Re: Re: RealTek RTL8139C

["Nandakumar NarayanaSwamy" <nanda_kn@rediffmail.com>]

Hi Matti Aarnio,

Thanks for your response.

Our target board is little endian and some initial code is working 
which is compiled in little endian mode. Also the PCI bus will 
also work in little endian mode.

Thanks again for your response.

With best regards,
Nanda



On Tue, 13 Aug 2002 Matti Aarnio wrote :
>On Tue, Aug 13, 2002 at 10:09:56AM -0000, Nandakumar  
>NarayanaSwamy wrote:
> > Hi All,
> >
> > Sorry for disturbing the list again.
> >
> > I am using RTL8139C in our target board which is based on 
>MIPS
> > IDT32334 processor.
>...
> > My doubt is whether this 8139too.c is tested with MIPS 
>processors?
> > Because in one of the article i found that the supported
> > processors are ARM, i386 etc.
>
>   ( Are you sure you don't want to use  8139cp.c  driver ? )
>
>   It might.  The issue is most likely the ENDIANITY of command
>   registers, and how they are supported.
>
>   The  i386 is little-endian machine, and most hardware is made 
>for that.
>   However in case of embedded systems, there are very system 
>dependent
>   details no how the host processor accesses PCI-bus, and what 
>happens
>   then...
>
>   Talk with your harware maker.  At least the  8139*.c  drivers 
>use
>   cpu_to_le32() and friends for these byte-order mapping issues, 
>but
>   if there happens some gratuitious byte-order wrap-around when 
>posting
>   IO from the CPU to the PCI bus, then you have major problems, 
>and
>   will need experienced kernel hacker to get you thru...
>
>   That is, if the CPU does post a big-endian (data byte order) 
>operation
>   into the bus, and the cpu<->bus host-bridge will not do any 
>magic byte-
>   order wrap-around, then the driver should just simply work.
>
>
>   Of course with MIPS processors you can have two different 
>byte-orders
>   active in the system (one at the time, of course).  You need 
>to know
>   which you are using.
>
>
> > Can any body throw some light on this?
> > with best regards,
> > Nanda