Re: TLP driver in generic

[Richard Earnshaw <Richard.Earnshaw%buzzard.freeserve.co.uk@localhost>]

On Sat, 04 Jun 2005 19:02:35 BST, Chris Gilbert wrote:
> This is a multi-part message in MIME format.
> 
> --Multipart=_Sat__4_Jun_2005_19_02_35_+0100_VR=9NsS5rsjGBlhG
> Content-Type: text/plain; charset=US-ASCII
> Content-Transfer-Encoding: 7bit
> 
> (some of you might just about remember this thread)
> 
> On Sat, 8 Feb 2003 18:42:20 +0000
> Chris Gilbert <chris%dokein.co.uk@localhost> wrote:
> 
> > On Sun, 9 Feb 2003 03:26:58 +0900 (JST)
> > Izumi Tsutsui <tsutsui%ceres.dti.ne.jp@localhost> wrote:
> > 
> > > In article 
> > > <200302081804.h18I4F611076%buzzard.buzzard.freeserve.co.uk@localhost>
> > > rearnsha%buzzard.freeserve.co.uk@localhost wrote:
> > > 
> > > > Odd, I just get:
> > > > 
> > > > tlp0 at pci0 dev 9 function 0: DECchip 21140A Ethernet, pass 2.2
> > > > tlp0: interrupting at irq 8
> > > > tlp0: Ethernet address 00:40:05:51:3b:b2
> > > > tlp0: unable to configure MII
> > > > tlp0: no media found!
> > 
> > The fun side effect there is that you can panic the kernel very easily.
> > 
> > > Does the card work on i386 (or other machines),
> > > and what the part number of the card?
> > 
> > Probably will, mine does.  I've not had time to figure out why.
> 
> Well 2 and a bit years later I've mangaged to get the card to work with the t
> ulip driver.
> 
> It appears that the MII bitbang sync code doesn't always work on cats.  I've 
> managed to patch the mii bitbang code to do the sync slightly differently, it
>  actually does a clock after changing direction, and then does the 32bits of 
> 1 for the sync.  This means that the mii code actually detects the phy on thi
> s card as an lxtpty (which is correct and the same as it was in one of my x86
> ).
> 
Yay! I now get:

tlp0 at pci0 dev 9 function 0: DECchip 21140A Ethernet, pass 2.2
tlp0: interrupting at irq 8
tlp0: Ethernet address 00:40:05:51:3b:b2
lxtphy0 at tlp0 phy 0: LXT970 10/100 media interface, rev. 0
lxtphy0: 10baseT, 10baseT-FDX, 100baseTX, 100baseTX-FDX, auto

> I'm not sure exactly what the problem is.  I suspect that the phy/21140 needs
>  a kick before it's used.  Looking at the de card, it actually sends 32bits o
> f one, then 4bits of one, then 01 (start frame) then the command, so perhaps 
> it's a known oddity of the cards.
> 
> The IEEE spec requires that 32bits of 1 are sent to allow the phy to sync, so
>  I'm guessing that the phy doesn't think it's seen 32 bits for some reason.
> 
> The patch against -current is attached.  The problem is I don't have the kit 
> to test the patch out.  Can people try the patch on their cats boxen (or othe
> r machines with tulip cards in)
> 


Well, it certainly finds the PHY now.  However, I still have problems 
sending bulk data through the interface.  TCP channels keep deadlocking 
and failing to recover properly; and they don't achieve magnificent 
throughput when connected either (maxing out at about 3MByte/s).

However, I wonder if this is all part of the PCI/ISA malaise that this 
machine suffers from:
- wd0 keeps downgrading to UDMA 1
- ISA soundblaster card eventually locks up the machine when used
- TCP connections hang on the de/tlp card

I have a good-old 3com ep0 card in an ISA slot and that seems to work 
fine, but it's a shame that nothing else seems to.

R.