Updated tg3 driver available

Jeff Garzik jgarzik at redhat.com
Tue Jan 21 14:59:00 CST 2003

On Tue, Jan 21, 2003 at 12:53:55PM -0800, Les Niles wrote:
> Jeff Garzik [mailto:jgarzik at redhat.com] wrote:
> > ...
> > For bcm5700 users, if it's working for you, fine and dandy.  I am a
> > Southerner, who believes strongly in the "if it ain't broke, don't fix
> > it" mantra.  However, just to reiterate, there continue to be 
> > known bugs
> > in the bcm5700 driver, over and above the slower performance 
> > it offers.
> > The internal driver locking functions are incorrect and could lead to
> > stack corruption at high loads.  The hardware errata workarounds are
> > incredibly heavy handed, and serve to severely limit performance in
> > exchange for working around hardware errata.
> Since this is a 10/100/1000 interface, does "high load" mean high 
> relative to 1 Gb/s, or high relative to the link-level max speed?  In 
> other words, are the bcm5700 bugs likely to be hit on a 100baseT network? 

This is a good question.  Link-level speed does indeed make a
difference.  It is _possible_ that these problems will be seen on
100baseT, but unlikely.  It is far more likely that gigabit speeds and
jumbo packets, or higher overall interrupt traffic on the box, will
trigger failures.

> Likewise, do the hardware errata work-arounds significantly degrade 
> performance when the interface is used for 100baseT? 

The hardware errata workarounds are "heavy-handed" only in the bcm5700
driver.  Thus, they only affect performance in bcm5700, not in tg3.
But to answer your question, yes, in bcm5700, performance at 100baseT is

Anything less than wire speed is unacceptable ;-)


More information about the Linux-PowerEdge mailing list