lkml.org 
[lkml]   [1998]   [Jul]   [23]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectNet code behavior under high load
Hi,

this is a enhanced version of a bug report posted to linux-smp today.

One problem hit me on a P166/96MB with kernel 2.0.35 and linuxthreads 0.7.
I can reproduce this failure to no end, so I think one may be concerned
about it.

It looks to me like there is some kind of overflow, because the IP is in
very abnormal code.

Code: 00000000 <_EIP>:
Code: 0 <_EIP+0>: 83 78 28 00 cmpl $0x0,0x28(%eax)
Code: 4 <_EIP+4>: 74 6b je 71 <_EIP+0x71>
Code: 6 <_EIP+6>: ff 05 b8 d1 1a incl 0x1ad1b8
Code: 11 <_EIP+11>: 00
Code: 12 <_EIP+12>: 8d b7 8c 00 00 leal 0x8c(%edi),%esi
Code: 17 <_EIP+17>: 00
Code: 18 <_EIP+18>: 90 nop
Code: 19 <_EIP+19>: 8b 00 movl (%eax),%eax
Code: 21 <_EIP+21>: 90 nop
Code: 22 <_EIP+22>: 90 nop
Code: 23 <_EIP+23>: 90 nop

Unable to handle kernel paging request at virtual address e2203a8f
current->tss.cr3 = 02f4b000, <r3 = 02f4b000
*pde = 00000000
Oops: 0000
CPU: 0
EIP: 0010:[__release_sock+17/136]
EFLAGS: 00010202
eax: 22203a67 ebx: 01bf10d8 ecx: 01bf11f0 edx: 0000006b
esi: 01bf1018 edi: 01bf1018 ebp: 00000001 esp: 026b4efc
ds: 0018 es: 0018 fs: 002b gs: 002b ss: 0018
Process webtest (pid: 20331, process nr: 328, stackpage=026b4000)
Stack: 01bf10d8 01bf1018 00000000 00145903 01bf1018 01bf1018 019abc40 019abc40
0014f3e5 01bf1018 00000000 019abc40 00000000 00135c08 019abc40 00000000
019abc40 ffffffe8 0014f03c 00000001 00136060 019abc40 9e7ffab0 00000000
Call Trace: [tcp_close+503/532] [inet_release+97/108] [sock_release+92/156] [inet_create+0/804] [sys_socket+268/308] [inet_create+0/804] [net_bh+9/284]
[sys_socketcall+252/636] [8390:ei_open+-125952/72] [system_call+85/128]
Code: 83 78 28 00 74 6b ff 05 b8 d1 1a 00 8d b7 8c 00 00 00 90 8b

This can be reproduced by running ftp://schell.de/pub/webtest-0.1.tar.gz
and start it with a thread number of at least 270. I ran the program two
hours with -c 250 and everything was fine. (only tested under 2.0.35)

If you need more info, further tests etc. let me know...

I've done a few more tests with webtest (run with up to 250 concurrent
threads). Here are the results for the server:

kernel card crash

2.0.35 SMP 3c900 1st) ~350000 2nd) ~650000
2.0.35 SMP Realtek (ne2k-pci) ~700000 hits - lockup
2.1.110 SMP 3c900 ~17000 hits - lockup
2.0.35 UP 3c900 still running (at 500000 now)

if(!lockup) a network restart would bring up the network interface again.

Client (P100) and server (2xP200/128MB) were connected over 10baseT.
Apache 1.3.0 was used as a server. I ran all above tests with vortex_debug
== 1, so I can't provide more info for the 3c59x driver. The `still
running' test is running with vortex_debug == 6, so syslogd takes ~30% of
cpu time ;)

Especially the result for 2.1.110 is terrifying. I didn't expect such a
bad one. Again, I would be willingly to do everything to help debug these
problems.

Bye,
Sascha




-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.altern.org/andrebalsa/doc/lkml-faq.html

\
 
 \ /
  Last update: 2005-03-22 13:43    [W:0.031 / U:1.052 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site