KWin5 memory leak?

René J.V. Bertin rjvbertin at gmail.com
Mon Nov 13 17:21:29 UTC 2017


On Monday November 13 2017 17:20:56 Martin Flöser wrote:

Hi,

Is it possible that many if not most of the users who might report this kind of issue use (very) fast disks combined with lots of RAM (and possible tiny amounts of swap space) nowadays? (Meaning they never notice anything out of the ordinary.)

>Please note that KWin maps memory from other applications (window 
>content).

How is that even possible? Do you mean KWin allocates memory to cache or clone the window content generated by other applications?

>It might be wrongly attributed and might just go away when the 

I look at the VM usage as reported by htop. I'm fairly certain that it looks at actual swap memory usage. It's been a while that I looked at its sources, though.
What is certain is that the issue is real. I can almost feel the disk work when swap usage goes over 800-900Mb and I do desktop swaps and the like ... or when I restart KWin and over 1Gb of space is given back to the free pool.

>compositor is killed. I'm quite certain you look at the wrong offender 
>here.

All I can say is that the memory is released when I restart KWin. I don't know exactly what is going on, but unless there are helper applications that are also restarted when KWin is restarted I don't see any other explanation or other offender. The only other possible memory hog I can think of here is the X xerver itself, but that would suppose it needs globs of memory for the window decoration.

In fact, what surprises me most is how well so much VM is given back. It's true that I use a swap partition so there's no such situation as "can't unlink this swap file yet because a tiny portion of it is still in use" but I'm more accustomed to swap continuing to grow without ever going back to amount used after just logging in.

Cheers,
R.


More information about the Plasma-devel mailing list