> > 
> > 	Performance with the new driver shows that peek transfer rates are
> > 	now about 1.1 MB/s (using dd with 64K buffers) which is somewhere
> > 	between 5 & 10 times better than before (I've lost my original
> > 	figures).  In normal operation I've seen up to around 900K/s; but
> > 	probably more importantly, the load on the cpu is much lower than
> > 	it was before.
> 
> This is about the same performance that is possible with Acorn's own
> driver under RISC OS.
> 
> The card should be able to deliver a lot more on synchronous transfers
> if that is ever implemented.
> 
> According to the spec in the manual, the card should be able to deliver
> 1.5 MB/s on asynchronous transfers and around 4 MB/s on synchronous IIRC.
> 
> Please post this information to this list if you think it is useful.
> 

Now that is interesting.  I had assumed that I was being limited by the 
I/O performance of the StrongARM being throttled by it's inability to do 
sequential accesses to uncached memory areas.

Can you tell me which manual you are referring to, I haven't seen anything 
like this mentioned in the technical ref manual that I used while writing 
this (for the A540 I think -- it's at home so I can't look at the moment); 
it did mention something about peek io performance to the podule bus being 
about 6 MB/s and I do know that SA's and uncached areas don't get on very 
well, so I assumed I was hitting this limit, not another one.

I also saw the driver saying "now asynchronous" during the startup phase, 
so I assumed that it was dropping back to asynchronous mode after finding 
that synchronous wasn't supported for one of the devices on the chain.  
I'll have another look at the code this weekend to see if there is 
anything further I can do.

Thanks once again for the pointer,
Richard.