Intel RNG API?

[R.D. Silverman]

Does VC++ provide any kind of API or system call that allows
one to access the hardware RNG in an Intel i3/i5/i7 processor?

Does one need the Microsoft Crypto API (CAPI) or the Next Generation
API (CNG) to do this?

[fivemack]

It looks as if CryptGenRandom from the cryptography API might do what you want ... the random number generator is only really accessible from kernel space (I've had a look at the Linux code for collecting random numbers, which does all sorts of weird PCI I/O operations).

http://msdn.microsoft.com/en-us/libr...(v=vs.85).aspx is the CryptAcquireContext call which you need to make in order to get a handle which you can pass to CryptGenRandom.

[retina]

Originally Posted by fivemack

... the random number generator is only really accessible from kernel space ...

Are you talking about the RDRAND instruction? Because that is available from user space, no special permissions required.


http://spectrum.ieee.org/computing/h...er-generator/0

[R.D. Silverman]

Originally Posted by retina

Are you talking about the RDRAND instruction? Because that is available from user space, no special permissions required.


http://spectrum.ieee.org/computing/h...er-generator/0

It appears though that RDRAND is not yet available; It seems to require
the (yet to be released) Ivy Bridge processor.

What about the intrinsic:


_rdrand_u32()

Is this available under VC++? What .dll or .lib needs to be linked?

[xilman]

Originally Posted by R.D. Silverman

It appears though that RDRAND is not yet available; It seems to require
the (yet to be released) Ivy Bridge processor.

What about the intrinsic:


_rdrand_u32()

Is this available under VC++? What .dll or .lib needs to be linked?

Lets step back a little.

What problem are you trying to solve?
How good, according to appropriate metrics such as speed, portability, and estimated entropy, does your RNG need to be?

Given the answers, we may be able to give more precisely targeted advice.

Paul

[R.D. Silverman]

Originally Posted by xilman

Lets step back a little.

What problem are you trying to solve?
How good, according to appropriate metrics such as speed, portability, and estimated entropy, does your RNG need to be?

Given the answers, we may be able to give more precisely targeted advice.

Paul

I just want to get/sample entropy bits from the h/w RNG; Speed is not an
issue in case I need to implement a Von-Neumann unbiaser. I will make as
many calls as I need to get the require number of entropy bits. Of course
I will not use them directly as a source of randomness; they will be passed
through a non-linear filter before use.

[fivemack]

Originally Posted by R.D. Silverman

What about the intrinsic:

_rdrand_u32()

Is this available under VC++? What .dll or .lib needs to be linked?

That's just a synonym for RDRAND without writing inline assembly, so again you need to wait for Ivy Bridge.

[fivemack]

Originally Posted by R.D. Silverman

I just want to get/sample entropy bits from the h/w RNG; Speed is not an
issue in case I need to implement a Von-Neumann unbiaser. I will make as
many calls as I need to get the require number of entropy bits. Of course
I will not use them directly as a source of randomness; they will be passed
through a non-linear filter before use.

The CryptAPI CryptGenRandom does the entropy collection (from a variety of sources) and non-linear filtering for you; why do you want to do it yourself? If you want to read the motherboard entropy source directly, you need a dedicated kernel driver; similarly if you get a USB entropy key like http://www.entropykey.co.uk/ you'd need a driver to get it to work under Windows.

[jasonp]

Originally Posted by fivemack

The CryptAPI CryptGenRandom does the entropy collection (from a variety of sources) and non-linear filtering for you; why do you want to do it yourself?

Has the CryptoAPI generator improved since this was published?

[Christenson]

Originally Posted by jasonp

Has the CryptoAPI generator improved since this was published?

As of 4:45PM ET today, that server is offline....what was in it? A proof that it wasn't very high entropy at all?

[CRGreathouse]

Originally Posted by Christenson

As of 4:45PM ET today, that server is offline....what was in it? A proof that it wasn't very high entropy at all?

From the Internet Archive:

3.9 The CryptoAPI Generator
Like the SSLeay/OpenSSL generator, the CryptoAPI generator uses separate functions for mixing in entropy and for the PRNG. The entropy-mixing stage hashes polled data using SHA-1, and the PRNG stage uses the SHA-1 output to key an RC4 stream cipher, as shown in Figure 15. The polled data consists of the traditional time and process ID and a few relatively poor additional sources such as the system’s current memory and disk usage, the static data consists of the previous hash output recycled for further use [44]. Unlike PGP, the CryptoAPI preserved state doesn’t appear to be protected by a user password.
This generator is, at best, adequate. The static data isn’t password-protected like the PGP seed and the polled data doesn’t provide much entropy, particularly if the generator is seeded at system startup, which is typically the case as Windows system components make use of CryptoAPI during the boot phase. In addition no steps appear to be taken to mitigate problems such as the fact the RC4 exhibits correlations between its key and the initial bytes of its output [45][46][47] as well as having statistical weaknesses [48][49][50], resulting in the generator leaking part of its internal state in the first few bytes of PRNG output or producing some predictable output during its operation.