This mail is related to use of newer kernels on single-cpu systems.
I don't address the reported problems related to networking in this mail.
Most of you wrote messages to linux-kernel regarding problems with
linux kernel 2.1.34 or higher,
so I write this to all of you because *-some-*, but not neccesary *-all-*
of your problems *-may-* be caused by running linux >=2.1.34 on a single
processor machine without an commented in "SMP = 1" line in the toplevel
makefile.
There have been made provosions to make __SMP__ to be a
config option CONFIG_SMP for "make config" and I think it's the time to do
so, and I will send a patch to linus.
Some of the expiriances were(You don't need to have all of them):
* Swap is not used at all.
* fork's failing due to not enough memory
* "top" doesn't show any correct cpu times, ever 100% idle, 0% busy
* Machine very slow at swapping with high-cpu tasks
* weird multi tasking when high-cpu task are running -> "slow performance"
* And fillally some solid locks.
I personally noticed this weird SMP on my i486 with on all kernels I tried
from 2.1.34 up to 2.1.38(I didn't get further yet, but will).
These problems went away as soon as I commented you the SMP = 1 line in
the makefile, and there has been some reports that others seen the same.
Linus submitted a nice mail to linux-kernel regarding the 2.1.34 kernel,
which I append to this mail, and probably one of the SMP-changes in it
show up the bug.
At my machine it looks like the timer-interrupt looses it's function to
count the process's times(top) and allow the sheduler to work.
This happens when a process like gcc or pine needs all cpu to compile or
sort a huge mail folder. It also looks like kswapd isn't working at this
time because when the system needs to swap for this process, it looks very
inefficient.
I'll try to put something into the timer interrupt which I can monitor to
verify it's function.
Bernhard
-- Bernhard Kaindl | email: edv@bartelt.via.at Bartelt Labor- & Datentechnik | bkaindl@netway.at Neufeldweg 42 | phone: +43.316.475328 A-8010 Graz, Austria | fax: +43.316.475328-43Linus wrote to linux-kernel:
>Subject: Linux-2.1.34 - the bleeding edge (fwd) > > Ha! > Do you want to be a _real_ man (or woman), and run interesting kernels > rather than those wimpy 2.0.x things? > > Now you can. What would you pay? 19.95? 14.95? Or even just 9.95? NO! > You can get a RealMan(tm) or RealWoman(tm) or even a RealThing(tm) > kernel for ABSOLUTELY FREE even if you don't fill in any subscription > information! (Void where prohibited by law. Entrants should be 5 years > or older or accompanied by a responsible adult. Only as long as > supplies last.). > > The new and improved 2.1.34 kernel is out there for grabs, and > implements such stunning new features as > > - on SMP you actually finally get _correct_ timing information for your > process (ie system/user times should actually be real, rather than > just guesses). > - the new interrupt architecture starts to work on sparc too (no > guarantees). > - the networking fixes from the boring^H^H^H^H^H^Hstable 2.0.x series > got upgraded. > - "intr_count" got terminated. VERY terminated. > > Anyway, what makes this kernel so extra interesting is that some of the > re-organizations probably means that it doesn't really compile very well > on non-x86 architectures, but you should not see this as a mis-feature > but as a challenge. When your friends are all running the 2.0.x > kernels, and you tell them you're running 2.1.34, see their eyes widen > in surprise at your daring and at your skills in actually being able to > compile it. > > You know that hot chick/hunk/thing/whatchamacallit down the street that > you've just been dying to impress? Here's your chance. > > Where it it, I can hear you ask? ftp.kernel.org, ftp.funet.fi, > ftp.cs.helsinki.fi and any other well-stocked ftp site in your > neighbourhood. > > Linus "v2.1.34 - the _cool_ kernel" Torvalds >