Shared Memory. SP1 SP0 SP1 SP0 CPU1 CRC calculation and CRC calculation and Comparator partition Comparator partition CPU0 CRC calculation and Comparator partition CRC calculation and Comparator partition Figure 29: Shared memory diagnosis pattern - Solution 2 - Scenario 2.4b: Additional considerations and solution.
Shared Memory. Modern multi-core devices contain different cache memories which can be applied for private use (e.g., L1 cache) or can be shared between the components of the device (e.g., L2 cache). Secondary cache or L2 cache memory is commonly used to improve the performance of the system when significant data traffic is generated by the processor. The use of L2 cache assumes the presence of a primary cache or L1 cache, which is usually coupled or internal to the processor (See Figure 3 and Figure 5). However, the use of the secondary cache memory implies ever greater complexity of the system and a new source of interferences which may lead to an undesirable behaviour of the system. For example, two cores may access (write/read) to the same shared memory at the same time, causing the blocking effect, which may cause the failure of the overall system. In single-core architecture domain, the IEC 61508 safety standard covers the failures caused by memory sharing such as the causal factors of the execution interference between elements of a single computer platform (See Annex F of IEC 61508-3 [20] Techniques for achieving non-interference between SW elements on a single computer). In addition, this safety standard recommends a set of measures and diagnostic techniques to detect the random failures of both variable and invariable memories. However, the measures and diagnostic techniques recommended by this standard are focused on single-core architectures where a resource (e.g., memory region) cannot be shared by more than one component at the same time. Instead, in multi-core architectures the sharing of a resource between more than one component (e.g., two cores) is a usual task. Therefore the measures and diagnostic techniques which are recommended by the IEC 61508 safety standard are not directly applicable to multi-core architectures, but have to be extended according to the given conditions. Different research studies propose techniques to solve or reduce as much as possible the interferences caused by the shared memories. For example, techniques to improve the performance of the system by reducing the memory interferences of the applications [33-35] are proposed. These techniques focus in scheduling policies which provide request prioritization and reduce the inter- partition interferences. Other solutions aim to control the mapping of application’s data to memory channels [36]. In short, the shared memories can be considered such as a recurrent source of interferen...
Shared Memory. Mem.1 Mem. SP1 CRC calculation and Comparator partition CRC calculation and Comparator partition Figure 28: Shared memory diagnosis pattern - Solution 2 - Scenario 2.4a: Additional considerations and solutions. Mem. Mem.1
Shared Memory. “Memory” between the communicating parties is essentially erased at the end of every session2, implying no mutual, long-standing history of shared session state to bootstrap from (this will be important later). In contrast, a CKA protocol constitutes a long, continuous session with an ongoing key agreement evolving throughout. The CKA session may in fact have a multi-year lifespan – with entity authentication occurring only once, at the 1 To avoid a potential terminology ambiguity around use of active adversary (active key compromise attacks or active impersonation attacks through message or key update injection), we will henceforth use active adversary strictly to refer to an adversary that is injecting messages or key updates, e.g. an adversarial Xxxxxx-Xxxxxxx share. Key compromise will be treated as separate action in an attack, and will be referred to as a compromise. 2 TLS [23] offers a session resumption option using a pre-shared key. As this mode is an option only and refreshment of keying material inside of the mode is a further sub-option in the standard we do not go into a detailed comparison.
Shared Memory. A CKA provides a continuous “memory” between the communicating parties. Thus, to combat the risk of compromise in such a long-lived session, the CKA can periodically refresh its secret values by dividing the entire session into a series of epochs, where the state from one epoch is input to the next. If communicating parties authentically communicate at epoch i and one of them is compromised, then the routine process of updating the key would lock out the attacker at epoch i + 1 (provided entity authentication is not broken by an adversary’s injected update). Future impersonation attempts would also be blocked. PCS is predicated on the requirement that the adversary is passive for one epoch, else the lack of re-authentication could lead to potentially years of an undetected man-in-the-middle (MitM) attacker. Epochs may be far shorter than a typical TLS session length, which significantly limits data exposure (i.e., FS and PCS guarantees are linked to epochs). In the first instance, session-based protocols have advantages under (1) but disadvantages under (2), while the reverse is true under a CKA. To illustrate this contrast between session-based and CKA protocols, consider the following case example comparing TLS and CKA security under passive and active attackers. Suppose, for simplicity, that two devices need to communicate on a regular basis for 3 years. The developer can choose between using a CKA, supporting asynchronous communication, or TLS, supporting synchronous communication only. CKA epochs will change over every time a party sends a message, while in the case of TLS sessions will be 1 hour in duration, thrice a day. Now, if an adversary compromises device A approximately one year in, we have the outcomes shown in Table 1, dependent on whether the adversary is passive or active immediately following the compromise (i.e., attempting impersonation along with key updates if necessary). For illustration, we assume that the attack is not detected. Protocol # Session Establishments Data Compromised / MitM Eavesdrop Capability Impersonation Capability PaAssdivve. CKA 1 1-2 msg. 1-2 msg. TLS 3285 1 hour of data 2 years AActdivve. CKA 1 2 years of data 2 years TLS 3285 1 hour of data 2 years Table 1: Simple example scenario, comparison under session state compromise. As seen, active or passive adversarial behavior immediately the time of compromise can have a drastic effect on the attractiveness of using a CKA protocol. Under a passive attacker, PCS in a...
