Re: Buffer problems

["Ronald L. Sluiter" <sluiter@aps.anl.gov>]

The following is from an old copy of the "ei" driver, but it
may still be relevant.  As can be seen below, the "command
field frozen" error message is from a software timeout that
indicates the 82596 ethernet chip is unresponsive.

/* ei82596End.c - END style Intel 82596 Ethernet network interface driver */

/* Copyright 1989-2000 Wind River Systems, Inc. */
#include "copyright_wrs.h"

/*
modification history
--------------------
02a,11jun00,ham  removed reference to etherLib.
01z,27jan00,dat  fixed use of NULL
...      ...    ...
...      ...    ...
/*******************************************************************************
*
* ei82596Command - deliver a command to the 82596 via SCB
*
* This function causes the device to execute a command.  It should be called
* with interrupts from the device disabled.  An error status is returned if
* the command field does not return to zero, from a previous command, in a
* reasonable amount of time.
*/

LOCAL STATUS ei82596Command(END_CTRL * pDrvCtrl, UINT16 cmd)
{
   int loopy;
   SCB *pScb;

   pScb = pDrvCtrl->pScb;

   for(loopy = 0x8000; loopy--;)
   {
   if(pScb->scbCommand == 0)  /* wait for cmd zero */
       break;
   }

   if(loopy > 0)
   {
   pScb->scbCommand = cmd;    /* fill in command */
   sys596ChanAtn(pDrvCtrl->unit);    /* channel attention */
   return(OK);
   }
   else
   {
   logMsg("ei driver: command field frozen\n", 0, 0, 0, 0, 0, 0);
   return(ERROR);
   }
}




Mark Rivers wrote:

> Kevin,
>
> I suspect you are seeing the same problem we have seen at Brookhaven.
> You can try to reproduce this problem by sending lots of large ping
> packets quickly to the IOC.
>
> Here's how I've done it on Mac OSX.
>
>  
>
> > ping -s1500 -l3000 <ioc>
> >    
>
> On Linux I think the command is the same.  Note that you must be root to
> do this.
>
> With this command I have managed to irreversibly hang MVME2700 CPUs, so
> that they needed to be rebooted.  This happens something like 10-30% of
> the time that the ping command is issued.  Other times it causes long
> "delays", where the IOC is not capable of network activity, but then it
> recovers in 30 seconds or so.
>
> On MVME172 processors I have only managed to induce that latter
> behavior, i.e. delays, but not complete hangups requiring reboot.  But I
> have not done extensive tests of this.
>
> Mark
>
>