Real Time Kernel Recompile now segfaults

[jheiv]

I followed BehaartesEtwas guide ( http://wiki.fragaholics.de/index.php/EN:Linux_Kernel_Optimization ) on how to compile a RTK (its amazing) but now I am getting segfaults every 10ish hours.

Kernel:
Linux localhost 2.6.29.6-rt24 #5 SMP PREEMPT RT Mon Oct 19 00:16:52 EDT 2009 x86_64 x86_64 x86_64 GNU/Linux



I pretty much followed it to the t. And its worth noting that (although performance was much worse) before the recompile there were no segfaults.

Has anyone experienced segfaults after a recompile and can target it to a certain setting?

Maybe some things to back off on that won't hurt performance so much?


Server OS: Centos 5.3 x64
Processor: Intel® Core™2 Quad CPU Q8200 @ 2.33GHz
Ram: 8GB
Game(s): counterstrike source (srcds)
Start Up Command: ./srcds_run -game cstrike -ip 72.249.124.50 -port 27015 -maxplayers 16 -tickrate 66 +fps_max 0 +map de_dust2 +servercfgfile server.cfg -autoupdate
Admin Mods: None

The core dumps don't give any insight -- they're the exact same segfaults I was getting a month or so on an Fedora 11 x86 box.

Thanks in advance!

[BehaartesEtwas]

can you post the back trace of the segfaults nevertheless? (are they always the same or always different?)

you can change virtually any option without hurting performance much apart from the realtime preemption setting. but for testing you can even try a lower preemption level.

I have segfaults on my machine sometimes as well, but very irregular and only during map change. Some time ago I did some intense testing to find the problem. It even existed for a clean unmodified srcds installation on a fresh installed linux without any kernel patch. Recreating the .nav files helped a bit (even when no bots are on the server), but it didn't fix this completely...

[jheiv]

(10-21-2009, 10:06 PM)BehaartesEtwas Wrote:  can you post the back trace of the segfaults nevertheless? (are they always the same or always different?)

you can change virtually any option without hurting performance much apart from the realtime preemption setting. but for testing you can even try a lower preemption level.

I have segfaults on my machine sometimes as well, but very irregular and only during map change. Some time ago I did some intense testing to find the problem. It even existed for a clean unmodified srcds installation on a fresh installed linux without any kernel patch. Recreating the .nav files helped a bit (even when no bots are on the server), but it didn't fix this completely...

Thanks Behaartes -- I'll probably get to that this weekend and I'll make sure to report back. I also think I'm going to roll back to the kernel you have listed on your site (2.6.26.8-rt16 instead of 2.6.29.6-rt24) since you have had success with it.

[ulx]

I updated to 2.6.31.4-rt14 #1 SMP PREEMPT RT Thu Oct 22 11:39:46 CEST 2009 x86_64 and all is running stable and smoothly - so no need to rollback to a quite old one, I think.