Re: CONFIG_RANDOM (compromise?)

Paul Gortmaker (gpg109@rsphy6.anu.edu.au)
Fri, 24 May 1996 03:29:57 +1000 (EST)


- From Theodore Y. Ts'o (tytso@mit.edu) Thu, 16 May 1996 18:58:48 -0400

> Well, at some level I'm assuming that Linus will have enough sense that
> he won't allow something stupid to go into the mainline kernel.

> Putting in a light version of /dev/random is an extremely bad idea,
> because now the application won't know whether or not /dev/random is
> really secure, and so applications won't use it.

Please note that I didn't do something so stupid as to indroduce a
weak /dev/random in my patch, nor would I advocate such an idea either.

> On a 4 megabyte machine, that's 0.4% of memory. On a 16 megabyte

A minor nit, but it is actually 0.4% of 4096kB, which is not really
a meaningful number. On a typical machine, you lose about 600kB code,
400kB data, and 384kB reserved. Which knocks you back to about 2.5MB
of memory on a 4MB box, before you even consider any dynamically
allocated kernel stuff. Which still leaves the issue of wasted CPU
cycles, if the device simply isn't being used.

- From Martin.Dalecki (dalecki@namu23.Num.Math.Uni-Goettingen.de)
Fri, 17 May 1996 10:14:27 +0200 (MET DST)

> mainly for nummerical prgramming, tex and X11 as an xterm, launcher. No
> one of those uses System V IPC. OK /proc can't be disabled anymore by
> this time.

Wrong. You can still compile without /proc -- even with networking
enabled. (I haven't tried to compile with networking disabled tho)

Oh but why would you want to build without /proc or IPC, right?
Think about setting up a teaching/student lab with 30 or more low
budget PCs running simply as xterminals. Not an uncommon scenario.
Now think what the kernel on these boxes really requires built in.

Paul.