Re: [Qemu-devel] virtio-rng and fd passing

[Anthony Liguori <anthony@codemonkey.ws>]

Peter Krempa <pkrempa@redhat.com> writes:

> On 03/01/13 21:04, Anthony Liguori wrote:
>> Eric Blake <eblake@redhat.com> writes:
>>
>>> Stefan Berger and I discovered on IRC that virtio-rng is unable to
>>> support fd passing.  We attempted:
>>>
>>> qemu-system-x86_64 ... -add-fd set=4,fd=34,opaque=RDONLY:/dev/urandom
>>> -object rng-random,id=rng0,filename=/dev/fdset/4 -device
>>> virtio-rng-pci,rng=rng0,bus=pci.0,addr=0x6
>>
>> Why are you using th rng-random backend instead of the rng-egd backend?
>
> There are two issues with using the egd backend for unprepared devices:
>
> 1) The backend spits out "\x20\x40" commands(egd blocking entropy 
> request for 0x40 bytes) so it really has to be used with some kind of 
> EGD server implementation otherwise it might feed your /dev/random with 
> predictable bytes if used directly.

Yes.  That's the point.

>
> 2) performance of the egd backend is terrible as I've reported here
> https://bugzilla.redhat.com/show_bug.cgi?id=915381 (yes I'm aware that I 
> probably should have filed a upstream bug too, but I was hoping Amit 
> would do it in the process)
>
> On my machine I managed to do 0.2KiB/s with the egd backend both with 
> using constant data as a high performance source, but also with a true 
> random number generator (in the Raspberry pi SoC, sources 107KiB/s of 
> entropy). The rng-random backend performs a bit better averaging
> 1.2MiB/s.

That's a bug that I wasn't aware of until you wrote this.  I'll look
into it next week.

Regards,

Anthony Liguori

>
> Peter