[Freifunk-Bonn] Q: haveged on MIPS-Systems vs. Cycle-Count == 0

[Gary Wuertz]

Jan,

Haveged is based on the www.irisa.fr havege project but is different in 
many ways. One such difference is that haveged will use use the 
clock_gettime() system call in circumstances where a hardware timer such 
as rdtsc cannot be used (such as MIPS). Since the resolution of 
clock_gettime() varies widely from implementation to implementation, the 
internal self test feature ensures that the timer resolution is 
sufficient for haveged output to pass the built-in version of the AIS-31 
statistical test suite - by default, AIS-31 procedure A and procedure B 
are run at initialization, and AIS-31 procedure B is run continuously 
for all subsequent output. This means for haveged version > 1.5, all 
haveged output fed into /dev/random has an empirical entropy estimate 
(Coron's method) of > 7.967 bits / byte and passes other AIS-31 tests or 
haveged throws a fatal error.

As the upstream, I don't know exactly what you have on hand. The best 
way to determine what you have is to execute haveged with additional 
arguments -r1 -v1

Note that for MIPS versions less that 1.7c, internal self test is broken 
(various data alignment problems).

AFAIK, there is no haveged mailing list. I will be glad to answer any 
other questions you have.

-Gary
On 10/09/2013 4:25 AM, Jan Lühr wrote:
> Hello,
>
> sorry for mailing you in person - I'ven't been able to find any mailinglist on
> haveged so far.
>
> We're using haveged on OpenWRT / mips in order to generate VPN keys during
> boot. Afaik havege takes the CPU cycle cout into account for accessing
> entropy:
> "During an initialization phase, the hardware clock cycle counter of the
> processor is used to gather part of this entropy"
> (http://www.irisa.fr/caps/projects/hipsor/)
>
> Are you aware of: https://lists.openwrt.org/pipermail/openwrt-devel/2013-
> September/021318.html ?
> Is there an impact on haveged on OpenWRT / MIPS?
>
> Thanks,
> Jan