The threat model for PC systems has changed dramatically over the years. Inserting a card (e.g., AGP, PCI, PCI-X, PCIe) was an action that required physical access to the system board, and the system could be protected with a case lock and intrusion alarms. It was regarded that a plug-in card that the user chose to install in their system should therefore be trusted. There have been many changes to this simple picture that result in a very different threat model today, for example:

• External connectors that are DMA capable, such as Thunderbolt and the USB-C connector, which makes internal data accessible without needing to open the case.
• The ability of devices to hide their purpose, such as a device that appears to be a USB-C charger, but that has malicious capability hidden such as wireless networking capability (a Trojan).
• Downloadable firmware that is used to implement devices (known as soft devices) that used to be hard-coded.
• Counterfeit devices and components in the supply chain.
• Malicious actors building back doors into genuine devices.
• Virtualization, where we have multiple operating systems sharing the same platform.
• Greatly increasing complexity opening the door to design errors and implementation bugs.
• Sophisticated attackers with resources to search for the smallest hole.
• Increased rewards for a successful system attack, whether monetary (perhaps from the theft of crypto), or from destructive intent to a target (such as a power grid).
• Increasing connectivity of systems, such that a platform that was isolated and therefore secure, is now online and attackable from the other side of the world.
• Device portability, such that physical access to a system in a secure data center is not required, the target is a laptop or phone class of system.
• Growth of sensitive data on devices, such that the system attacker might be a “good guy” such as law enforcement trying to prevent a terrorist attack.
• An increasing complex world where the definition of “good guy” and “bad guy” are not clear.

We have moved from a world where a simple I/O device installed in a system was reasonably safe, to a world where literally everything could be used as a possible attack vector. The weak points may come from design limitations, from designs being used outside of the environment that they were intended for, and from implementation defects.

The simple attack over PCIe has moved from being:

• a simple DMA read/write of system memory, or
• an errant interrupt causing execution of unintended code

to being a question of the inability to trust the PCIe fabric itself. For example, has an attacker been able to switch on the debug facilities in a firmware-implemented PCIe switch, and use the switch as a malicious TLP generator? (The answer is Yes).

This PCIe focused security course starts by looking at some of the simpler forms of attacks such as DMA, and some of the defenses again those attacks (IOMMU), and also some of the holes in these defense mechanisms. We then move onto looking at some of the additional features added to help protect the fabric (such as source validation in Access Control Services), and then onto the newest features for evaluating trust, link encryption, and point-to-point encryption. We also look at the supported for a TEE (Trusted Execution Environment) VM.

This is not a platform security course as there are many other topics of relevance to security that we will not discuss, such as UEFI, hypervisors, operating systems, processor security features, etc.

This class looks at potential attacks on “legacy” PCIe and some of the mitigations (ACS, IOMMU, etc), and then explores the newer features (CMA, IDE, SPDM, TDISP, etc). This class is based on the ECNs to PCIe 4.0 and 5.0, and also covers the changes from PCIe 6.0 and 6.1.

All of MindShare's classroom and virtual classroom courses can be customized to fit the needs of your group.

The MindShare PCIe 5.0 / 6.x class or equivalent knowledge is recommended. Basic knowledge of processor (Intel/AMD/ARM) and computer architecture.