Bug 59904

Got a similar problem with Half-Life with Steam-linux

(In reply to comment #0)
> Created attachment 73689 [details]
> crash backtrace
> 
> A regression was introduced into Mesa that causes 3DMark2000 to crash in the
> high detail helicopter benchmark. A bisect revealed this patch:
> 
> 68a097596efdf314ec60487a1761cac789ae2367 is the first bad commit
> commit 68a097596efdf314ec60487a1761cac789ae2367
> Author: Brian Paul <brianp@vmware.com>
> Date:   Thu Jan 24 14:44:09 2013 -0700
> 
>     util: add some defensive coding in u_upload_alloc()
> 
> The terminal output when the crash occurs is this:
> 
> radeon: mmap failed, errno: 12
> radeon: mmap failed, errno: 12
> wine: Unhandled page fault on read access to 0x00000034 at address
> 0x7cdd1e9b (thread 001f), starting debugger...
> 
> A backtrace of the crash is attached. I do not know where exactly the "mmap
> failed" lines are written.

+1 

I just bisected to this commit after getting the same running nexuiz on rv790 (with and without llvm).

Extra info from dmesg -

[drm:r600_cs_packet_next_reloc_mm] *ERROR* Relocs at -1073723904 after relocations chunk end 4 !
radeon 0000:01:00.0: bad SET_CONTEXT_REG 0x28980
[drm:radeon_cs_ib_chunk] *ERROR* Invalid command stream !

I started to get regular OOM-kills after that commit when i try to play minecraft. Memory usage with openarena was increased too. Looks like everything is ok with just compositing desktop without running other 3d apps. I have hd4330 (rv710)

Also OOM kills here, bisected:
68a097596efdf314ec60487a1761cac789ae2367 is the first bad commit


in Xonotic 64 bit and Spring 64 bit games after several minutes of play I run out of my 8GB ram.

(In reply to comment #2)
> (In reply to comment #0)

> 
> +1 
> 
> I just bisected to this commit after getting the same running nexuiz on
> rv790 (with and without llvm).

My issue is fixed by mesa commit -


commit 87592cff57feef29565150b9203e220b50623f30
Author: Marek Olšák <maraeo@gmail.com>
Date:   Mon Jan 28 02:47:24 2013 +0100

    gallium/u_upload_mgr: fix a serious memory leak
    
    It can eat all memory and crash in a matter of minutes with r600g.

for me also, thanks

I'm seeing this with 72916698b056d0559263e84372bb45cd83a1c2c2 and an HD6870

radeon: mmap failed, errno: 12
radeon: mmap failed, errno: 12
Crash log saved as '/home/bruno/.Heroes of Newerth/game/crash_2.6.35.0_00.log'
Segmentation fault

72916698b056d0559263e84372bb45cd83a1c2c2 contains the patch that it is supposed to fix this issue, but it happens to me from time to time while playing HoN.

Though I have nothing in dmesg...

(In reply to comment #7)
> I'm seeing this with 72916698b056d0559263e84372bb45cd83a1c2c2 and an HD6870
> 
> radeon: mmap failed, errno: 12
> radeon: mmap failed, errno: 12
> Crash log saved as '/home/bruno/.Heroes of
> Newerth/game/crash_2.6.35.0_00.log'
> Segmentation fault
> 
> 72916698b056d0559263e84372bb45cd83a1c2c2 contains the patch that it is
> supposed to fix this issue, but it happens to me from time to time while
> playing HoN.
> 
> Though I have nothing in dmesg...

Are you running a 32-bit app on a 64-bit distro?  If so, make sure you've also updated the 32-bit version of the 3D drivers.

Running 64 bit and the game is 64bit.

With mesa in debug mode I get this:

Mesa: User error: GL_INVALID_ENUM in glBindTexture(target)
Mesa: 85 similar GL_INVALID_ENUM errors
Mesa: User error: GL_INVALID_OPERATION in glFramebufferTexture2DEXT(non existant texture)
radeon: mmap failed, errno: 12
Crash log saved as '/home/bruno/.Heroes of Newerth/game/crash_2.6.35.0_02.log'
Segmentation fault

context mis-match in pipe_sampler_view_release()

The 3 first line seem to appear at the game start, so it's mprobabl not the problem. But the "context mis-match in pipe_sampler_view_release()" might be important.
I can't seem to get a backtrace even with -g though ...

OK, it looks like it only happens when I'm on linux 3.8, it's OK when running linux 3.7.
Also when using 3.8 it does look like a memory leak too ! So I experience the same symptoms.

I'm not seeing the problem anymore. I guess it has been fixed by http://cgit.freedesktop.org/mesa/mesa/commit/?id=5c86a728d4f688c0fe7fbf9f4b8f88060b65c4ee .
Closing.