Improper Protection Against Voltage and Clock Glitches

Put the processor in an infinite loop, which is then followed by instructions that should not ever be executed, since the loop is not expected to exit. After the loop, toggle an I/O bit (for oscilloscope monitoring purposes), print a console message, and reenter the loop. Note that to ensure that the loop exit is actually captured, many NOP instructions should be coded after the loop branch instruction and before the I/O bit toggle and the print statement. Margining the clock consists of varying the clock frequency until an anomaly occurs. This could be a continuous frequency change or it could be a single cycle. The single cycle method is described here. For every 1000th clock pulse, the clock cycle is shortened by 10 percent. If no effect is observed, the width is shortened by 20%. This process is continued in 10% increments up to and including 50%. Note that the cycle time may be increased as well, down to seconds per cycle. Separately, the voltage is margined. Note that the voltage could be increased or decreased. Increasing the voltage has limits, as the circuitry may not be able to withstand a drastically increased voltage. This process starts with a 5% reduction of the DC supply to the CPU chip for 5 millisecond repeated at 1KHz. If this has no effect, the process is repeated, but a 10% reduction is used. This process is repeated at 10% increments down to a 50% reduction. If no effects are observed at 5 millisecond, the whole process is repeated using a 10 millisecond pulse. If no effects are observed, the process is repeated in 10 millisecond increments out to 100 millisecond pulses. While these are suggested starting points for testing circuitry for weaknesses, the limits may need to be pushed further at the risk of device damage. See [REF-1217] for descriptions of Smart Card attacks against a clock (section 14.6.2) and using a voltage glitch (section 15.5.3).

During the implementation phase where actual hardware is available, specialized hardware tools and apparatus such as ChipWhisperer may be used to check if the platform is indeed susceptible to voltage and clock glitching attacks.

Review if the protections against glitching merely transfer the attack target. For example, suppose a critical authentication routine that an attacker would want to bypass is given the protection of modifying certain artifacts from within that specific routine (so that if the routine is bypassed, one can examine the artifacts and figure out that an attack must have happened). However, if the attacker has the ability to bypass the critical authentication routine, they might also have the ability to bypass the other protection routine that checks the artifacts. Basically, depending on these kind of protections is akin to resorting to "Security by Obscurity".

Many SoCs come equipped with a built-in Dynamic Voltage and Frequency Scaling (DVFS) that can control the voltage and clocks via software alone. However, there have been demonstrated attacks (like Plundervolt and CLKSCREW) that target this DVFS [REF-1081] [REF-1082]. During the design and implementation phases, one needs to check if the interface to this power management feature is available from unprivileged SW (CWE-1256), which would make the attack very easy.