On Sun, Dec 23, 2007 at 04:08:13PM -0500, Steven M. Bellovin wrote:
> > 
> The problem appears to be from 1.150 of if_wm.c; when I revert it to
> 1.149, my speed doubles.  The commit message refers to 82543 and newer
> chips, and claims a 4x speed improvement.

No, it doesn't (claim a 4x speed improvement).  If your system is
dropping interrupts, however, this change make it considerably worse,
since there will now be many, many less interrupts (!).  If there's not
such a problem, I'm of course interested in fixing whatever is wrong,

One quick thing you could try is to set RADV, RDTR, TIDV, and TADV all
to zero, and let me know what happens.

> My problems are on an 82801H and a 82540EP.

What PHY is detected and attached for each of these?

Can you post a complete dmesg, please?

-- 
  Thor Lancelot Simon	                                     tls@rek.tjls.com

  "The inconsistency is startling, though admittedly, if consistency is to
   be abandoned or transcended, there is no problem."	      - Noam Chomsky