am29xxxxxv2: Support for buffered writes

["Rene Nielsen" <rbn at vitesse dot com>]

The attached patch adds support for buffered writes to the am29xxxxx V2
driver.
If cyg_am29xxxxx_init_cfi() is used to initialize the device, then the
write buffer size is auto-detected.
If the flash supports buffered writes, then am29_hw_program_buffered()
will get called, otherwise the "good old" am29_hw_program() will be
called.

I have only tested it with a flash hooked up in 16as8 mode, so please
scrutinize the changes before applying, and - if possible - test with
other flash combinations.
Perhaps I should make a CDL option that guards the use of buffered
writes (default 'off') until it's been verified by the community.

I have two questions:
1) Why is data programmed to the flash verified by both
   cyg_am29xxxxx_program() and cyg_flash_program()? Isn't
   that a waste of CPU cycles?

2) For am29xxxxx, the number of retries when programming
   is given by CYGNUM_DEVS_FLASH_AMD_AM29XXXXX_V2_PROGRAM_TIMEOUT.
   This is OK, but what bothers me is that this is *not* the number
   of retries per written buffer or written word, but the number
   of retries per call to cyg_am29xxxxx_program(), whether 1 byte
   or a full sector is being written. This means that if a flash
   sector is bigger than
CYGNUM_DEVS_FLASH_AMD_AM29XXXXX_V2_PROGRAM_TIMEOUT
   then it's not possible to write such a sector in one go.

Regards,
Rene Schipp von Branitz Nielsen 
Vitesse Semiconductors

Attachment: am29xxxxxv2_2009_04_24.patch
Description: am29xxxxxv2_2009_04_24.patch