xx-sources

Gentree · Posted: Fri Sep 03, 2004 5:19 pm Post subject:

Hi,

I recently move over to xx because it seems a lot more consistant in it's support of reiser4 than than most other kernels.

One sad loss is bootsplash. It's a detail but technically but it makes the system look as sharp as it is on the inside.

I've spent quite some time looking into gensplash but cant even get splashutils to compile:

xiphux · Posted: Fri Sep 03, 2004 5:29 pm Post subject:

I could try and make a bootsplash/fbsplash selectable option, but it might take some time.

However, that splashutils compile error is quite simple to fix; I actually posted the solution in the gensplash thread.

Go edit /usr/src/linux/include/linux/fb.h. At line 305, you'll see:

Gentree · Posted: Fri Sep 03, 2004 8:02 pm Post subject:

Thanks xiphux,

that got me through the compile pb.

now to see about the rest of gensplash.

_________________
Linux, because I'd rather own a free OS than steal one that's not worth paying for.
Gentoo because I'm a masochist
AthlonXP-M on A7N8X. Portage ~x86

xiphux · Posted: Sat Sep 04, 2004 5:31 am Post subject:

If anyone's interested, I'm in the middle of committing a tree that's updated to -mm3. High res timers are back, the completely-out-of-line spinlocks are replaced with an architecture agnostic version, there's a little bit of the fast suspend/shutdown and deferred resume stuff, and other little things. I don't know if it's just a placebo effect, but for some reason it feels really good on my comp. (low latency, etc) Then again, very rarely is there an actual noticable improvement; it's almost always just placebo and the person's ego talking. But whatever.

Let's just hope CVS isn't retarded this time.

trevorj · n00b Joined: 13 May 2003 Posts: 68

current cvs does...

Gentree · Posted: Sat Sep 04, 2004 5:02 pm Post subject:

xiphux, thanks for the splashuitls fix.

I posted a bugreport so at least the fix is noted and could help find a more permanent correction. Spock basically says it pb with mm headers not gensplash. I'm not in a position to comment. Here's his reply to the bugreport.

xiphux · Posted: Sat Sep 04, 2004 5:39 pm Post subject:

*sigh*

Regardless of what's right and what's wrong, the fact is that the way modelist is now is in the mainline kernel now, starting with 2.6.9-rc1. Antonio Daplas chose to write the features (mode list as linked list), Linus chose to include it, and we just have to deal with it, plain and simple. We have to change our code accordingly. To adamantly deny that there is anything wrong with your code is just a matter of ego, and if he continues to refuse to update splashutils to deal with the change, then gensplash is going to fall behind.

Unless, of course, you want to go complain to Linus that the fbdev developers are wrong. Personally, I don't see any 'bug' in the way modelist is handled at present. It makes the various handler functions that were written for modelist easier to deal with.

Gentree · Posted: Sat Sep 04, 2004 7:12 pm Post subject:

xiphux · Posted: Sat Sep 04, 2004 7:40 pm Post subject:

I can kind of understand where he's coming from, though. When you finally get something complicated like gensplash working (or vesafb-tng, or any of his advanced fb-related projects), it's hard to admit that you did something incorrectly. But gensplash on the whole is an impressive project, so this is really just very minor.

The only reason I don't exactly prefer bootsplash is that it only works with vesafb for me. I run xserver on radeonfb, so this way, I can have both a splash and my xserver. The -sp version of bootsplash (the one that removed the vesafb-specifics) would display the splash on radeonfb, but would oops and lock up whenever I switched consoles (like, when I would start xserver).

Plus, I have so many test-phase kernels in my boot partition that being able to put the whole thing right into the initramfs makes it one less initrd for me to deal with.

If fbsplash still has incompatibilities after some time (I dunno... probably before or by the time 2.6.9 comes out), then I'll start working on adding bootsplash.

I don't update from portage very much at all. But then again, I use gcc 3.5 so I couldn't merge a lot of the updates even if I wanted to.

I can burn cds and dvds fine using atapi and cdrecord/growisofs. I haven't really tried using the packet writing driver, though; I don't really have a need to. I don't really see why it wouldn't work, though.

Evil Dark Archon · Posted: Sat Sep 04, 2004 8:14 pm Post subject:

with the cvs pull i have, the external modules i install (nvidia-kernel and svgalib) have an undefined symbol pci_find_class error, and they warn of an implicit decleration of pci_find_class.
_________________
This post has been over explained for newb-informing purposes.

Registered Linux user 347334
Abit AV8-3rd eye, AMD Athlon64 3500+ 90nm, ATI Radeon x850 pro

DaMouse · Apprentice Joined: 18 Jul 2003 Posts: 233

https://forums.gentoo.org/viewtopic.php?t=216985

*sigh*

-DaMouse
_________________
I am the dark lord DaMouse I eat souls and wear the dressing gown of evil.

trevorj · n00b Joined: 13 May 2003 Posts: 68

gensplash fails to work for me with cvs. it shows the splash screen right after the splash init script runs, then the screen goes black, comes back 5 seconds later, then this really cool looking fading effect happens with the splash, and then it just leaves the terminal unusable

vesafb-tng works though. I also encountered data loss under this kernel ( thank god I made a backup the day before, which I rarely do to begin with ), so let's hope it was just an old reiser4 partition that suffered from older bugs, and not the kernel

_________________
\\ trevorj

xiphux · Posted: Mon Sep 06, 2004 4:16 am Post subject:

Huh. That's weird. The fbsplash in there is a little old, though, so I might rip it out and re-add the newest version.

The data loss is also weird; I haven't experienced any data loss in a very long time (not counting reiser4's atomicity reverting files). On what kind of scale are you talking about? Just several k or megs, or on a larger scale?

In other news, there's been an experimental patch posted on lkml which will allow forked files on the vfs layer. It's a little hard to imagine... basically you have a directory with X number of files in it, but if you try to access that directory as a file (with cat, whatever), it'll "fork" and access a file inside the directory that you specified (with a symlink to the special path '...'). It's not really the way it should be done, but at least it's a step towards the file/directory abstraction that reiser4 started. And at the very least, it'll give something to play with until more programs support reiser4's plugin setup.

I applied the patch fine; I just need to compile, reboot, and test a little to make sure nothing lights on fire, and then I'll put it into cvs. Should be interesting.

trevorj · n00b Joined: 13 May 2003 Posts: 68

data loss on a scale of quite a few files being lost, and I can't even ls the ones it threw into lost+found ( hence I can't rm them! ) although fsck tells me no errors exist anymore.

here's the result from ls -l :

xiphux · Posted: Mon Sep 06, 2004 1:18 pm Post subject:

Have you tried rebuilding the fs and/or superblock?

Forking wouldn't really break things, because in the directory you have to specifically link the file that you want to '...'. So if you don't bother doing that, then it's just the same as any old directory.

trevorj · n00b Joined: 13 May 2003 Posts: 68

Yes, actually I had to do both to fix the fs. I'll do it again later today.

Ah, I see that now, I must of read it too quickly. Reiser4 really is a step forward for filesystems =)
_________________
\\ trevorj

xiphux · Posted: Wed Sep 08, 2004 7:02 am Post subject:

Well, ixnay on the vfs forks, they didn't work. And I don't know quite enough about filesystem code to fix it.

I'm committing the update to 2.6.9-rc1-mm4 right now. Unfortunately there was a big rework of the fbdev stuff, and vesafb-tng doesn't apply anymore and I'm not quite sure how to fix it. I guess we'll just have to wait for spock.

I actually might get to that bootsplash option sooner than I thought. Something in gensplash went retarded on my computer, so now anytime I use a splash image, when text scrolls on the screen, it doesn't "clear" the text as it goes by. So I'm left with long trails of junk filling up the screen.

The more time I spent fixing gensplash, the more I started to dislike it. I don't like the idea of using entire userspace libraries in/with the kernel (klibc, libpng, libjpeg). The jpeg decompressor in bootsplash should be lighter and faster. While I do see the merit of moving as much of it as possible to userspace (easier to deal with changing kernels), if you think about the amount of stuff being put into the initramfs... that's one _massive_ image.
I also believe that there are ways to put the bootsplash code into the kernel _without_ being intrusive. It's just a matter of doing it correctly.
Plus, I was looking through the fbsplash and bootsplash code (since fbsplash is closely based on bootsplash), and there's some coding in there that I personally think is horrid. (Includes via relative paths, for example... that's a disaster waiting to happen. Plus, kernel headers shouldn't be compiled into userspace programs/libs)

So I think I'll try including bootsplash again alongside gensplash (mutually exclusive, of course). And maybe clean up some of the code nastiness and move code out to be less intrusive, while I'm at it. I guess I'll just have to see what happens.

4nykey · Apprentice Joined: 11 Feb 2004 Posts: 180

xiphux · Posted: Wed Sep 08, 2004 2:45 pm Post subject:

Is there a line that says what function the oops occurred in?

Also, in the stack trace, what are the top couple of lines? (The human-readable parts, I mean)

Gentree · Posted: Wed Sep 08, 2004 5:52 pm Post subject:

xiphux,

I have been using xx for a couple of weeks now , I am still running 2.6.8.1-xx1 since I like to stay one step back from the bleeding edge . (At least until it starts to coagulate a bit!)

Anyway it seems very stable and does most of what I want.

One thing I am starting to notice however is it seems a little less responsive that 'old' 2.6.6-love4 I used to use.

In particular I am playing an mp3 with command line mplayer.

It seems that anything moderately heavy , like a large scan, seems to make the music glitch from time to time.

Do I need to start playing around with schedulers or nicing stuff? Is this normal behaviour?

Thx.

_________________
Linux, because I'd rather own a free OS than steal one that's not worth paying for.
Gentoo because I'm a masochist
AthlonXP-M on A7N8X. Portage ~x86

Gentree · Posted: Wed Sep 08, 2004 6:16 pm Post subject:

PS,

I am now posting from my old love kernel. Damn , I had forgotten just how snappy it was.

No jumping on mplayer , aMule snaps from one tab to another without the slightest delay (this I had got used to waiting 2 seconds for !) no delay in activating any window.

My only pb with this old kernel is that it does not support the newer R4 format and I have my entire portage tree on an R4 partition.

I chose xx for it's solid record on Reiser4 support but I would really like to get back the snappy system responce I had in love-sources if that is possible with a couple of tweeks in xx.

BTW I had configured love to use cfq if that is relevant .

Thanks again.

_________________
Linux, because I'd rather own a free OS than steal one that's not worth paying for.
Gentoo because I'm a masochist
AthlonXP-M on A7N8X. Portage ~x86

4nykey · Apprentice Joined: 11 Feb 2004 Posts: 180

xiphux · Posted: Wed Sep 08, 2004 8:57 pm Post subject:

Gentree, what scheduler did you use?

Also, the cfq in xx is the ionice-supported version, and I was the one who hacked it up to work with the new cfq as well as the runtime selectable schedulers, so I could very well have done something wrong.

So as far as you can tell, does the latency issue seem to be related to cpu usage or disk i/o?

4nykey: You know, now that I think about it, I think I have seen that bug once before. I'm looking into it now.
[edit]
4nykey, could you see if the same thing happens with a plain -mm kernel? Because I can't really think of a change I made that would cause that, but I'm starting to suspect the reiser4 patch.
[/edit]

Gentree · Posted: Wed Sep 08, 2004 10:25 pm Post subject:

Evil Dark Archon · Posted: Wed Sep 08, 2004 10:46 pm Post subject:

hey xiphux, could you please include the new cfq v2 io scheduler, preferably as an option?
_________________
This post has been over explained for newb-informing purposes.

Registered Linux user 347334
Abit AV8-3rd eye, AMD Athlon64 3500+ 90nm, ATI Radeon x850 pro