X.org Lockups (part 2)

Message

praseodymium · Post by **praseodymium** » Wed Jul 06, 2005 2:52 pm

I haven't. But it's not a good comparison either: nv doesn't do 3D accel anyway, so you could also say that 3D accel crashes Xorg.

ctford0 · Post by **ctford0** » Sat Jul 09, 2005 4:09 am

alaa wrote:Well everyone, one thing I know for sure is that when I moved from nvidia to nv in my xorg.conf file my pc has been rock solid for over a month now.

What does this suggest? A problem with the nvidia driver that's what.

Am I 100% sure? No

Has anyone had xorg lockup with nv?

There are just as many ppl having lockups using the ati supplied linux drivers.

this has to be an X/kernel problem. This has been going on for a couple of months now and I'm convinced that it has to be some type of miscommunication between X and the kernel itself.....

chris

l33tmmx · Post by **l33tmmx** » Sat Jul 09, 2005 2:11 pm

alaa wrote:Has anyone had xorg lockup with nv?

As I said on the post preceding yours, my system locked up regularly regardless of whether I used 'nv' or 'nvidia' drivers

pankkake · Post by **pankkake** » Sat Jul 09, 2005 4:53 pm

alaa wrote:Has anyone had xorg lockup with nv?

I'm having X lockups (100% CPU usage, able to reboot via ssh) with nvidia driver. Now I'm trying nv, I'll keep you informed.

drefo · Post by **drefo** » Sat Jul 09, 2005 9:33 pm

ctford0 wrote:
There are just as many ppl having lockups using the ati supplied linux drivers.

this has to be an X/kernel problem. This has been going on for a couple of months now and I'm convinced that it has to be some type of miscommunication between X and the kernel itself.....

chris

I agree: it seems like a more generic problem between X and the kernel
when using certain features of ati/nvidia graphics hardware. I have an
older ati card (radeon 7500) and I see regular lockups using the xorg
provided `ati' driver.

I've completely re-installed gentoo several times, have compiled
everything with CFLAGS="-O0 -g" and USE="debug", but the crashes I see
occur in a non-reproducible fashion and are not easily debuggable,
at least for me.

Running X remotely under gdb, when it works, shows either a corrupted
stack after the crash or it gives an apparently reasonable stack dump
but which originates in widely differing parts of X. Sometimes
unexpected signals are received by X, and I've seen SIGILL and SIGSEGV
more than once ... which suggests that X is clobbering kernel memory.

I've tried switching to the generic fbdev driver for Xorg. While
this seems to be marginally more stable than using the `ati' driver,
(so far I've seen no crashes using web browsers), I can still cause
the system to lock totally (ssh impossible and no response even to SyS
Rq) just by running `glxgears.'

The exact same problems are present for me on several recent linux
distributions (SuSE, debian, Slackware) as well as on FreeBSD 5.4.

So I vote for a general Xorg/kernel problem, possibly having to do
with some aspect of the GART code, which remaps video memory.

I suppose it could even be a cache consistency problem. To rule that
out, it would be interesting to know what architectures it occurs
on. If it doesn't occur on amd or amd64 but occurs on x86, with
exactly similar graphics hardware, for example, that could be a hint.

David

heavyjoost · Post by **heavyjoost** » Sat Jul 09, 2005 10:44 pm

If this is a X/kernel error, could it work without lockups in older kernels? Or with an older Xorg/XFree version but I believe that didn't work.

beatryder · Post by **beatryder** » Sun Jul 10, 2005 12:26 am

drefo wrote:
ctford0 wrote:
There are just as many ppl having lockups using the ati supplied linux drivers.

this has to be an X/kernel problem. This has been going on for a couple of months now and I'm convinced that it has to be some type of miscommunication between X and the kernel itself.....

chris
I agree: it seems like a more generic problem between X and the kernel
when using certain features of ati/nvidia graphics hardware. I have an
older ati card (radeon 7500) and I see regular lockups using the xorg
provided `ati' driver.

I've completely re-installed gentoo several times, have compiled
everything with CFLAGS="-O0 -g" and USE="debug", but the crashes I see
occur in a non-reproducible fashion and are not easily debuggable,
at least for me.

Running X remotely under gdb, when it works, shows either a corrupted
stack after the crash or it gives an apparently reasonable stack dump
but which originates in widely differing parts of X. Sometimes
unexpected signals are received by X, and I've seen SIGILL and SIGSEGV
more than once ... which suggests that X is clobbering kernel memory.

I've tried switching to the generic fbdev driver for Xorg. While
this seems to be marginally more stable than using the `ati' driver,
(so far I've seen no crashes using web browsers), I can still cause
the system to lock totally (ssh impossible and no response even to SyS
Rq) just by running `glxgears.'

The exact same problems are present for me on several recent linux
distributions (SuSE, debian, Slackware) as well as on FreeBSD 5.4.

So I vote for a general Xorg/kernel problem, possibly having to do
with some aspect of the GART code, which remaps video memory.

I suppose it could even be a cache consistency problem. To rule that
out, it would be interesting to know what architectures it occurs
on. If it doesn't occur on amd or amd64 but occurs on x86, with
exactly similar graphics hardware, for example, that could be a hint.

David

I used to run a 7500, and I NEVER had any problems with it
I used the "radeon" driver and it worked fine. I think the "ati" is for an older tnt card...

drefo · Post by **drefo** » Sun Jul 10, 2005 1:08 am

beatryder wrote: I used to run a 7500, and I NEVER had any problems with it
I used the "radeon" driver and it worked fine. I think the "ati" is for an older tnt card...

Yeah, I used to use the "radeon" driver too, and I never had any problems before I started using
2.6 kernels and the newest Xorg release.

I noticed looking at Xorg.0.log, using Xorg 6.8.2, if you specify "ati" as the driver in xorg.conf, it actually ends up
loading the radeon driver module ... there seems to be no real difference in the log output if you specify radeon instead,
so I think the new "ati" Xorg supplied driver must include code to detect all the possible `radeon' cards.

Anyway "ati" seems to work, 3D acceleration is on and all, but it makes the system unstable.

heavyjoost wrote: If this is a X/kernel error, could it work without lockups in older kernels? Or with an older Xorg/XFree version but I believe that didn't work.

I just installed SuSE 8.2/XFree86 4.3.0 on a separate partition and am trying my best to
crash the system. If this works, I'm going to try reverting to an earlier gentoo.

(This is with kernel: 2.4.20-4GB)

Update to come.

pankkake · Post by **pankkake** » Sun Jul 10, 2005 2:50 pm

Still crashing with "nv".
I'm going to disable agpgart/dri kernel support (altough my card is pci-e).

drefo · Post by **drefo** » Mon Jul 11, 2005 5:35 am

Update:

OK, my system still crashes even running SuSE 8.2 with the old X.

I was totally surprised by that, since I know that I formerly had no problems.

So I left memtest86+ running overnight, and at some point it crashed
with an `Unexpected Interrupt' error, dumping the registers and stack to the
screen. That's very bad, needless to say.

It seems I've got a hardware problem after all, either cpu, motherboard
or memory. I just replaced the memory, and I thought things were OK, since
I had already seen memtest86 run cleanly once through. But apparently things
are not quite right. I suppose it might be heat related, since it takes a while
to show up.

Anyway, I think that the crashes I'm seeing are probably not at all related
to what everyone else is seeing, even though most of the symptoms seemed to fit.

Sorry for the confusion.

For anyone with an older system who's seeing these kinds of lockups I'ld
say it's worth doing an extended run of memtest86.

cybertronix · Post by **cybertronix** » Mon Jul 11, 2005 5:22 pm

Here's my version of the X lockup story. I was having trouble with this in both Mandrake 10.0 and Gentoo until I noticed a lot of error messages about wmfire in my console after I started X (didn't happen in Mandrake though). I removed wmfire and was stable for a while, so I assumed it was that. i installed a newer version than what is in portage and all of a sudden everything was happy. I had a 50 day uptime before having to shut down to move my computer.

After the move I never got quite that large an uptime due to a number of hardware upgrades (but no X lockups) and eventually I bought myself a new hard drive. Re-installed Gentoo on it, used the newer version of wmfire, and started getting X crashes again. At least I assume they're still X crashes as each time it has happened when I wasn't actually at the computer. One time the screensaver was frozen on the screen as was the computer, and the other times I came home and it was locked solid but the screen had already shut off due to powersave mode. It appears that it happened again today because my web site hosted on that computer has suddenly stopped responding again. Can't ping or ssh either. My experience has been that if I catch the lockup quickly enough I can ssh in and sometimes restart X, although sometimes that locks the computer too. After a long enough time though it stops responding altogether.

So I guess I'm still looking for that one element that changed from my original Gentoo installation to my new one. Fortunately I still have the old partition around so I can check settings if need be.

Good luck to everyone in figuring this out. Hopefully we can get it licked and go back to never having to reboot.

Oh, and I have a couple of Suse boxes at work that are coming up on 50 days themselves (basically since they were installed) without a hitch. Of note is the fact that one is using the i810 drivers and the other is using the nv driver.

Gergan Penkov · Post by **Gergan Penkov** » Tue Jul 12, 2005 6:12 pm

It is probably offtopic, but after switching off RenderAccel, i don't have any lockups anymore, but ... What is interesting in recent times :
1)I hear a beep, when I start X (strange),
2) sometimes galeon hangs, but could not lock up the system
3) today I tried the glxgears, first as a normal user - it was around 700-800, which is way too little (I use NVidia GForce4 440MX - but I don't play, and it suffices me), because earlier runs have showed more (how much I don't know but it was in 1000s, probably 1200 or sth like this). Then I tried it as a root, it was a little faster, but what was very strange were the return of the beeps at regular intervals, so long the glxgears ran.
4) After this I could not switch to console with f1-f6, the display simply goes in poweroff, switching back shows X.
Have someone such an experience and what it could be - probably the kernel, I use 2.6.12-love1 (no splash and vesafb)?

yaneurabeya · Post by **yaneurabeya** » Wed Jul 13, 2005 7:58 am

Considering all of the apparent issues, it's most likely something with the OpenGL tie-ins with the kernel and X. Still no issues with FreeBSD after many weeks of testing with the latest nVidia driver 1 month ago (haven't updated since I'm living off-campus now though...).

Pinkis · Post by **Pinkis** » Wed Jul 13, 2005 10:03 am

Yes I had lockups too. I've tried various driver versions, played with xorg.conf but in vain. There was no way to do the job with nvidia provided drivers but "nv" was ok. Now I'am pretty sure that in my case it was a hardware problem cause after switching my old GF2 MX400 to another old GF FX 5200 the problem vanished. And I must add that gentoo was not the only one OS that had problems with previous video card (oh yes I have windows xp in mind). So the moral of the strory is that maybe I say maybe some of you guys have faulty hardware too.

Annie · Post by **Annie** » Wed Jul 13, 2005 2:36 pm

I had the same random freezes for a while. So, I downgraded from a Nvidia 7664 module to the version prior to that. And I decided to disable RenderAccell and "GLX" within my Xorg.conf for a while. Till now I noticed that KDE loads faster (I could notice some lag while KDE was trying to render some images) and the lockup didn't occur for almost 20 minutes at this point. Usually it occurred whenever I started Konqueror and trying to post to some discussion boards...

Let's see if it stays this stable....

pankkake · Post by **pankkake** » Wed Jul 13, 2005 2:58 pm

3 days (no reboots) without lockups. I have disabled AGP GART and DRI support in the kernel and added

Code: Select all

Option "NvAGP" "1"

in my xorg.conf. I also have RenderAccel and Hardware cursor disabled, but I had lookups with this too.

sbbeebe · Post by **sbbeebe** » Fri Jul 15, 2005 3:36 pm

UPDATE
the solution below helped, but didn't prevent lockups from re-ocurring.
----
Well, obviously there's an xorg / kernel bug underneath all this. Therefore there is no solution here, but I was able to get a usable system by:

1) updating to latest nvidia-kernel and nvidia-glx (7667).
2) setting NvAGP = "0" in xorg.conf. Used to have it set to "1".
[edit]
3) turning off the hardware cursor (HarwareCursor = "False").
[/edit]

As with most other posters in this thread I tried all kinds of things - changing use flags and recompiling, upgrading / down grading various packages, but I pretty much ended up back where I started (a gentoo sources 2.6 kernel based system, using stable packages) with the above two changes. When I was getting lockups, all I had to do was run glxgears and then try to move glxgears window and it would lockup every time.

Here's some other relevant info:

Kernel 2.6.12-r6 gentoo sources

Xorg version 6.8.99.14
snip from xorg.cong...

Code: Select all

Section "Device"
        Option     "DigitalVibrance"         "20"    # <i>
        Option      "NoLogo"                   "True"  # [<bool>]
        Option      "NvAGP"                    "0"     # <i>
        Option      "HardwareCursor"      "False"  # [<bool>]
        Identifier  "Card0"
        Driver      "nvidia"
        VendorName  "nVidia Corporation"
        BoardName   "NV18 [GeForce4 MX 440 AGP 8x]"
EndSection

Running either gnome (2.10) or xfce4 (4.2.2) - doesn't really matter.

hth

whoracle · Post by **whoracle** » Sat Jul 16, 2005 9:37 am

I had some similiar problem, the computer hanging both with X running and not. But only when connected to the network i think.. After a couple of months i got pissed off and tried windows on the computer, it ran quite fine and I went back for gentoo and now it works fine too!? I have absolutely no idea what could have caused this strange behaviour! But happy I am

now i run:
nvidia-kernel: 1.0.6629-r4 (driver version?)
kernel: 2.6.12-r4
driver I use is vesa!? will be changed now to see if any difference occurs.

jtox · Post by **jtox** » Sat Jul 16, 2005 5:17 pm

This happens to me as well and I'm suprised that this thread has lasted so long without any clear solution ... Out of curiosity, would anyone still having problems want to post the size of their virtual memory, and what kind of DRI/GART settings they do (or don't in the case of nvidia) have - along with whatever ACPI settings are used? Also, for the person who can reboot from ssh, can you post up your X error log of the approximate time when X locked up? Has anyone tried debugging+ssh? Or perhaps waiting for X to lock while on a separate tty?

It usually happens to me with xscreensaver starting and running kaleidoscope (yes I turned it off!) ... aside from glxgears, anyone else have any separate cases?

renrutal · Post by **renrutal** » Sat Jul 16, 2005 11:52 pm

jtox wrote:It usually happens to me with xscreensaver starting and running kaleidoscope (yes I turned it off!) ... aside from glxgears, anyone else have any separate cases?

I do, my locks usually happen when I'm scrolling pages, documents, anything with a rolling bar. It's not always or even often, but they happen. Abiword is my main concern, it does crash Xorg a lot more often than anything else.

If I had to bet in a problem, I'd say it's the speed. Abiword is really fast, dunno if they optimized it a lot in the page rendering area and Xorg is not handling it fast enough. Or maybe the video card isn't handling Xorg->Kernel->Video Card requests. I'd say the lack of syncrony is making it crash, someone must have "optimized" something in Xorg by taking out some security/stability checks/code that were making Xorg slower.

zeus_045 · Post by **zeus_045** » Sun Jul 17, 2005 5:33 pm

I tried posting this on my gentoo machine but it locked up !!

Locksup 95% of the time when using gaim, but has locked when browsing and well, posting the first version of this.
Tried switching back to the `nv' driver, lockups are just a frequent.

a7n8x-delux
amd xp 2500
nvidia gforce 4 ti 4200

My box use to lockup in the shell as well (no X) until I disabled APIC, so far no lockups have happened, so thats a seperate issue afaik. The lockups in X are quite random, felt like posting this, even with the lack of logs and knowledge, because it may help somehow...

I have another box running slack with the nvidia drivers, running a gefore 3 ti 4200, and its running gun, not a single crash. MIght switch cards see what happens.
I'm also going to put slack on there, and see what happens as well.

My laptop (tecra m2 - fx 5200), which also ran slackware and fedora at one stage with nvidia drivers ran gun also, once I put gentoo on there I was getting random lockups on starting X. Running `nv' only gets me a stable system, no lockups. Disabling framebuffers also sorted out a corrupted screen on closing X with the bonus of yet another random lockup(running nv allowed me to keep the framebuffers and avoid this issue). Also would like to add I tried out ubuntu, had exactly the same problems.

Over and out.

azuriel · Post by **azuriel** » Tue Jul 19, 2005 12:01 am

Okay, my freezes have stopped. Before, it would commonly lockup (no keyboard input) when using firefox, and just firefox. I could still ssh in and reboot, but that was about it. However, after setting

Code: Select all

Option "NvAGP" "1"

this has stopped happening. I've been getting these lockups with a variety of versions of Xorg, kernels, window managers and nvidia drivers, so I think this line in xorg.conf is the key.

These are my relevent xorg.conf settings:

Code: Select all

Section "Device"
    Identifier  "NVidia GeForce4 MX440"
    Driver      "nvidia"
    Option      "RenderAccel" "true"
    Option "NvAGP" "1"
Option "NoLogo" "true" #Turns off NVidia logo on startup

EndSection

Versions of some packages:

xorg-x11-6.8.99.14
firefox-1.0.5
nvidia-kernel-1.0.7667

Hope this helps someone else! Those lockups were totally disabling my computer usage.

TheCarNinja · Post by **TheCarNinja** » Wed Jul 20, 2005 5:55 am

I figured i would add my own two cents like the others in the hope it may help in some way and also since im about as desperate as the rest of you to try and solve this.

Specs:
IBM T22 Thinkpad
900mhz
IXSavage Card
no opengl
no dri
Jackass Gentoo 2005.0
2.6.12-r4 kernel
X.org(6.8.2r2) + XFCE4

Also this is a brand new Gentoo install.

I've been having very similar if not exactly the same problems that all you have been describing on this/these thread(s) for quite some time; I'd say about a year ago. I changed kernels and then bam, i was hit with this spontaneous freezing bit. Sometimes while using the box i'll get about 5 minutes of usability out of it before it locks up on me, sometimes ill get a day or even three days. Its really random and doesn't seem to have much to do with what i do, browsing, coding, typing ect. It always goes down though. I run 2 other gentoo boxes and they dont have these problems (although now it honestly seems like I'm tempting fate saying that).

The freeze is the same, the machine locks up, if theres music that gets stuck on whatever pitch it was last playing, and all inputs have no effect. I've tried sshing into the box but all i get is no route to host. Reading through the thread I saw someone suggested adding some kernel options in grub.

Code: Select all

idle=poll noapic nolapic

After adding that line i did some thorough testing of the box and it died after about 2hrs. I was really pushing it too, which means it probably had a good impact on the system. This time however i was already sshed into it figuring that before sshd dies (if it did) i would at least (hopefully) get some error message before it went. This is what i recieved:

Code: Select all

Broadcast message from root (vc/3) (Tue Jul 19 22:22:00 2005):

The system is going down for reboot NOW!
Read from remote host: Connection reset by peer
Connection closed.

Needless to say i was very surprised. I thought i had to tell it to reboot before but now it seems its trying to guess my wishes! My machine is either intelligent or insane! My guess is the latter (i hope). I can only assume that whatever killed the box killed my init too? I don't know. Any ideas?

Oh if i forgot something crucial let me know, its 2am and im exhausted.

pankkake · Post by **pankkake** » Wed Jul 20, 2005 3:59 pm

pankkake wrote:3 days (no reboots) without lockups. I have disabled AGP GART and DRI support in the kernel and added
Code: Select all
Option "NvAGP" "1"
in my xorg.conf. I also have RenderAccel and Hardware cursor disabled, but I had lookups with this too.

10 days uptime, no lockups

. The next time I restart X or my computer i'll try to re-enable RenderAccel and HWCursor.

Luka21 · Post by **Luka21** » Thu Jul 21, 2005 2:11 pm

Could it be the fault in X accelleration system and some drivers using hardware acceleration?

In my case using ati proprietary driver with radeon 9000 mobility resaults in frequent freezing when using 2D, but is stable for playing most of the games (ut2004, warcraft III, rtcw, ...).

Setting agpmask and agpmode didn't fix anything for me, but if I intentionaly set
opengl-update xorg-x11
which renders dri useless for fglrx driver,
my xorg stopps/reduces freezing and 2D movement is still accellerated.

edit: it is still freezing, but less frequent

I think this particular problem will actullay be resolved - atleast ati marked both of my tickets for freezing as resolved.

X.org Lockups (part 2)

Work around