That's very strange. I have no idea why this would matter, but I wonder
if it's an issue of CPU affinity. For some reason, using taskset to
explicitly set the CPU affinity of srcds solved a couple of problems for
me in the past. I use it to set the affinity of L4D to both cores of my
Core2Duo like this:
taskset -c 0,1 ./srcds_run ...
Do you have a second box you could test this on to see if the problem
exists in two places?
- Dave
Chris Boot wrote:
> Dave,
>
> The core dumps are mostly useless, I can't get gdb to load the symbols
> at all and it just complains about invalid addresses. My kernel logs a
> stream of messages when I start the server though, for example:
>
> Oct 29 14:03:12 yuna kernel: [4204278.388117] srcds_linux[4377]:
> segfault at 297c2e7f ip f63862f0 sp fff3e500 error 4 in
> vphysics_linux.so[f61a5000+47e000]
> Oct 29 14:03:12 yuna kernel: [4204278.443767] srcds_linux[4368]:
> segfault at 990a160 ip f63a87c0 sp fff3e600 error 4
> srcds_linux[4373]: segfault at c19bb84b ip f6138011 sp fff3e410 error 4
> in datacache_linux.so[f60dc000+bb000]
> Oct 29 14:03:12 yuna kernel: [4204278.461375] srcds_linux[4371]:
> segfault at 6acad9d4 ip f613803d sp fff3e410 error 4 in
> datacache_linux.so[f60dc000+bb000]
> Oct 29 14:03:12 yuna kernel: [4204278.461375] srcds_linux[4370]:
> segfault at 67b066c ip f647e7bb sp fff3e6d0 error 4 in
> vphysics_linux.so[f61a5000+47e000]
> Oct 29 14:03:12 yuna kernel: [4204278.491647] srcds_linux[4378]:
> segfault at c324482e ip f6138069 sp fff3e410 error 4 in
> vphysics_linux.so[f61a5000+47e000]
> Oct 29 14:03:12 yuna kernel: [4204278.526002] srcds_linux[4380]:
> segfault at 5ecb65a2 ip f6138011 sp fff3e410 error 4 in
> datacache_linux.so[f60dc000+bb000]
> Oct 29 14:03:12 yuna kernel: [4204278.567814] srcds_linux[4381]:
> segfault at fcfd4e48 ip f63a8730 sp fff3e600 error 4 in
> vphysics_linux.so[f61a5000+47e000]
> srcds_linux[4379]: segfault at 1e39205f ip f647ea22 sp fff3e710 error 4
> in vphysics_linux.so[f61a5000+47e000]
> Oct 29 14:03:12 yuna kernel: [4204278.751870] srcds_linux[4395]:
> segfault at c9750ad7 ip f63862f0 sp fff3e500 error 4 in
> vphysics_linux.so[f61a5000+47e000]
>
> However, I've noticed that the crashed child processes restart
> immediately and seem to stay up, I now have 9/10 forks on my server full
> of players so they are clearly working (much to my consternation, see my
> sv_steamgroup_exclusive post). Very strange.
>
> Cheers,
> Chris
>
> David A. Parker wrote:
>> Signal 11 is a segmentation fault, so it looks like the child processes
>> are trying to perform some sort of memory access at an illegal address.
>> I'm afraid I'm at a loss as to why this would be happening. Do the
>> core dumps show anything odd?
>>
>> - Dave
>>
>> Chris Boot wrote:
>>
>>> David,
>>>
>>> I'm using the stock Lenny kernel:
>>> Linux yuna 2.6.26-2-amd64 #1 SMP Wed Aug 19 22:33:18 UTC 2009 x86_64
>>> GNU/Linux
>>>
>>> My system is completely up-to-date. Besides, surely kernel features
>>> should never cause a crash in userland unless userland is doing
>>> something very wrong...
>>>
>>> What I see is like this:
>>>
>>> bo...@yuna l4d2 $ ./demo_dir1_run.sh
>>> Auto detecting CPU
>>> Using SSE2 Optimised binary.
>>> Server will auto-restart if there is a crash.
>>> #
>>> #Console initialized.
>>> ***VPK: FastFindFile Attempting to use full path with VPK file!
>>> File:
>>> /srv/steam/l4d2/demo_dir1/left4dead2_demo/left4dead2/addonlist.txt
>>> #***VPK: FastFindFile Attempting to use full path with VPK file!
>>> File:
>>> /srv/steam/l4d2/demo_dir1/left4dead2_demo/left4dead2/addonlist.txt
>>> #Game.dll loaded for "Left 4 Dead 2 Demo"
>>> Server is hibernating
>>> ConVarRef test_progression_loop doesn't point to an existing ConVar
>>> Game supporting (2) split screen players
>>> maxplayers set to 18
>>> #01:Unknown command "mat_bloom_scalefactor_scalar"
>>> #02:Unknown command "mat_bloom_scalefactor_scalar"
>>> #03:Unknown command "mat_bloom_scalefactor_scalar"
>>> #01:---- Host_NewGame ----
>>> #04:Unknown command "mat_bloom_scalefactor_scalar"
>>> #02:---- Host_NewGame ----
>>> #03:---- Host_NewGame ----
>>> #04:---- Host_NewGame ----
>>> Child 0 aborted with signal 11
>>> #01:Unknown command "mat_bloom_scalefactor_scalar"
>>> #01:KeyValues Error: RecursiveLoadFromBuffer: got NULL key in file
>>> scripts/population.txt
>>> Child 1 aborted with signal 11
>>> #02:Unknown command "mat_bloom_scalefactor_scalar"
>>> #02:---- Host_NewGame ----
>>> #04:Host_NewGame on map c5m1_waterfront
>>> #03:Host_NewGame on map c5m1_waterfront
>>> Child 0 aborted with signal 11
>>> #01:Unknown command "mat_bloom_scalefactor_scalar"
>>> #01:Network: IP 87.117.228.64, mode MP, dedicated Yes, ports 27015 SV / 0 CL
>>> #01:Network console: IP 87.117.228.64 port 9001
>>> #01:---- Host_NewGame ----
>>>
>>> Notice all the "Child N aborted with signal 11" messages?
>>>
>>> Thanks,
>>> Chris
>>>
>>> David A. Parker wrote:
>>>
>>>> Hi Chris,
>>>>
>>>> Just wanted to let you know that I'm running a server with 6 forks on
>>>> 64-bit Debian Lenny and I am not seeing this problem. All six forks
>>>> stay up and there is no core dumping going on. My first guess would be
>>>> that your problem is related to your kernel configuration, such as high
>>>> resolution timers, dynamic ticks, etc. Is your Lenny installation up to
>>>> date? Are you using a custom kernel?
>>>>
>>>> - Dave
>>>>
>>>> Chris Boot wrote:
>>>>
>>>>
>>>>> Hi,
>>>>>
>>>>> Hmm, after more testing I can't seem to run any number of forks
>>>>> properly, with children core dumping regularly when I use a forked
>>>>> server... A non-forked server seems to run just fine. Anyone else have
>>>>> this problem?
>>>>>
>>>>> I'm using a 64-bit Debian Lenny server (HP DL140 G3).
>>>>>
>>>>> Can I run several instances out of the same install directory without
>>>>> using forks? E.g. running several different servers on different ports
>>>>> all out of the same directory?
>>>>>
>>>>> Cheers,
>>>>> Chris
>>>>>
>>>>> Chris Boot wrote:
>>>>>
>>>>>
>>>>>> Hi,
>>>>>>
>>>>>> I'm trying to run a forked server with 10 child processes, but if I try
>>>>>> and fork more than 2 children all the children crash out with a core
>>>>>> dump. Any ideas?
>>>>>>
>>>>>> My command-line is like this:
>>>>>>
>>>>>> ./srcds_run \
>>>>>> -console \
>>>>>> -game left4dead2 \
>>>>>> -debug \
>>>>>> -pidfile /srv/steam/l4d2/demo_dir1/left4dead2_demo/srcds.pid \
>>>>>> -fork 3 \
>>>>>> -netconport 90## -netconpassword <snip> \
>>>>>> +ip <snip> +hostport 27014+## \
>>>>>> +map c5m1_waterfront
>>>>>>
>>>>>> My server.cfg is like this:
>>>>>>
>>>>>> hostname "<snip>"
>>>>>> rcon_password "<snip>"
>>>>>> sv_allow_wait_command 0
>>>>>> sv_alltalk 0
>>>>>> sv_alternateticks 0
>>>>>> sv_cheats 0
>>>>>> sv_clearhinthistory 0
>>>>>> sv_consistency 1
>>>>>> sv_contact "[email protected]"
>>>>>> sv_downloadurl ""
>>>>>> sv_lan 0
>>>>>> sv_log_onefile 0
>>>>>> sv_logbans 1
>>>>>> sv_logecho 0
>>>>>> sv_logfile 0
>>>>>> sv_logflush 0
>>>>>> sv_logsdir "logs"
>>>>>> sv_maxcmdrate 100
>>>>>> sv_unlag 1
>>>>>> sv_maxunlag .5
>>>>>> setmaster add 68.142.72.250:27011
>>>>>> setmaster add 72.165.61.189:27011
>>>>>> sv_steamgroup <snip>
>>>>>> sv_steamgroup_exclusive 1
>>>>>>
>>>>>> Many thanks,
>>>>>> Chris
>>>>>>
>>>>>> Jason Ruymen wrote:
>>>>>>
>>>>>>
>>>>>>
>>>>>>> As you've already seen, the dedicated server files for the Left 4 Dead
>>>>>>> 2 Demo are now available. The game name is left4dead2_demo. We're
>>>>>>> still doing some final checks for the client, but that should be going
>>>>>>> live soon.
>>>>>>>
>>>>>>> Jason
>>>>>>>
>>>>>>>
>>>>>>> _______________________________________________
>>>>>>> To unsubscribe, edit your list preferences, or view the list archives,
>>>>>>> please visit:
>>>>>>> http://list.valvesoftware.com/mailman/listinfo/hlds_linux
>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>>
>>>>>> _______________________________________________
>>>>>> To unsubscribe, edit your list preferences, or view the list archives,
>>>>>> please visit:
>>>>>> http://list.valvesoftware.com/mailman/listinfo/hlds_linux
>>>>>>
>>>>>>
>>>>>>
>>>>> _______________________________________________
>>>>> To unsubscribe, edit your list preferences, or view the list archives,
>>>>> please visit:
>>>>> http://list.valvesoftware.com/mailman/listinfo/hlds_linux
>>>>>
>>>>>
>>>>>
>>>>
>>>>
>>> _______________________________________________
>>> To unsubscribe, edit your list preferences, or view the list archives,
>>> please visit:
>>> http://list.valvesoftware.com/mailman/listinfo/hlds_linux
>>>
>>>
>>
>
> _______________________________________________
> To unsubscribe, edit your list preferences, or view the list archives, please
> visit:
> http://list.valvesoftware.com/mailman/listinfo/hlds_linux
>
--
Dave Parker
Utica College
Integrated Information Technology Services
(315) 792-3229
Registered Linux User #408177
_______________________________________________
To unsubscribe, edit your list preferences, or view the list archives, please
visit:
http://list.valvesoftware.com/mailman/listinfo/hlds_linux
