Re: CF Ethernet performance?

From: Nicolas Pitre <nico.a.t.cam.org>
Date: Tue Oct 10 2000 - 00:31:34 EDT

On Mon, 9 Oct 2000, Russell Nelson wrote:

> John Dorsey writes:
> > All CF ethernet cards are not slow. The Socket LP-E CF+ card is used
> > almost universally by kernel developers on the SA-1110 and SA-1111
> > development boards. Even with the various bug workarounds for that
> > card and the PCMCIA implementation on the SA-1111 board, I get
> > error-free, low-latency ping response on a production ethernet LAN.
>
> I see now that it's not a speed thing. The ping fails on certain
> packet lengths but not others. I'd guess it's either a driver or
> hardware problem. Probably a driver problem.

Hey guys!

I probably found it.

I was quite anoyed by my Socket LP-E CF+ card always spitting "eth0:
interrupt dropped" at every 10 seconds or so while playing MP3's over NFS,
accompanied with the corresponding audio hicks of course.

So I decided to re-visit the underlying low-level SA1100 interrupt
handling code of mine... yet just to find a possibility of loosing
interrupts... and guess what? ;-)

I fixed it and now my network card is spinning packets __a lot faster__
without any lost interrupt anymore.

This problem might have affected any CF or PCMCIA devices on iPAQ and
Assabet (not Neponset), whether it is network cards, ATA cards, etc.

The fix is included in diff-2.4.0-test8-rmk5-np2 that I'm releasing at the
moment.

Russell: Please tell me when you have the chance to test it if your card
performs better.

Nicolas
Received on Mon Oct 9 21:24:23 2000

This archive was generated by hypermail 2.1.8 : Tue May 04 2004 - 09:42:17 EDT