Latest if_bfe - I think it works!

[Mikko Työläjärvi]

On Wed, 3 May 2006, Mike Silbersack wrote:

> Ok, I got it working with an artificial 128MB limit.  I had to fix two major 
> issues:
>
> 1.  It was requiring more bounce buffers than busdma was prepared to give it, 
> so I reduced the size of the rings.
> 2.  It was leaking bounce buffers whenever you transmitted a multi-fragment 
> buffer, which pretty much means any non-ping packet.
>
> and scottl fixed #3 with rev 1.77 of busdma_machdep.c
>
> 3.  busdma wasn't telling us when we ran out of bounce buffers.
>
> However, now that #1 and #2 are fixed, #3 shouldn't come into play ever, as 
> far as I can tell.
>
> I'm writing this e-mail via ssh into pine on the machine which has the 
> artificial 128MB limit set, looking at hw.busdma sysctl I can see that it's 
> bouncing pages all over the place, so I can say that it seems to be good to 
> me.  Please test.

I concur.

Initial tests (on -current with a fresh kernel) running concurrent
cvsup, portupgrade, copying stuff back and forth over the LAN and
random web surfing works, as far as I can tell, without problems.
No panics, no error messages, no hangs, no hickups of any kind.

If it crashes, you'll be the first to know :-)

    Thanks,
    /Mikko

P.S. I'm pleased to the how much faster bfe gets the link up compared
      to the bcm4sbxp (ndis) driver.  If only someone could do the same
      to "re"... :-P