network hangs I/O system?

Nick.SIMON at syntegra.com Nick.SIMON at syntegra.com
Mon Apr 2 07:01:46 UTC 2001


Is it possible your system is performing a RTEMS-panic?

If the network stack runs out of MBUFs, it tries to allocate more;
unfortunately, as these can only be allocated at startup, the system panics
instead.  This could be down to heavy network traffic tying up buffers.  If
you're really unlucky, you could be leaking mbufs :=(   Compare your network
code with that in 4.5.0 - I think there are probably a fair number of
changes.

I have bodged our system to lose frames rather than stop.  I can dig out
these changes for you if necessary (but not for 4.0.0!) but you may like to
increase the number of buffers and clusters first.


-- Nick Simon 

> -----Original Message-----
> From: Thomas Doerfler [mailto:Thomas.Doerfler at imd-systems.de]
> Sent: 02 April 2001 07:42
> To: rtems-users at OARcorp.com
> Subject: network hangs I/O system?
> 
> 
> Hello,
> 
> in my current project my customer has found out a very strange 
> behaviour: I have built a system based on RTEMS-4.0.0 including 
> ethernet networking, hardware is the MC68360. 
> 
> When my customer powers my systems up with disconnected network, they 
> run fine for many days. When my customer connects my systems to his 
> plant network (with rather heavy load and a high broadcast rate), 
> then the systems hang after some time, although they themselfs are 
> more or less idle (waiting for a connect from the network). It takes 
> between 2 hours and 5 days, before the systems hang. Some internal 
> tasks still seem to run (e.g. task to retrigger of software watchdog 
> :-(( ), but all the I/O activity (including serial V.24 devices) is 
> inactive. (Even a disconnect from the network does not wake the 
> systems up again, only a power cycle helps.)
> 
> In my local office network (with much lower traffic) I could not 
> repeat this error. 
> 
> Now my questions: 
> 
> - Did anybody out there find the same behaviour during development 
> (or test phase)?
> 
> - Is it possible that the "multiple event bug" discussed recently (by 
> Zoltan etc.) can cause this behaviour concerning the network task?
> 
> - Is it possible that the network task falls into a condition so that 
> it consumes all CPU performance, even when there is no more traffic 
> on the network connector?
> 
> - Is there any "global I/O" semaphore that might block the whole I/O 
> system (networking, serial termios drivers AND serial non-termios 
> drivers)?
> 
> Any hints welcome :-)
> 
> Bye,
> 	Thomas.
> 
> --------------------------------------------
> IMD Ingenieurbuero fuer Microcomputertechnik
> Thomas Doerfler           Herbststrasse 8
> D-82178 Puchheim          Germany
> email:    Thomas.Doerfler at imd-systems.de
> 


******************************************************************************

Check us out at http://www.syntegra.com

***********************************************************************



More information about the users mailing list