AT91 patches

[Jonathan Larmour <jifl at eCosCentric dot com>]

I have been working on trying to get the AT91 changes incorporated for the 
last while. The big problem, as described before, is the flash driver. 
Thomas's patch appeared to add support for different flash sector sizes. 
However it impacted flash device access in redboot that didn't need this.

Now, support for differing flash sector sizes is a good thing when the 
hardware has it. The obvious benefit is the reduced overhead for FIS 
directories and fconfig areas.

With that in mind I tried to do what I could to fix this by making it 
configurable at least. To that end I made a bunch of changes in the 
generic driver layer that were fine. I also made quite a few changes 
analogous to what Thomas did in the redboot flash.c.

However a few things became apparent:

* I was making more and more changes that would be quite difficult to test 
in the absence of any hardware with differing sector sizes (although I 
realised I could spend some time getting the synth flash driver to pretend 
to do this). What's more the changes could well break normal targets, not 
just the AT91. That's unacceptable.
* RedBoot's flash code has the idea of fixed block sizes very very 
ingrained. All the code can be changed only by breaking lots of 
fundamental design assumptions. It mixes numbers of sectors with sizes of 
sectors freely, including in CDL options and configuration from the drivers.
* The more changes I had to make, the more it would end up being as 
invasive to fixed sized flash as Thomas's patch, just in a different way.
* Thomas's patch wasn't complete anyway. It would loosely appear to work, 
but things would go wrong when flash limits exceeded certain boundaries 
(primarily where the sector size switches) or you want control over what 
goes into what sized blocks. It would end up as a somewhat schizophrenic 
RedBoot. It's fundamental really because RedBoot would still be oriented 
towards a fixed size.

So my conclusion is that I can't see any way to reasonably apply the 
existing flash driver with "true" mixed sector sizes at all. The only 
sensible way to do it in the near term is to change the driver to use the 
same bootblock system as the existing drivers, which means they are read 
and written in chunks of the "larger" sector size.

What's more that isn't something I can do either, not having an AT91 EB40 
board to test something like this. So unless someone out there is willing 
to donate one, I can't see the flash driver being included for a while 
until someone with one does it.

I can do the rest of the eb40 port (and probably will soon), but without 
the flash driver, it's somewhat crippled in terms of redboot support.

Suggestions?

In general, my recommendation is that in the not too distant future we 
will need a new flash system, designed from scratch to support multiple 
different devices any or each with different sector sizes, and NAND flash. 
To be honest, that requires a rewrite, mostly of the RedBoot code, and I 
can't foresee any compatibility with the existing hardware flash drivers 
as the layering isn't too good - too many dependencies and assumptions 
between the layers (e.g. passing _down_ the number of blocks in the device 
from the generic driver to the hardware one, and using the single 
flash_info global). At least not without a painful number of ifdefs. 
Instead it would be a big switch between the old flash system and a new 
one - they could exist in the codebase at the same time, but new low level 
drivers would be needed for the new one. Based on the old ones for sure, 
so it's not really starting over, but the structure would be different.

Of course no-one's likely to write such a system unless there's enough 
impetus, so I'm not going to hold my breath.

Jifl
--
eCosCentric       http://www.eCosCentric.com/       <info@eCosCentric.com>
--[ "You can complain because roses have thorns, or you ]--
--[  can rejoice because thorns have roses." -Lincoln   ]-- Opinions==mine