Cure for the latest catalyst's memory leak (ie. catalyst 13.8 & 3.10 kernel)
jacob from AMD has found that the memory leak which we were experiencing when using ie. catalyst 13.8 beta 2 with combination of linux kernel >= 3.10 was actually caused by a community patch (the one which was "fixing" the access to proc files for non-root users).
He also write a patch which seems to work well, giving proper non-root access without memory leaks.
If you look into 13.8 beta2 source (./common/lib/modules/fglrx/build_mod/firegl_public.c), you see that the "problematic community patch" was integrated into 13.8b2 and failed with in AMD QS with 3.11 / 3.10 or was not tested at all :-)
This is also the reason why jacob's patch is applying onto 13.8b2.
Yeah whatever. The community user wrote that patch, to make his graphics card usable with stable kernels. It would never be happened, when there would be an AMD driver in time, that would support the current stable kernel, should not be too hard to accomplish.
ATM (I'm kinda afk) I'm unable to check catalyst' 13.8 beta sources and tell where 3.10 kernel support came from. Althought it's probable that it came from AMD itself, since both community patches for 3.10 kernel that I know about (so kolasa' and gentoo' for catalysts 13.4 and 13.6) didn't have problems with memory leaks.
But let me tell the story from it's very begining...
After catalyst 13.8 beta release (whith initial 3.10 kernel support, which as said above probably came from AMD) a new bug was found - problems with access to ati's proc files for non-root users. Some community user write a patch (simple, 2 lines patch) which was fixing it and so non-root user could run opengl apps (this patch was never added to catalyst). So after we were finally able to get it working right we noticed a huge memory leaks, but we didn't know where it originally came from, we were thinking it's catalysts fault.
Like said on 1st post it turned out that the problem with memory leak wasn't catalyst' fault - but this simple, 2 line patch.
Ofcourse nobody's saying that AMD did everything right here, because obviously it didn't, since there should be no problems with non-root access at all.
Also nobody's saying that it's user's (who write the patch) fault that he tried to save our asses and gave us the code which seemed to work well.
I opened this thread only to help users and maintainers with the memory leak :>
Yes - I got your point. My point was in short: Community should not have to patch a binary closed source driver at all.
Couldn't agree more!
Originally Posted by fritsch
I am really happy, that fglrx is not needed anymore, as the OSS readeon made big step forwards within the last weeks (Not sure about gaming though).
radeonsi is still lacking the opengl performance. OpenCL support is kinda missing. But yes, overall great step forward! Half a year ago nobody would dare to think that it will look so well in this year.