Stephen,

Do you have a version of the PXA2xx SPI contoller driver more recent
that the one posted at the end of October 2005?  There are some SPI (and
other) API changes required for 2.6.16-rc1.

I've attached my attempt (PIO works but DMA doesn't) if it's of any use.

The patch also:
  - includes support for the different clock on the PXA27x
  - calculates the correct clock divisor (at least on the PXA27x...)
  - allows null_cs_control for PIO transfer.

I'm currently using SSP3 on the PXA27x with the slave chip select GPIO
line configured as SSPSFRM3 instead of a GPIO.  This works fine provided
that each spi_message consists of a single spi_transfer.  With more than
 one transfer they're not back-to-back and SSPSFRM3 is deasserted
between transfers.

It looks like you're waiting for the transmit buffer in the controller
to empty before switching to the next transfer.  Is it possible to
switch to the next transfer immediatly upon exhausting the transfer's
tx_buf?  Perhaps (in the DMA case) by chaining several DMA descriptors
together?

At the higher SPI clock rates available on the PXA27x (up to 13 MHz with
the internal clock) PIO mode doesn't seem to feed the transmit buffer
fast enough resulting in gaps between each byte/word of the transfer.  I
would assume using DMA would not show this?

David Vrabel
-- 
David Vrabel, Design Engineer

Arcom, Clifton Road           Tel: +44 (0)1223 411200 ext. 3233
Cambridge CB1 7EA, UK         Web: http://www.arcom.com/