prng: how to seed the random number generator?

[OlegHahm]

As @haukepetersen stated in today's meeting, this problem is two-fold:

1. How to multiplex the PRNG initialization for different sources for the source (and boards without a hardware PRNG)?
2. How to generate a source for randomness for boards without a hardware PRNG

(edit: the hardware RNG isn't "pseudo-" ;-) /martine)?

[miri64]

I implemented something like that a while back and opened a PR for it now: #5153.

[miri64]

(it offers a solution for both problem parameters).

[miri64]

Fixed issue description.

[miri64]

A value read with DEV_RANDOM_READ can then be used to initialize the random number generator in auto_init or somewhere like that.

[kaspar030]

There's also #4294. It can use drift between two crystals when there are no "natural" random sources.

[miri64]

We could include that into #5153. I still have a bad feeling about your spiking of the PNRG there.

[kaspar030]

Spiking?

[miri64]

The introduction of entropy into the result chain of the PNRG you did there.

[jnohlgard]

Some recent developments on Linux can be relevant here as well:

• http://phoronix.com/scan.php?page=news_item&px=LRNG-dev-random-V2
• http://www.chronox.de/lrng/doc/lrng.pdf
• http://lkml.iu.edu/hypermail/linux/kernel/1604.3/00029.html

[stale]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. If you want me to ignore this issue, please mark it with the "State: don't stale" label. Thank you for your contributions.

[maribu]

@Einhornhool could maybe take a look at this. Out of my had I would say: Use puf_sram or a true hardware RNG.

[Einhornhool]

For multiplexing different sources, the mbedTLS entropy module could be an option (#15671).
As a source of randomness on boards without a HWRNG, puf_sram can be used (like @maribu said).

[maribu]

Looking in the code (auto_init_random() in sys/random/random.c), puf_sram or a hardware RNG are already used to seed the PRNG, when available. The fallback is to used luid_get() (which would at least result in different boards using different seeds) or a default constant.

I guess this means we should make try to get puf_sram supported for all boards and everything will be fine.