[tulip] interupt mitigation on a DFE-570TX - where's the secr
Fri, 30 Mar 2001 14:48:55 -0800
Great, that hack worked - e.g. it got the mitigation message in my syslog.
Now the question is - how can I tell it's really working?
My issue is that once this box (it's a linux router/firewall) sees about
20Kpps, it starts really bogging down. It also shows me about 22K int/s (as
shown by 'vmstat -3'). My hopes are that interrupt mitigation reduces the
number of int's bogging down the CPU, and allows it to route more
Does that make sense?
> -----Original Message-----
> From: Donald Becker [mailto:email@example.com]
> Sent: Thursday, March 29, 2001 12:21 PM
> To: Roisman, Dani
> Cc: Tulip List
> Subject: Re: [tulip] interupt mitigation on a DFE-570TX - where's the
> secret sauce?
> On Fri, 23 Mar 2001, Roisman, Dani wrote:
> > I'm using version 0.92t of the tulip driver, and it isn't
> performing as if
> > it's doing interrupt mitigation. I'm comparing the pps
> that I'm sending it
> > to the interrupt/sec info that I get from vmstat or looking at
> > /proc/interrupts.
> > Browsing through tulip.c, this line leads me to believe
> that I should see a
> > message sent to syslog (I'm logging *.*) about interrupt
> mitigation when I
> > load the driver:
> > printk(KERN_NOTICE "%s: Set interrupt mitigate paramters %8.8x.\n",
> There are two conditions when the interrupt mitigation
> register is set:
> When the chip attempts to do too much work during a single
> You will see the infamous "Too much work at interrupt" message.
> When set by an ioctl() call
> You will see the message above.
> > Is there something I'm missing???
> It's a hack, but the ioctl() call has been put into the mii-diag
> program from
> You can set interrupt mitigation by running
> mii-diag eth0 -G 0x8b240000
> Donald Becker firstname.lastname@example.org
> Scyld Computing Corporation http://www.scyld.com
> 410 Severn Ave. Suite 210 Second Generation
> Beowulf Clusters
> Annapolis MD 21403 410-990-9993