Re: [Madwifi-devel] Making OpenHAL endianess-agnostic

[Nick K. <mic...@gm...>]

2007/2/24, Pavel Roskin <pr...@gn...>:
> Quoting Nick Kossifidis <mic...@gm...>:
>
> > > I think I know why madwifi-old-openhal is not working on PowerPC.
> > >
> > > There are some registers OpenHAL sets differently on big-endian
> > > machines.  Presumably, it allows using faster I/O without byte swapping.
> > > Check for BYTE_ORDER in the OpenHAL headers.
> >
> > Yup there is how you set that behaviour, it's used for descriptors etc
> > (part of the header i made and waiting for a yes or no to commit
> > it)...
>
> You can post it to the list first.  I prefer the approach when we convert one
> function at a time and actually test the new code, including constants.
>
> > #define AR5K_CFG_ADHOC                0x00000020      /*not found on 5210*/
>
> Such comments are not quite useful.  Do you know for sure that 5210 doesn't
> support it?  Do you know what that bit does?  I think it would be better to
> write about 5210 in a separate comment and try to provide the meaning of the
> value.  It's better it you put question marks there to encourage others to put
> the description there.

It says that this flag is not fount in 5210 registers header, i
haven't got a 5210 to know for sure, we'll have to fill in the gaps...

> > #define AR5K_CFG_TXCNT_S      11
>
> What is that?  How is it used?  I think we should limit the header to things we
> know and use.  We don't need things just picked up somewhere with description.
> I guess it's the shift value to guess the count, but I don't see it used.
>

All the _S are used by some macros, it's how many bits we shift (_M is
for mask).

> > It doesn't swap the registers, it swaps the descriptors etc that are
> > being directly assigned (pointers as function args) and i think that's
> > something we need to keep because swapping each descriptor/data we
> > get/send will have a huge impact. That's the only programming being
> > done on the card that has to do with endianess as far as i know.
>
> I don't understand why we need to store any function pointers in the hardware
> and why byte-swapping them would have a huge impact.  I would prefer to start
> with something that works and then do optimizations is needed.
>

O.K. let's start with what you think is better but i suggest that we
use that functionality in the future since the hardware supports it.

> > Do what you think is best, you know more than i do, I just believe
> > that swapping descriptors/buffers etc while the card can do this for
> > us is not wise, it's the register reads/writes we should work on.
>
> As I said, most drivers don't handle big-endian hosts specifically.  And I'll
> rather start with that.  It's better than to do optimization of something that
> is not working in hope that it would work.
>

ACK


-- 
As you read this post global entropy rises. Have Fun ;-)
Nick