入以及微架构侧信道相关的一类新漏洞引起了关注。瞬态执行漏洞背后的技术并不新，但这些技术的组合应用比以往认为的可能情况更为复杂，且安全影响更为严重。针对瞬态执行漏洞的变体，人们已经提出并实施了众多缓解措施。虽然基于异常的Meltdown型瞬态执行漏洞已被证明易于处理，但Spectre型漏洞和其他基于推测的瞬态执行漏洞对对策的抵抗力要强得多。一些提议的缓解措施已被硬件供应商和软件开发人员广泛采用，但结合这些通常部署的缓解措施并不能产生有效且全面的解决方案，它只能防御一小部分变体。多年来，新提出的缓解措施趋向于更有效、更全面且性能更好的方法，然而，尽管旧缓解措施安全收益有限且性能损失高昂，但仍然最受欢迎。如果我们继续这样下去，我们可以预见未来多代硬件将因新变体被发现而不断增加的缓解层导致性能受损，且仍然容易受到已知和未来变体的攻击。
英文摘要：
A new class of vulnerabilities related to speculative and out-of-order execution, fault-injection, and microarchitectural side channels rose to attention in 2018. The techniques behind the transient execution vulnerabilities were not new, but the combined application of the techniques was more sophisticated, and the security impact more severe, than previously considered possible. Numerous mitigations have been proposed and implemented for variants of the transient execution vulnerabilities. While Meltdown-type exception-based transient execution vulnerabilities have proven to be tractable, Spectre-type vulnerabilities and other speculation-based transient execution vulnerabilities have been far more resistant to countermeasures. A few proposed mitigations have been widely adopted by hardware vendors and software developers, but combining those commonly deployed mitigations does not produce an effective and comprehensive solution, it only protects against a small subset of the variants. Over the years, newly proposed mitigations have been trending towards more effective and comprehensive approaches with better performance, and yet, older mitigations remain the most popular despite limited security benefits and prohibitive performance penalties. If we continue this way, we can look forward to many generations of hardware debilitated by performance penalties from increasing layers of mitigations as new variants are discovered, and yet still vulnerable to both known and future variants.
机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。
