[Mesa-dev] Possible memory leak with `glxMakeCurrent`?

[Eero Tamminen]

Hi,

On 15.09.2016 11:00, Itai wrote:
> What I should have said is I have never written anything in Mesa. I
> tried making a small C program reproducing the problem, but I couldn't
> get the same result (Not entirely sure what the JRE is doing).
> I have run valgrind on the sample Java program, which produces a lot of
> extraneous data, but I hope there is enough there to find the problem.

On a quick look I would say that the Java program / VM doesn't at exit 
get properly rid of the context(s) it creates.  Additionally, I think 
it's creating compatibility profile, because largest allocations are 
from swrast context creation (AFAIK added for compat fallbacks).



> At the time the program was stopped, it's memory consumption exceeded
> 1.5GB, only about 200MB of them accounted for by the JVM (which means
> about 1.3GB were in native/mesa code).

I would recommend killing the process with a signal that Valgrind can 
catch, but program doesn't (e.g. SIGCPU is something most programs don't 
catch :-)).  That way you see the actual run-time leaks instead of being 
confused with issues program creates with its exit code.  A lot of 
programs don't properly free their resources at exit and people often 
confuse Valgrind reports of these application's exit issues with 
run-time leakage.


	- Eero

> This was run using Mesa 12.0.2-1 on Debian testing. Attached is the
> GZipped valgrind output, with hope it could help to find the problem, or
> at least write a proper (non-Java) test case
>
> On Thu, Aug 18, 2016 at 6:08 AM, Michel Dänzer <michel at daenzer.net
> <mailto:michel at daenzer.net>> wrote:
>
>     On 18/08/16 05:39 AM, Itai wrote:
>     > (Posted initially in mesa-users, but got no reply - the list seems dead.
>     > Couldn't find any bug report, and sadly not well versed enough in mesa
>     > to file one myself).
>
>     FWIW, there's no need to be versed in Mesa to file a bug report. :)
>
>
>     > Following an investigation of a memory leak with JavaFX on some Linux
>     > configuration, it looks like there is a possible memory leak when using
>     > `glxMakeCurrent`.
>     > Sadly, I myself don't know enough about OpenGL/Mesa to describe it
>     > fully, but I'm hoping someone here can understand it well enough to make
>     > a proper bug report.
>     >
>     > Here is a link to the discussion on the openjfx-dex list:
>     > http://mail.openjdk.java.net/pipermail/openjfx-dev/2016-August/019577.html
>     <http://mail.openjdk.java.net/pipermail/openjfx-dev/2016-August/019577.html>
>     >
>     > Here is a forum post describing a non-Java way to reproduce this same
>     > issue:
>     > http://www.gamedev.net/topic/679705-glxmakecurrent-slowly-leaks-memory/
>     <http://www.gamedev.net/topic/679705-glxmakecurrent-slowly-leaks-memory/>
>     >
>     >
>     > This was possibly not an issue in older versions of Mesa, as the bug
>     > does not appear on older Linux installations (I'm using Mesa 11.2.2,
>     > where the bug is present)
>
>     Does it still happen with current Git master? There have been some fixes
>     in this area recently.
>
>     If it still happens, the output of running an affected application in
>     valgrind --leak-check=full (with debugging symbols available for at
>     least all Mesa binaries) would be useful.
>
>
>     --
>     Earthling Michel Dänzer               |               http://www.amd.com
>     Libre software enthusiast             |             Mesa and X developer