memory leak? massive slowdowns opening apps

[vorpal]

Hi.

Sometimes after opening an application, the system slows to a crawl for about 1 to 2 minutes, then returns to normal. Mouse movement becomes jerky, redrawing the desktop background can take many seconds, etc.

This is always reproducible by starting X and then opening eg. several console windows, or instances of konqueror. Once the slowdown occurs, no new applications will open.

At this point, even running 'vi' in a terminal results in an unceremonious (and immediate) "segmetation fault" !!

Switching to a console (ctr+alt+f2) running top, I find that one or more recently started processes has started to gobble up huge amounts of memory. 200M, 400M, 600M, rapidly increasing. The processor usage is at 100%, divided between the rogue processes and kswapd.

The processes reach a size of 2000M and either crash, or in some cases, suddenly return to normal memory usage, releasing all the used memory. The slowdown stops.

This problem occurs occaisionally under drivers 2960 and 3123, and only when several applications are running.

However, under 4191, this slowdown can happen opening even one application, and sometimes even while KDE is starting.

Reverting to the kernel driver fixes the problem. I can open zillions of windows, no probs.

Is this a memory leak?
Might it be due to the race condition in the module's interface code?
Is this related to the troubles people are having with slowdowns in Nautilus? Has some Nautilus person checked memory usage during slowdowns?
Is this likely to be fixed by the patches at www.minion.de? What do they do and who should use them? Do they work with 2.4 kernels?

The system is:
-Mandrake 9
-kernel 2.4.19-16mdkenterprise (compiled with 4G -mem support, I believe)
-XFree 4.2
-gcc 3.2
-KDE 3.0.3 (I'll try it under gnome)

-using .src.rpm versions 4191, 3123, 2960

-Athlon 1700
-1G ram, 6G swap
-64M GF2MX400

Thankyou,

jvf

[empty]

I doubt it is a memory leak but just a bad driver release.. this is the same problem im getting from 4191, looks like its normal. =\
Bad redwraw and 2D rendering, causing huge CPU spikes but no memory spikes. I cant get 3123 drivers and lower working either, XFree86 loads the NVIDIA screen 4 times then crashes. Know what the problem could be ? but for your problem its normal from the gist of many people in here.

[vorpal]

I'm surprised you're having trouble with the 3123 set. What hardware are you using? Which type of install?

Also, if you run top during your slowdowns with 4191, do you find processes gobbling lots of memory? or is it just lower performance all round?

[crimsun]

The output of ``vmstat 5'' during one such period of heavy swapping would be quite helpful (let it run for a couple minutes).

[vorpal]

Here it is. You learn something new about linux everyday ;)

notes:
A: Started 4 instances of konsole
B: Started 4 more instances - the 5th won't open
C: Now one of them starts to eat up main memory..
D: ..and swap..
E: then, after 95 seconds, everything's OK

Free memory drops from 780M down to not much, then back up to 940M (some other stuff stays swapped out). Swap usage goes from zero, gradually up to 1.2G, then suddenly down to 115M as the process dies.

This happens infrequently with 2960/3123, very frequently with 4191, and I've never reproduced it under the kernel drivers.

procs memory swap io system cpu
r b w swpd free buff cache si so bi bo in cs us sy id
0 0 0 0 812124 9544 103140 0 0 105 23 191 274 6 1 93
2 0 0 0 812116 9552 103140 0 0 0 7 209 530 8 2 90
2 0 0 0 796212 9560 103144 0 0 0 54 220 2470 52 11 37 (A)
1 0 0 0 785604 9580 103148 0 0 0 100 195 4903 47 13 40 (B)
1 0 0 0 177532 9600 103312 0 0 32 83 216 3719 38 41 21 (C)
1 1 3 118388 10412 6076 53956 275 19778 287 19881 813 2016 9 27 64 (D)
2 0 1 194412 10360 5884 49728 951 14631 956 14638 508 1407 7 80 12
1 0 3 272272 10312 5884 45500 394 15999 394 16002 484 1082 4 81 15
2 1 1 347076 10320 5884 41020 572 14550 572 14554 485 1370 2 85 13
1 1 2 437744 11240 5884 37152 779 18286 779 18314 546 1663 6 86 8
1 0 2 518732 10860 5760 33284 401 16510 411 16564 506 1732 5 87 9
1 0 1 581568 10260 5344 30656 181 12178 190 12182 427 1556 7 80 13
0 1 2 651712 12196 4984 30084 161 14146 161 14149 473 1165 3 77 20
1 0 3 725984 10396 4456 28760 388 14664 389 14668 518 855 1 70 29
1 0 3 792900 10028 4128 28268 446 14503 446 14506 484 1210 5 77 18
1 0 2 857256 10392 4128 28276 447 12749 454 12752 488 1634 3 80 17
1 0 2 909248 9948 3900 28228 163 10398 163 10401 392 1368 9 70 21
0 1 1 969404 9964 3720 27672 179 12250 179 12254 445 1382 3 78 20
0 2 1 1020740 9828 3664 27356 91 10147 91 10150 383 1314 7 71 22
1 0 2 1069552 8452 3564 27192 55 9748 55 9751 367 1311 5 69 27
1 0 2 1124020 10552 3544 27044 142 10478 142 10482 402 1357 4 76 20
1 1 1 1179668 10884 3424 26696 122 11514 122 11518 405 1398 11 74 15
1 0 2 1249692 11964 3376 26364 99 13294 99 13298 437 1388 5 86 9
1 0 0 115656 944188 3668 28624 693 2218 1212 2249 316 608 3 21 76 (E)
1 0 0 115360 943944 3680 28644 42 0 47 6 190 336 6 1 93
1 0 0 115328 943388 3688 28644 87 0 87 3 192 330 4 1 95
2 0 0 115208 943208 3696 28644 34 0 34 3 188 333 8 2 90
0 0 0 114984 942672 3704 28644 106 0 106 3 191 329 4 1 95
0 0 0 114972 942212 3712 28644 90 0 90 3 195 343 6 1 93
0 0 0 114776 941636 3724 28720 98 0 114 31 199 338 5 0 95
0 0 0 114664 941432 3732 28720 39 0 39 3 188 311 5 1 93
1 0 0 114664 941032 3740 28720 78 0 78 3 190 319 6 1 93
1 0 0 114664 940740 3748 28720 58 0 58 3 189 317 5 1 94

[crimsun]

Thanks. Your system is spending a considerable amount of time thrashing. I'll dig into Mandrake's current kernel source to see if they've applied any of Andrea's patches for this. The recourse may have to be switching to a vanilla kernel (2.4.20) and applying Andrea's latest (-aa1) vm stubs against it.

[empty]

I beleive that might be my problem with the 4191s aswell. But with the 3123s .. hardware is Gf2MX 32mb, Athlon 1ghz, 256mb SDRAM on ASUS A7V133. Ive tried SRPMs and source tarballs, tried `make SYSINCLUDE=/path/to/source', still no luck. Just crashes loads the nvidia screen 4 times then crashes, every time I see the mouse icon ( with the watch ) come up. Ill post the output of `tail -n 20 /var/log/XFree86.0.log' - any help would be appreciated. I cant even go back to old drivers without reinstalling =\

[empty]

here it is:

(II) Initializing built-in extension RENDER
(II) [GLX]: Calling GlxExtensionInit
(II) Keyboard "Keyboard1" handled by legacy driver
(**) Option "Protocol" "ExplorerPS/2"
(**) Mouse1: Protocol: "ExplorerPS/2"
(**) Option "CorePointer"
(**) Mouse1: Core Pointer
(**) Option "Device" "/dev/usbmouse"
(**) Option "ZAxisMapping" "6 7"
(**) Mouse1: ZAxisMapping: buttons 6 and 7
(**) Mouse1: Buttons: 7
(II) XINPUT: Adding extended input device "Mouse1" (type: MOUSE)


no more after that, just goes into console.