From: Sune Kirkeby <[EMAIL PROTECTED]>
Subject: Re: entropy-0.2 [Was: Re: /dev/{,u}random driver]
Date: Mon, 12 Jun 2000 02:14:23 +0200
> I was thinking of implementing only the bare-bones in-kernel, probably
> just providing the raw data (inter-interrupt timings, etc) to
> priviledged user-space processes. Then it should be a simple matter to
> make my current implementation grab it's input from the kernel.
That sounds nice to me. Perhaps the kernel should have an "entropy"
device like "kmesg", which just stores random source messages in a
queue. That is a cleaner way.
But I don't think it's very bad to have the whole code in the kernel
itself, since the code size is negligible and we will steal the code
from Linux anyway. It is a problem on aesthetic sense rather than a
technical deficiency. In fact, IIRC, gnumach-char should have already
supported a Linux-compatible random device driver. So you may use it
whenever you like, unless you want to remake the driver.
IMO, a more serious problem is that the current situation about GNU
Mach is somewhat confusing. There are at least three variants, that
is, gnumach-1.2 and later which have glue code with Linux's block and
net drivers, gnumach-char which has glue code with Linux's char
drivers additionally, and oskit-mach which doesn't have the glue code
natively but uses OSKit instead. So it is difficult for contributors
to decide on which Mach they should work. Apparently, Roland and/or
Thomas should announce to which direction the maintainers are planning
to go. Even I don't know what they think.
Okuji
