JVM crash on 64bit Ubuntu

[avm1979]

Getting a jvm crash report from a player.

Apparently, the game starts up ok and then crashes when he tries to interact with the menu (not much to go on, but the crash is reproducible and happens at that exact point). The custom mouse cursor also appears to be corrupted (reports box around it).

The OS is: Linux 64bit Ubuntu 11.04. The video card is AMD's ATI 5770 HD, LWJGL version 2.7.1.

Would someone mind taking a look at it to see if it indicates where the problem might lie? Not having much luck making heads or tails of it. Thank you!

[Matzon]

Native frames: (J=compiled Java code, j=interpreted, Vv=VM code, C=native code)
C  [+0x430]  __kernel_vsyscall+0x10
C  [fglrx_dri.so+0x1837b23]
[error occurred during error reporting (printing native stack), id 0xb]

looks like we can't really do much about that ...

[avm1979]

Ok, thank you for looking at that! Next time I'll know just where to look in that wall of text.

[pjohnsen]

Not sure if it's much help, but from the log it looks like it's running on a 32bit JVM

[avm1979]

Hmm... that ought to be just fine, though, afaik? I've certainly run a 32 bit JVM on a 64 bit machine without any trouble.

What's werid, though, is that this player reports that it works using OpenJDK, but not Hotspot. For some values of "works" - there are other issues (cursor weirdness, sound (OpenAL) not working)

[dhdd]

Hi guys

I am seeing this problem for months now and I have narrowed it down to this:
a.) it only happens under 32bit
b.) it only happens with ATI graphics cards, no matter what catalyst version from 10.8 onward
c.) it happens with every Ubuntu version from 8.04 onward

Under Ubuntu 64 bit it just doesn't happen. I have been looking into this for months now and there are dozens of threads about that problem around the interwebz. No fix, because apparently everyone is just blaming another. You LWJGL guys just say that it cant possibly be LWJGL because in the stacktrace is nothing about LWJGL. Ubuntu developers say it must be ATI because they are calling things wrong and ATI says that Ubuntu is to blame after all they're at the top of the stacktrace.

Everybod should just pull their head out of their a**, stop blaming and start fixing.

EDIT:PS: it must have something to do with LWJGL because all the bugreports are from apps that use LWJGL. It doesn't have to be a bug in LWJGL but it is definitely involved in it.

ANOTHER EDIT: http://ati.cchtml.com/show_bug.cgi?id=17

[Matzon]

As much as I agree that the blameshifting game is a bit lame, it doesn't change the fact, that IF the call does not crash INSIDE the LWJGL native binary, then it's not our fault.

The crash happens because of this:

C  [+0x424]  __kernel_vsyscall+0x10
C  [fglrx_dri.so+0x18219d3]
C  [libpthread.so.0+0x5cc9]

so, basically fglrx_dri.so is making a syscall that is invalid.

Now, the big question is, why is it doing that?

Since LWJGL really doesn't do anything, except pass arguments onward to OpenGL, then its relatively fair to say that the issue must either originate from the caller, which in this case is minecraft.
If minecraft is doing something wrong, then the drivers should set a GL error flag and then handle that.
IT SHOULD NOT CRASH.

Since it is doing that, then the issue must be inside either fglrx_dri.so or whereever __kernel_vsyscall+0x10 is.

My bet is on fglrx_dri.so doing something wrong.

Also, if you look at: http://ati.cchtml.com/attachment.cgi?id=107, you will see that it also happens on non-LWJGL games. It does, however, happen on OpenGL games.

[dhdd]

Hi Matzon.

Thanks for clearing that up for me. Now I can tell the systems integration guys to start looking into Linux issues. 

And sorry for coming along rude. Last week was a bitch.

Greetings