Memory leak in Xorg with Quadro FX 1600M/PCIe/SSE2

For quite a while with several driver updates (even up to the latest one 310.19), I’m getting a memory leak in Xorg (on current Debian testing x86_64 with KDE). It slowly increases RAM usage until it eats it all (which can take a number of days for 4 GB).

I’m running it on a laptop (HP Compaq 8710w) using the laptop screen and external monitor attached in TwinView mode to make one big continuous desktop. Mobile GPU is Nvidia Quadro FX 1600M/PCIe/SSE2.

See the bug report attached below (jpg extension added to bypass strange forum restriction…).

Typically, when the memory usage of the Xorg process grows over time, it’s caused by some X client (such as kwin) leaking resources. Can you try running xrestop to see if the number of resources used by some particular client is growing?

I noticed that “Misc” resources for plasma-desktop are gradually growing:

Here is xrestop output after the system has been running for a few days:

xrestop - Display: localhost
          Monitoring 30 clients. XErrors: 0
          Pixmaps:   17462K total, Other:    4672K total, All:   22135K total

res-base Wins  GCs Fnts Pxms Misc   Pxm mem  Other   Total   PID Identifier    
1c00000    47   11    1   40  307    10903K      9K  10913K 16183 kwin
2000000    73   73    1   44 189711      435K   4450K   4886K 16193 plasma-desktop
2400000     5    2    1  305  348     1903K      9K   1913K 16210 krunner
4000000    39   56    1 2825 3296     1686K     80K   1766K 21200 untitled - Geany
3a00000    16   73    1 3027 3402     1615K     82K   1698K 16694 Buddy List
1600000    12    8    1   26   49      243K      2K    245K 32291 Screenshots
3800000    16   63    0   73   79      213K      3K    216K 16624 SWT
3000000     2    2    0   56   64      125K      1K    126K 16528 kerneloops-applet
2a00000    17    7    0   10   33      120K      1K    121K 16445 kmix
1400000     3    3    0    6   13       72K    456B     72K 16190 knotify4
1000000     3    3    0    2  603       24K     14K     38K 16172 kglobalaccel
0c00000     4    3    0    2   19       24K    624B     24K 16164 kded4
1200000     3    3    0    2   11       24K    408B     24K 16176 kactivitymanagerd
0e00000     3    3    0    2   11       24K    408B     24K 16174 kwalletd
3600000     0    1    0    2    8       24K    216B     24K   ?   
2600000     0    1    0    2    8       24K    216B     24K   ?   
0000000     2    0    2    0  152        0B      5K      5K   ?   
0800000    23   23    1    2  124        0B      4K      4K 16255 Qt-subapplication
1e00000     0    1    1    0    3        0B      1K      1K   ?   
0200000     0    1    1    0    0        0B      1K      1K   ?   
1800000     4    3    0    1   11        0B    432B    432B 16181 Qt-subapplication
4200000     3    3    0    1   11        0B    408B    408B 18136 apper-sentinel
2200000     3    3    0    1   11        0B    408B    408B 16200 kuiserver
2e00000     1    1    0    0    0        0B     48B     48B   ?   
2c00000     1    1    0    0    0        0B     48B     48B   ?   xrestop
0a00000     1    1    0    0    0        0B     48B     48B   ?   
0400000     1    1    0    0    0        0B     48B     48B   ?   
3400000     0    1    0    0    0        0B     24B     24B   ?   
1a00000     0    1    0    0    0        0B     24B     24B   ?   
0600000     0    1    0    0    0        0B     24B     24B   ?

There Misc is 189711 for plasma-desktop. In comparison, after I restart kdm, that Misc value is around 390 but growing slowly.

I’d say that’s a leak where the developers of plasma-desktop aren’t releasing resources, not in the Xserver?

Recent update in KDE on Debian testing seems to have fixed this issue.