On Tue, Nov 27, 2018 at 4:10 AM Nico Huber nico.huber@secunet.com wrote:
Hi Grant,
I don't know how it is supposed to work on Haswell, but can give you some pointers anyway.
tl;dr I don't think you are looking for a PCI device.
If intel-rng support is required, a PCI device will be advertised because of how the linux kernel binds PCI devices to drivers. See "alias" field of "modinfo intel-rng" as an example.
One of the search results pointed at a response that said "load intel-rng driver" as the solution to this problem... that's the only reason I'm exploring this path.
Am 27.11.18 um 08:11 schrieb Grant Grundler:
Asus Chromebox (Panther) with Celeron 2995U processor is supposed to have a HW Random Number Generator: https://ark.intel.com/products/75608/Intel-Celeron-Processor-2955U-2M-Cache-...
(Intel calls it Secure Key)
But "modprobe intel-rng" is failing with "No such device" (Debian 4.18.0-2-amd64 kernel).
This driver is for very old Firmware Hub (FWH) hardware which would be controlled through the LPC PCI device. You have such a PCI device (00:1f.0) but there's no FWH to be expect with Haswell.
Hrm. OK. But that would explain why intel-rng driver only binds with PCI devices.
What you are probably looking for is the RDRAND instruction. I don't know if it can be controlled by the firmware, but would check first if your OS is prepared to make use of it.
Linux kernel has supported RDRAND for a long time. There is even a public debate about *excluding* RDRAND use since some people were hypothesizing that RDRAND was "compromised" by Intel so "goverment agencies" could break encrypted traffic which used RDRAND exclusively to generate encryption keys. Linux kernel does NOT exclusively use RDRAND and Ted Tyso made compelling arguments that RDRAND would still add "entropy" to key generation.
What I don't know is how linux figures out it can or should use RDRAND. RDRAND appears to be a "CPU feature":
arch/x86/include/asm/cpufeatures.h:#define X86_FEATURE_RDRAND ( 4*32+30) /* RDRAND instruction */
And as notedin original email, Intel says this CPU (Celeron 2995U) supports "Secure Key" which is the new marketing name for HW RNG support (could be only via RDRAND now).
Why do I care about HW RNG? Because of this: ... [ 8.560270] r8169 0000:01:00.0 enp1s0: link up [ 8.560287] IPv6: ADDRCONF(NETDEV_CHANGE): enp1s0: link becomes ready [19039.712644] random: crng init done [19039.712649] random: 7 urandom warning(s) missed due to ratelimiting [19044.485625] wlp2s0: authenticate with ... ...
Yes, several *hours* until the crng was initialized and then wpa_supplicant could start talking on WIFI. :(
The length of the delay varies...shortest was 7 minutes.
Well, even without a hardware rng, I wouldn't expect that.
Exactly. I didn't either. My NUC5 completes typically in 3 second from the time the kernel is loaded. But this is a different CPU (Intel Core i5 6260) and completely different firmware (If Coreboot was available for this, I'd prefer Coreboot).
With antennas available, I would say after 10s for the paranoid there should be enough entropy available. But that's probably just how I'd do OS development (and depends on what the wifi driver can do).
I don't know if the kernel has access to any radios (or antennas) until the 80211 link is brought up... which in turn won't happen until wpa_supplicant is running. So something else is wrong here. My suspicion is still on Coreboot not providing something that tells the linux kernel a quick method to generate random numbers.
I saw Matt DeVillier's response as well and I'll follow up once I've updated the SeaBIOS firmware, installed rng-tools5, and determined which CPU features are advertised by both Panther and NUC CPUs. For some reason my "phone home" (SSH) is getting rejected right now. :(
cheers, grant
Nico