stix stix.id.au wrote:
> > Ah, I notice there is an obvious bug in
dp83932var.h.
> > I guess sonic chip on your LC card is in 16bit
mode, right?
>
> Yes, it is. And with that obvious bug fixed, it appears
to work
> fine,
Thanks, I'll commit it soon.
> although with a lot of FIFO overruns - probably
somewhat
> to be expected on an old system like this.
You could try to find optimized TX/RX FIFO threshold values
in if_sn_nubus.c with the sc->sc_dcr member. Old one
used
TX/RX = 16bytes/16bytes (DCR_TFT0|DCR_RFT1), but arc port
uses conservative values, 28bytes/4bytes
(DCR_TFT0|DCR_TFT1).
> Note: I left the previous patch in, too. Not sure if
that was the
> idea or not.
The DP83932C manual says in the CAM enable regsiter (CER)
section:
>> Under normal operations the user does not access
this register.
>> Instead the user sets up this register through the
last entry in
>> the CAM descriptor area. The SONIC loads the CE
register during
>> execution of the LCAM Command.
so I'll also commit the previous patch, which sets
"camvalid"
via LCAM command rather than writing CER directly.
---
Izumi Tsutsui
|