@NicHartley Naturally I don't need an exhaustive list. An answer providing a single example of low-hanging fruit would suffice, although a more detailed analysis would be nice too. — forestJul 2 at 21:06
Basically I'm looking to see if my hypothesis is correct (that untrusted input passed through DMA buffers would allow a DMA-capable but IOMMU-restricted device to compromise the kernel).
@MechMK1 It's not always about bias. If the RNG is seeded using the current time, which is common for low-quality RNGs, then the keyspace, no matter how many characters are in the password, is severely limited, enough that exhaustive search becomes practical.
@Xander That isn't true. Rainbow tables are still useful for certain ciphers, like A5/1.
Also, rainbow tables are still valuable for WPA2, since the ESSID is used as the salt.
And, as we all know, a lot of people just use the default name "linksys" as their ESSID. :P
@Xander To put it into perspective, it would take some hundred thousand high-end CPU years to break A5/1, but it takes a negligible amount of time with a several TiB rainbow table.
