2.6.5-love4 aka "Dum Diddly Doo"

Message

zerojay · Post by **zerojay** » Mon Apr 12, 2004 5:44 am

After manually patching the kernel sources and compiling, it's working well here too. I used -Os for my kernel cflags.

MG-Cloud · Post by **MG-Cloud** » Mon Apr 12, 2004 6:06 am

(for example compiling kernel on background with web surfing/mp3 takes twice the time to complete, heavy disk load gets the machine to its knees...)

I've also been experiencing these problems. I thought it was just a performance deficiency of Linux as a whole rather than attributing it to love-sources (I just started using linux fulltime in late 2003/early 2004, and love-sources has been my kernel since I've started. I was only using a kernel without nick's scheduler for a short amount of time before). This will be interesting to experiment with (eg the lack of pre-emptable kernels).

However, what's the official word on this (steel? lovechild?)? Is there any reason why perhaps nick's scheduler will interfere with preempt to the point where two things that are supposed to have drastic benefits on a system's performance level can slow it down? Or have I missed earlier posts that say to disable "preemptible kernel" with nick's scheduler?

I'm a little concerned here because the ability to have a preemptible kernel was one of the major aspects that 2.6 was touted for.

Thanks =)

EDIT: Also, I've just recently installed xorg-x11. To renice X, wrapper.c should be changed to launch /usr/X11R6/bin/Xorg instead of /usr/X11R6/bin/XFree86, correct?

ed0n · Post by **ed0n** » Mon Apr 12, 2004 7:31 am

It works here.

MG-Cloud · Post by **MG-Cloud** » Mon Apr 12, 2004 7:42 am

ed0n wrote:It works here.

Hmm... interesting. I may be experiencing something of an isolated case then

Just to see if there's anything in common with the other people who posted problems with preemptible kernels and Nick's scheduler - I'm running an AMD Athlon XP 1800+ (~1.53GHz) with 768mb of pc2100 ddr ram... epox 8kha+ motherboard, msi gf2ti video card (UGH), Maxtor 7200rpm 60gb hard drive, sblive!.

dub.wav · Post by **dub.wav** » Mon Apr 12, 2004 7:47 am

Here's a bash script I wrote to automatically download the latest love-sources patch. For the ones using ebuilds, it should be easy to extend it to download the ebuild, copy it to $PORTDIR_OVERLAY, and digest it.

Code: Select all

#!/bin/bash
love_url="http://www.public.iastate.edu/~jpcox"

for patchdir in patches ../patches
do
	[ -d "$patchdir" ] && cd "$patchdir" && break
done

set -- `wget -q -O - "$love_url/?M=D" \
| grep "HREF" | grep ".bz2" | head -n1`
file=${6/>*/}
file=${file/'HREF='/}
file=${file//\"/}
if [ ! -f "$patchdir/$file" ]
then
	wget "$love_url/$file"
	echo
	echo "$file downloaded."
else
	echo "$file already exists."
fi

Pink · Post by **Pink** » Mon Apr 12, 2004 8:04 am

Everything works for me except vmware, can't compile the module.

Ati compiled in just fine and dandy - Have to go back to love3 as i can't do without vmware (damn you university assignments!).

We do appreciate your work Steel...

VolcomPimp · Post by **VolcomPimp** » Mon Apr 12, 2004 10:06 am

just did a stage 1 again and it runs fine until I get to compiling the ATI drivers like everyone else... I tried 1 or 2 of the newest (the ones that
d/l in the rpm) as well as a couple old ones which were from a guide that
I used 2 get 3d acceleration working in the past....
errors on the old ones and the new ones seem to install seemingly fine
until you try and use the module in which its nowhere in sight.

- Added the DRM development tree to -mm kernel. Please Cc
dri-devel@lists.sourceforge.net on any bug reports.

looks to me like an -mm problem...

sindre · Post by **sindre** » Mon Apr 12, 2004 12:17 pm

I disabled preempt since the loss shouldn't be too big anyway. Preempt doesn't help worst-case latency which is the real issue.

Zviratko · Post by **Zviratko** » Mon Apr 12, 2004 12:36 pm

I got vmware and ndiswrapper to compile!!

Replaced scripts/Makefile.modpost and /usr/src/linux/Makefile with respective files from -love3 and everything works....

I recommend switching them back if you want to recompile your kernel though

PrakashP · Post by **PrakashP** » Mon Apr 12, 2004 12:56 pm

@sindre

Actually it is what preempt is all about: It should make worst case better, though making average and best case a lite bit worse due to overhead.

Zviratko · Post by **Zviratko** » Mon Apr 12, 2004 1:05 pm

nope, preempt DOES NOT improve worst case latency - just smoothes all latencies and reduces average

PrakashP · Post by **PrakashP** » Mon Apr 12, 2004 1:08 pm

Hmm, I thought I read something about this in lkml a month or so ago. But I could be mistaken...

sindre · Post by **sindre** » Mon Apr 12, 2004 1:15 pm

Andrea Arcangeli wrote:"...keep preempt turned off always, it's useless..."
"...preempt just wastes cpu with tons of branches in fast paths that should take one cycle instead..."
"...Takashi Iwai did lots of research on the preempt vs lowlatency and he found that preempt buys nothing..."
"...infact 2.4-aa has a lower max latency than 2.6 stock with preempt enabled..."
"...Sure, I agree it improves average latency, the problem is that there are nearly no application that cares about average, what matters is the worst case latency only..."
"...PREEMPT=y doesn't and cannot improve the worst case latency. This is true today like it was true 4 years ago..."

Adrew Morton wrote:Preempt is overrated. The infrastructure which it has introduced has been useful for detecting locking bugs.

It has been demonstrated that preempt improves the average latency. But not worst-case, because those paths tend to be under spinlock.

Robert Love wrote:Yes, the absolute worst case latency probably remains because it tends to occur under lock.

Quotes taken from this kerneltrap-story.

Beware that there are also comments in favour of preempt which I haven't quoted.

ktech · Post by **ktech** » Mon Apr 12, 2004 1:22 pm

Zviratko wrote:I got vmware and ndiswrapper to compile!!

Replaced scripts/Makefile.modpost and /usr/src/linux/Makefile with respective files from -love3 and everything works....

I recommend switching them back if you want to recompile your kernel though

For those of you that cannot compile drivers (ati-drivers, vmware... etc), this solution works.

Only don't forget tu edit Makefile and change love3 to love4 after doing that changes.

PrakashP · Post by **PrakashP** » Mon Apr 12, 2004 1:25 pm

@sindre

Ok, you got me then.

Well, I'll try kernel without preempt then.

cold_flame · Post by **cold_flame** » Mon Apr 12, 2004 1:59 pm

any reason why lirc patch was dropped?

darkless · Post by **darkless** » Mon Apr 12, 2004 3:07 pm

Zviratko wrote:I got vmware and ndiswrapper to compile!!

Replaced scripts/Makefile.modpost and /usr/src/linux/Makefile with respective files from -love3 and everything works....

I recommend switching them back if you want to recompile your kernel though

If everything works for you, then that's great. However, you didn't catch all the related changes.

I made a patch which reverts the relevant patch in -mm4:
http://www.sundebo.dk/dark-sources/reve ... port.patch

Zviratko · Post by **Zviratko** » Mon Apr 12, 2004 3:54 pm

darkless: I did notice there's a patch that actually moves the module to the modpost of kernel make modules_install - that however means that vmware is uninstallable (at least not without modifying the script which is not something an average user is supposed to be doing) - in my opinion, the patch just b0rks a few things - I am suggesting a fix :)

Markus_T · Post by **Markus_T** » Mon Apr 12, 2004 6:25 pm

Working fine here with:
CONFIG_CFLAGS_STRING="-O3 -funroll-loops -fpeel-loops -funswitch-loops"
and gcc 3.4.0

ktech · Post by **ktech** » Mon Apr 12, 2004 8:32 pm

Markus_T wrote:Working fine here with:
CONFIG_CFLAGS_STRING="-O3 -funroll-loops -fpeel-loops -funswitch-loops"
and gcc 3.4.0

OFF-Topic, but... how is gcc 3.4 performing in gentoo?

Markus_T · Post by **Markus_T** » Mon Apr 12, 2004 9:21 pm

ktech wrote:
Markus_T wrote:Working fine here with:
CONFIG_CFLAGS_STRING="-O3 -funroll-loops -fpeel-loops -funswitch-loops"
and gcc 3.4.0
OFF-Topic, but... how is gcc 3.4 performing in gentoo?

AFAICT it is performing well. I'm primarily interested in the new C++ parser, but there are a lot of other improvements too. Visit http://gcc.gnu.org/gcc-3.4/changes.html for further details.

infirit · Post by **infirit** » Mon Apr 12, 2004 10:05 pm

Great work on the best kernel for linux

Any change of reiser4 getting in again?

Evil Dark Archon · Post by **Evil Dark Archon** » Mon Apr 12, 2004 10:48 pm

as soon as it is fixed to a point where it can be applied to a kernel version >=2.6.5-rc*-mm* then it will probably be included in -mm and since both love-sources and evil-sources are based on -mm they will both have reiser4 in them.

rfujimoto · Post by **rfujimoto** » Mon Apr 12, 2004 10:53 pm

I've been using the mm tree since test3 or so, finally decided to give the love-sources a try.

So far so good, thanks for the work

jonnii · Post by **jonnii** » Mon Apr 12, 2004 11:38 pm

worked a treat.

thanks.

2.6.5-love4 aka "Dum Diddly Doo"

lirc?