1.0.3123 and XFree 4.2

[lpnotfr]

If I use the 3123 drivers with XFree 4.2, I experience random crashes of the server (sometimes 5 crashes in 1 minute, sometimes no problems during a whole day). The problems disappear when I use < 3123 drivers (2880 for example)

Does anybody experience the same problems ?

[Sebi]

Same here. Installed it weeks ago, without problems. Now I updated my KDE (and in general did lots of compiling) and have had about 6 crashes in the last two days.

I logged in to the system via ssh and saw that the process state of the X server is "RL", approx. 50% cpu usage; virtual size was huge. Will try the old driver and see what happens. Seems to be a memory leak to me.

Sebi

[lpnotfr]

I downgraded to XFree 4.1 and it's
much better now

[Sebi]

Originally posted by lpnotfr
I downgraded to XFree 4.1 and it's
much better now
Why didn't you keep the older driver release? I did only a short read of the changelog but couldn't see any major changes btw. 3123 and 2880?!

I'll first give mem=nopentium a try as I have a athlon system and check my shmem config.

Will post my results here.

Sebi

[lpnotfr]

That's because I experienced crashes even with 2880 and XFree 4.2

2880 and XF4.1 is a lot more stable here

[bwkaz]

Originally posted by Sebi
the process state of the X server is "RL", approx. 50% cpu usage; RL -- it's currently runnable (normal), and has pages locked into memory (also normal for X -- not for other things, but for X).

virtual size was huge. Also normal for X. On my system (X 4.2.0 from source, 3123 drivers), which doesn't experience these lockups, X's VSZ is ~300MB according to ps.

Although no one has posted their video card, motherboard (and chipset), processor, or kernel version...

Mine: GF4 Ti4200, Biostar (KT333) (don't get this motherboard...), Athlon XP1800+, and 2.4.19 from kernel.org.

[lpnotfr]

You're right. I'll try to be more precise:

Tested with three different 2.4.19 kernel: vanilla, ck13 with preemptible and low latency enabled, ck13 with low latency only.

3123 + XF4.2.1 -> crash
2880 + XF4.2.1 -> crash (less)
3123 + XF4.1.0 -> crash
2880 + XF4.1.0 -> seems stable

GF2 MX, ASUS A7V (KT133), athlon thunderbird 1GHz

One of my friends experience the same crashes with XF4.2.1 and 3123. But just once a day (it happened to me every hour !!!). Equivalent hardware BUT he has a Geforce4.

Same with you: KT133, 2.4.19, XF4.2 and geforce4 -> stable

sebi has a geforce2 and it doesn't work

Seems to me that NVidia now writes its drivers with only the newest cards in mind :(

[bwkaz]

Hmm... my other system, an i815 motherboard, P3-800 processor, with kernel 2.4.19-pre8 (and a bunch of iptables-related patches), and driver version 2960, is stable as well. I haven't used X on that machine for a while, though, so I never upgraded to 3123. It's got a generic Dell-installed GF2 GTS 32MB card.

Of course, I have never used any MX series card, but I'm sure there are people who have, and for whom it has worked...

When you say "crashes of the server", do you mean that it crashes and brings you back to a command-line, or do you mean that it locks up?

If you boot to runlevel 3 (by editing /etc/inittab and changing the 5 in the initdefault line to a 3, then rebooting), then do a startx until you see the crash, is there anything relevant in the XFree86.0.log log file?

[lpnotfr]

I haven't used X on that machine for a while, though, so I never upgraded to 3123. It's got a generic Dell-installed GF2 GTS 32MB card.

Well, it's not a very useful testbed... ;)


If you boot to runlevel 3 (by editing /etc/inittab and changing the 5 in the initdefault line to a 3, then rebooting),

I'm using Debian :)

It really crashes. And badly, there is nothing relevant in the logs.

[bwkaz]

Yeah, you're probably right, but I figured I'd throw that out there anyway. 2960 does work on the thing.

I don't really have any ideas for anything else to look at, either... :(

[Klaus-P]

"
3123 + XF4.2.1 -> crash
2880 + XF4.2.1 -> crash (less)
3123 + XF4.1.0 -> crash
2880 + XF4.1.0 -> seems stable "


Hmmm, yeah and all versions are compiled and linked, etc, with the
same compiler, etc ,...?

I've read about warnings about those issues. Some gcc compiler versions
are not always the best choice also the latest ones.

Try to check different versions and take care of all the other
necessary stuff, header files, etc,....

[lpnotfr]

I'm running debian. Though XFree & libraries & kernel are compiled with the same compiler (gcc 2.95) and use a consistent set of libraries & tools.

I don't know what they use to build the OpenGL driver. I just know that it is built for XFree 4.0.2 (!)

Ho: my friend is using gentoo. He had the same behaviour with a fresh install (all libraries & executables just being built).

[Sebi]

Yesterday I checked my configuration and updated it to the following - without any crashes so far (fingers crossed...;)):

- added "mem=nopentium" to boot parms
- reinstalled the glx driver as I had a "failed to load glx extension" (selfcompiled)
- built agpart as module instead of kernel included, set NVagp to auto -> uses agpart module; had "failed to verify agp usage" before
- reinstalled nvdriver module (selfcompiled)
- added "tmpfs /dev/shm tmpfs..." to fstab and checked shared mem. config

Now my XFree log is clean again and I had no lockups. Had one X-restart, but maybe caused by another program (server killed and immediately restarted).

Config:
- Athlon 1000 FSB 133
- ECS K7VZA (KT 133A)
- 2.4.19 from kernel.org
- gcc 2.95.3
- 384 MB RAM
- Geforce 2MX
- XFree 4.2.1
- nvdriver/glx 3123

All software on my system is compiled by myself using the gcc 2.95.3 as I use linux from scratch system.


Sebi

[Klaus-P]

" I don't know what they use to build the OpenGL driver. "

F.i libgGL.so.1.0.3123 and libGL.core (GLX package) are compiled with gcc 2.91.66.

Do simply a grep or use a binary editor with a search option and you will see.

[Sebi]

Unfortunately, all I've tried so far was completely with no result. The only thing that affects is disabling antialiasing support for QT3 - but then fonts look like hell. So I switch back to the nv driver module - I don't use 3D that much...

Originally posted by bwkaz
Also normal for X. On my system (X 4.2.0 from source, 3123 drivers), which doesn't experience these lockups, X's VSZ is ~300MB according to ps.

I'm quite sure this is NOT normal. Moved the whole configuration (incl. X, QT, KDE) to another PC with different Card (NeoMagic chipset). Size (according to PS/TOP) here is about 24000 (NOT MB!!).
Disabling almost everything in NVidia server didn't change everything; right after starting X by simply invoking "X" (i.e., with absolutely no apps) I get a Size of approx. 250 MB!

Switched back to the nv module gave me a Size of <70000, RSS 15M with a running X and apps started. IMHO this definetly is a problem of the X driver.

Sebi

[lpnotfr]

OK. I have good news.

I'm trying 4194. It seems stable (just a few hours of test though)

But a friend of mine sent me a patch for the 3123 kernel driver. With this patch, X worked 24h without crashing. Seemed to be stable :))

If 4194 still crashes, I'll put this patch on a ftp. Just one problem: what is exactly the license of the nvidia kernel driver ?