2.6.6-rc1 caused dedicated Quake 3 server to core dump
For years now, I've been running a dedicated Quake 3 server on my linux box. Last night, was the first time I've ever seen a core dump with Quake. 2.6.5 was and is now running fine, but under 2.6.6-rc1 it core dumped within about 30 minutes with the attached dmesg output. I am also including my config's for 2.6.5 and 2.6.6-rc1. The only things I'm aware of as being different is CONFIG_POSIX_MQUEUE=y and the relabeling of usbhid (2.6.6-rc1) from just hid (2.6.5). If any additional information is needed, please let me know. Also, please CC as I'm no longer on this mailing list.
Thanks,
Steve
Am Sa, den 17.04.2004 schrieb Stephen Lee um 11:14 Uhr -0500:
Quoted 4 lines
For years now, I've been running a dedicated Quake 3 server on my linux box. Last night, was the first time I've ever seen a core dump with Quake. 2.6.5 was and is now running fine, but under 2.6.6-rc1 it core dumped within about 30 minutes with the attached dmesg output.
Could this be something similar to what I've been seeing?
Hmm, some days ago I compiled a program and the resulting binary was broken. It always crashed at a certain point. I recompiled it and then everything was fine. I've never seen this before on this machine. It is 2.6.5 patched with 2.6.5-mcsomething + some patches from 2.6.5-mmsomething (I don't remember exactly, but all of the VM changes and driver updates, nearly everything except the CPU scheduler changes and the actual objrmap patches). The machine was under extreme swap load while compiling the program and that resulted in a broken binary.
I had tested another kernel before (with some 2.6.4-rc3-mmsomething patches) and I tried to compile php and it failed because a source file had 3*1024 bytes of nullbytes in it after unpacking the tar file. I think the machine was swapping too.
The patches in in 2.6.6-rc1 that are actually being used on the machine are (from jgarzik's RFT mail):
* non-exec stack support
* much better block I/O unplugging (I/O scalability)
* lots of VM work (often related to I/O or Hugh's rmap/anonvma stuff)
* queue congestion hooks
* DM, MD fixes (some related to the queue congestion/unplugging changes) (and other patches from the unstable dm tree, but there are no other core changes)
* reiserfs fixes and features (+ some other of Chris Mason's patches)
* readahead tweaks and fixes
* writeback tweaks
* direct-IO and AIO fixes and speed-ups
This is a IBM ThinkPad T40 and the hardware is absolutely stable. config attached.
Stephen Lee <[email protected]> :
Quoted 4 lines
For years now, I've been running a dedicated Quake 3 server on my linux box. Last night, was the first time I've ever seen a core dump with Quake. 2.6.5 was and is now running fine, but under 2.6.6-rc1 it core dumped within about 30 minutes with the attached dmesg output. I am
Which binary module are you using ?
--
Ueimor
Ah, I meant to include that in my first email. It's the latest NVidia driver 1.0-5336. I know, I know, but it's worked fine with 2.6.5 and all previous. Also, just the dedicated quake 3 server core dumped, the quake 3 client running on the same machine did not crash. At the time of the crash, there were four clients connected to the quake 3 server. I realize the NVidia driver could have corrupted some memory some where else, but if that's the case, I would have thought it would have shown this behavior previous, and not just with 2.6.6-rc1. Unless of course, something in 2.6.6-rc1 specifically altered something the NVidia driver makes use of. I guess. :-)
Steve
-----Original Message-----
From: Francois Romieu [mailto:[email protected]]
Sent: Saturday, April 17, 2004 1:00 PM
To: Stephen Lee
Cc: [email protected]; [email protected]
Subject: Re: 2.6.6-rc1 caused dedicated Quake 3 server to core dump
Stephen Lee <[email protected]> :
Quoted 4 lines
For years now, I've been running a dedicated Quake 3 server on my linux box. Last night, was the first time I've ever seen a core dump with Quake. 2.6.5 was and is now running fine, but under 2.6.6-rc1 it core dumped within about 30 minutes with the attached dmesg output. I am
Which binary module are you using ?
--
Ueimor
Quoted 11 lines · Steve Lee
Ah, I meant to include that in my first email. It's the latest NVidia driver 1.0-5336. I know, I know, but it's worked fine with 2.6.5 and all previous. Also, just the dedicated quake 3 server core dumped, the quake 3 client running on the same machine did not crash. At the time of the crash, there were four clients connected to the quake 3 server. I realize the NVidia driver could have corrupted some memory some where else, but if that's the case, I would have thought it would have shown this behavior previous, and not just with 2.6.6-rc1. Unless of course, something in 2.6.6-rc1 specifically altered something the NVidia driver makes use of. I guess. :-)
Hi Steve,
Unfortuately the nvidia driver means nobody is likely to look into your problem - it is just not a good use of their time. It would be very helpful if you could try to reproduce the oops without the nvidia driver loaded though.
It should be possible if the nvidia driver isn't the cause of the problem.
Nick