The productivity falls stronly after few work hours

[AndrewM]

Through a 10-20 works hours the productivity of the graphic system falls stronly (depending on intensity of manipulation with windows).

Examle:

At the first start of the Xorg:

$ vmstat -m | grep nvidia
       nvidia   823  3712K       -   105299  16,32,64,128,256,512,1024,2048,4096

$ glxgears
29852 frames in 5.0 seconds = 5970.293 FPS
29936 frames in 5.0 seconds = 5987.193 FPS
29853 frames in 5.0 seconds = 5970.447 FPS

$ env __GL_YIELD="NOTHING" __GL_FSAA_MODE=0 \
     __GL_SYNC_DISPLAY_DEVICE='DFP-0' __GL_SYNC_TO_VBLANK=1 \
    compiz --replace ccp  --loose-binding --ignore-desktop-hints

$ glxgears
6598 frames in 5.1 seconds = 1302.401 FPS
6309 frames in 5.0 seconds = 1261.758 FPS
6607 frames in 5.0 seconds = 1321.286 FPS

. . . 

In a hour intensive work:

$ vmstat -m | grep nvidia
       nvidia  8302 10231K       -  2450095  16,32,64,128,256,512,1024,2048,4096

$ glxgears
4976 frames in 5.0 seconds = 994.781 FPS
5170 frames in 5.0 seconds = 1033.955 FPS
4806 frames in 5.0 seconds = 961.000 FPS
5656 frames in 5.0 seconds = 1131.097 FPS

After two-three hours a FPS is equal to 700, after several more hours a FPS is equal 500 and so on. During the day FPS goes down to 100-200 and at the end work becomes impossible. The restart of X doesn't solve the problem, only complete restart does.

it's possible to force this problem:
1. turn on "Wobbly Windows Plugin" in Compiz;
2. To load a processor on 100% (for example, by the program sysutils/cpuburn);
3. During few minutes windows must be in action and watch the glxgear utility

$ uname -rms
FreeBSD 6.2-STABLE i386

$ pkg_info -Ex xorg-server nvidia-driver compiz-0 kde-3
compiz-0.6.2
kde-3.5.8
nvidia-driver-100.14.23
xorg-server-1.4_3,1

$ dmesg | grep CPU
CPU: AMD Sempron(tm) Processor 2800+ (1760.18-MHz 686-class CPU)

$ sudo nvclock -i
-- General info --
Card:           nVidia Geforce 7600GS
Architecture:   NV4B B1
PCI id:         0x392
GPU clock:      480.000 MHz
Bustype:        PCI-Express

-- Pipeline info --
Pixel units: 3x4 (011b)
Vertex units: 5x1 (11111b)
HW masked units: None
SW masked units: None

-- Memory info --
Amount:         256 MB
Type:           128 bit DDR
Clock:          796.500 MHz

-- PCI-Express info --
Current Rate:   16X
Maximum rate:   16X

-- Sensor info --
Sensor: GPU Internal Sensor
GPU temperature: 48C

-- VideoBios information --
Version: 05.73.22.50.00
Signon message: GeForce 7600 GS VGA BIOS
Performance level 0: gpu 400MHz/memory 800MHz/1.15V/100%
Performance level 1: gpu 400MHz/memory 800MHz/1.15V/100%
VID mask: 0
Voltage level 0: 1.15V, VID: 0

$ nvidia-settings -q ConnectedDisplays -q RefreshRate | grep Attribute
  Attribute 'ConnectedDisplays' (inspirra.localdomain:0.0): 0x00010002.
  Attribute 'RefreshRate' (inspirra.localdomain:0.0; display device: CRT-1): 75,00 Hz.
  Attribute 'RefreshRate' (inspirra.localdomain:0.0; display device: DFP-0): 60,00 Hz.

[zander]

Please generate and attach an nvidia-bug-report.log file; when you say that the problem can be forced, do you mean that the symptoms can be replicated or that the system slows down permanently?

[AndrewM]

Starting Xorg+KDE+Compiz

$ env __GL_FSAA_MODE=0 \
    __GL_SYNC_DISPLAY_DEVICE='DFP-0' \
    __GL_SYNC_TO_VBLANK=1 \
    compiz --replace ccp --ignore-desktop-hints --loose-binding 2>&1 &

$ nvidia-bug-report.sh 
$ mv nvidia-bug-report.log 1-nvidia-bug-report.log

$ LANG=C ; (date ;  vmstat -m | grep  nvidia)
Tue Dec 25 06:16:47 MSK 2007
       nvidia   816  3711K       -     2232  16,32,64,128,256,512,1024,2048,4096

$ glxgears
6901 frames in 5.0 seconds = 1380.188 FPS
6969 frames in 5.0 seconds = 1393.605 FPS
6948 frames in 5.0 seconds = 1389.464 FPS

Right after the system booting, all works well. But in course of time, the graphic system slows down.

The symptoms can be replicated:

It is possible to reproduce symptoms thus: it is necessary to activate "Water Effect" in Compiz on 30 minutes . After it, the graphic system does not restore the productivity

$ glxgears
6901 frames in 5.0 seconds = 1380.188 FPS
6969 frames in 5.0 seconds = 1393.605 FPS
6948 frames in 5.0 seconds = 1389.464 FPS

Turn on "Water Effect Plugin" in Compiz, and toggle "rain". (only for the repetition of the symptoms)

$ while sleep 30 ; do  vmstat -m | grep nvidia ; done
       nvidia   833  3713K       -    77900  16,32,64,128,256,512,1024,2048,4096
       nvidia   829  3712K       -    77969  16,32,64,128,256,512,1024,2048,4096
       nvidia  2291  4276K       -    79812  16,32,64,128,256,512,1024,2048,4096
       nvidia  4000  4329K       -    81534  16,32,64,128,256,512,1024,2048,4096
       nvidia  5521  4377K       -    83095  16,32,64,128,256,512,1024,2048,4096
       nvidia  7086  4426K       -    84678  16,32,64,128,256,512,1024,2048,4096
       nvidia  8674  4477K       -    86506  16,32,64,128,256,512,1024,2048,4096
       nvidia  9775  4511K       -    88649  16,32,64,128,256,512,1024,2048,4096
       <...>SKIP 44 LINES<...>
       nvidia 78911  8793K       -   169842  16,32,64,128,256,512,1024,2048,4096
       nvidia 78748  6668K       -   173264  16,32,64,128,256,512,1024,2048,4096
       nvidia 78766  6670K       -   173501  16,32,64,128,256,512,1024,2048,4096
       nvidia 78767  6670K       -   174111  16,32,64,128,256,512,1024,2048,4096

Turn off  "Water Effect Plugin" in Compiz

$ glxgears
689 frames in 5.0 seconds = 137.793 FPS
835 frames in 5.0 seconds = 166.987 FPS
770 frames in 5.0 seconds = 153.459 FPS
771 frames in 5.0 seconds = 152.894 FPS

$ nvidia-bug-report.sh
mv nvidia-bug-report.log 2-nvidia-bug-report.log

Maybe, that it related to the option "--loose-binding". But without this option Compiz works very badly permanently.

[AndrewM]

An analogical situation takes a place when I use the google-earth program:

$ pkg_info -Ex google-earth
google-earth-4.2.205.5730

[AndrewM]

I've discovered that if "__GL_SYNC_TO_VBLANK=1" env variable is not used, there's no no memory leak; but it's still leaking when I use google-earth

[AndrewM]

There's no memory leak with a new 196.12 driver if the googleearth to use without Compiz.

But, there's memory leak if using:
# compiz (with General compiz option: sync_to_vblank=ture) & googleearth

And there's memory leak if using:
# mplayer -vo gl2 file.avi (with a variable: __GL_SYNC_TO_VBLANK=1 OR nvidia-setting Attribute: 'SyncToVBlank' is 1)

[uOpt]

Unloading and reloading the kernel module should make it work without a reboot. Still gotta nuke the X11 server, though.

[AndrewM]

Originally Posted by uOpt

Unloading and reloading the kernel module should make it work without a reboot. Still gotta nuke the X11 server, though.

The module unloads without problems if a memory leak did not happen. But it is impossible to unload the module when a memory leak happened...

1. I stop the Xorg, hal, dbus, etc...
2. $ kldunload nvidia
kldunload: can't unload file: Device busy

[zebdits]

Originally Posted by AndrewM

1. I stop the Xorg, hal, dbus, etc...
2. $ kldunload nvidia
kldunload: can't unload file: Device busy

Did you try this?

fstat /dev/nvidia0

[AndrewM]

Originally Posted by zebdits

Did you try this?

fstat /dev/nvidia0

Yes, sure.
After stop of the Xorg-server, this file (/dev/nvidia0) is not using by any process.
Nevertheless, unloading the module is impossible:

$ fstat /dev/nvidia0
USER CMD  PID FD MOUNT INUM MODE SZ|DV R/W NAME
===EMPTY===
$ kldunload nvidia
kldunload: can't unload file: Device busy

[uOpt]

Anything in /var/log/messages or dmesg about this?

[AndrewM]

Originally Posted by uOpt

Anything in /var/log/messages or dmesg about this?

$ newsyslog -F /var/log/messages
$ reboot
$ xinit
$ twm ; glxgears ....
I leave the Xorg: CTRL+D
$ kldunload nvidia
nvidia0: detached
$ kldload nvidia
nvidia0: <GeForce 7600 GS>......
nvidia0: [GIANT-LOCKED]
$ xinit
$ compiz --replace ccp --loose-binding ......
I leave the Xorg: CTRL+D
$ fstat /dev/nvidia0
USER CMD PID FD MOUNT INUM MODE SZ|DV R/W NAME
===EMPTY===
$ kldunload nvidia
kldunload: can't unload file: Device busy
$ nvidia-bug-report.sh
See attached log.