Barking up wrong tree? (was Re: buildworld problem on cyrix 166)

[Charles Swiger]

On Apr 28, 2004, at 12:07 PM, John Mills wrote:
[ ... ]

I seem to recall that the earlier Cyrix processors had hardware flaws 
(particularly with their L2 cache?) which may be causing your problems, 
although the kernel config file mentions workarounds for specific 
models which might help.

> Q1) Alternative possible cause: I once had a [non-fBSD] problem where a
> linking a large library required a larger amount of "live" RAM than I 
> had.
> Swap space wouldn't do. As in this case, it always failed at the same
> point because the GCC linker couldn't lay out the required image in my
> limited RAM. My "sandbox" fBSD box is definitely 'RAM-challenged' 
> (49MBy
> if I remember). Could this happen on a 'buildworld' step? Specifically 
> my
> 'buildworld's were failing around 'libncurses'.

FreeBSD has working virtual memory.  Unless a program goes out of it's 
way to obtain locked pages (ie, uses mlock()), this should not be an 
issue.

If you have another machine available with more memory, perhaps use 
that to compile on, and just perform the installation step onto your 
sandbox...

> Q2) Implications of problem: I was rebuilding my kernel [RELENG_4_8 on 
> the
> last go-round] in case there had been security fixes in the code since 
> the
> CD images were released, not because of any observed problem. Is this a
> real concern? If so, is there a source of up-to-date 'generic' builds I
> could use instead, or some more brain- and less RAM-intensive approach?

There have been meaningful security advisories since 4.8 was released, 
yes, although what comes to mind would require updating userland and 
not just the kernel (ie, OpenSSL).

4.9 is available now, obviously, and 4.10 will be out shortly.  I seem 
to recall that there also is a snapshot FTP server with intermediate 
builds available as well, but I don't recall a more specific pointer to 
it now.

-- 
-Chuck