sys/auto_init: cpuid as seed for random_init

[cgundogan]

Choose a sane seed number instead of defaulting to 0. This PR utilizes the cpuid, if present.

Rationale: Trickle timers rely on random offsets to generate increasing intervals. This way, if a network protocol relies on trickle timers, packets of different nodes will not be sent at the same time, but with a little offset applied.

Currently, in case of RPL, many nodes send out DIO messages at the same time, because the "randomness" is the same for all nodes (same seed, and booted at same time).

[cgundogan]

I am aware of #4294, #4989, #5150 and #5153. My solution here is more or less a temporary "quick-fix" (given that this discussion is several years old now), hence the ugly label.

[OlegHahm]

IMO it makes sense to have initialize the PRNG with something that is at least different on each board.

[kaspar030]

Well, if a CPUID is available, this is the way to go. Simple, and for crypto safety we'd need sth else anyways.

thoughts:

• AFAIR random_init() is available for all prngs, so should we initialize them all?
• why are you adding up the cpuid? wouldn't be using the first four bytes directly, then XOR'ing the rest, result in slightly more distributed numbers?
• I'd prefer to just call random_auto_init() in auto_init.c and move the code to sys/random/auto_init_random.c, or whatever location seems suitable.

[jnohlgard]

@kaspar030 why would xor be more random than add? Addition is only an xor with carry.

@cgundogan I would suggest that you concatenate some bytes or integrate some bit shifts to get to the full width of the seed value, otherwise the upper 20-something bits will always be zero for a 32 bit seed value. Another option would be to use one of the functions provided by sys/hashes and sys/checksum.

[miri64]

@cgundogan how about using #5153 instead to get a random value and add CPUID as a fallback random value?

[kaspar030]

@cgundogan Could you factor out the code from auto_init somewhere into sys/prng, so auto_init.c gets less cluttered? Also maybe call it for every PRNG, not just tinymt.

[kYc0o]

PRs about RNG should be postponed, since there are several addressing similar issues.

[miri64]

Postponed due to feature freeze

[OlegHahm]

@cgundogan, what's up?

[PeterKietzmann]

CPUID is old-school, luid is the thing (see #7547 for example)! @cgundogan thanks for your initial input on this!