This weekend I decided to upgrade from my trusty 2.4.16 + preempt to
something a bit more recent. The first thing I tried was
2.4.18-pre3 +
O(1) + preempt + lock-break,
which fell over miserably on my system (eg if it didn't panic at boot I
could consistently cause a panic with a gmake -j5). After
some "testing" on
IRC with
Robert Love (ie, boot, wait
until panic and hard lock, write entire panic on piece of paper, reboot
into stable kernel and type up panic, post panic to web for Robert to
run ksymoops on, apply patch to kernel, reboot, repeat), I finally
decided to take a break from
preempt
for the week and give low latency a shot. I'm running
2.4.17
+ rmap +
O(1) + low latency now, and it
seems to be pretty stable. I'll try out preempt + lock-break again
this weekend and let everyone know what happens. By the way, the
problems I mentioned above with preempt have to do with SMP and
preemption near interrupts -- all you single-processor users have
nothing to fear.
There has been a really amusing thread on the LKML for the last few weeks regarding automatic kernel build configuration. The thread started as a question from ESR about probing the ISA bus, and progressed (devolved?) into an amusing war of metaphors between Alan Cox and ESR. Essentially, ESR used a hypothetical "Aunt Tillie" to represtent the typical technology-impaired "mom and pop" Linux user, who, for some unspecified reason, opts to compile her own kernel instead of using the modular vendor-supplied, vendor-tested, and vendor-supported version. After the Aunt Tillie scenario made relatively little progress against Alan Cox and a handful of other people, ESR added "Nephew Melvin" to the list of hypothetical characters. Anyway, Alan Cox's latest contribution to the argument:
From: Alan Cox <alan@aunt-tillie.org> ... pabs@gw-uunet:~> whois aunt-tillie.org ... Registrant: Melvin? What do I type here? 10 Alan Turing Road Surrey Research Park Guildford, Surrey GU2 7YF UK
Be sure to check out http://www.aunt-tillie.org/ as well.