[U-Boot-Users] [PATCH] use CFI-Flash and board depended driver together

[Tolunay Orkun]

Tolunay Orkun wrote:
> Jens Scharsig wrote:
>> Hello,
>>
>> i need to use cfi and board flash driver together. This Patch contains a 
>> simple way to do this with a minimun changes at cfi_flash.c. All changes 
>> will only enable if CFG_FLASH_BOARD_DRIVER defined.
>>
>>
>> CHANGELOG
>>
>> * Rename flash_print_info, flash_erase, write_buff and
>>   flash_real_protect to cfi_xxxx, if  CFG_FLASH_BOARD_DRIVER is
>>   defined. This will allow use cfi driver together with board a
>>   dependent flash driver (see README.board-dependent-flash for
>>   an example)
> 
> I have to object to this patch. I will be providing a testing patch 
> tonight or tomorrow that will address this issue properly which is 
> teaching cfi_flash driver to deal with non-cfi flash. There is no need 
> to have two flash drivers...

I've started with this patch but ran into some roadblocks so it did not 
happen last friday as promised. I have been thinking better solutions that 
avoids two sets of flash drivers.

Basically, I was trying to detect the port and chip width by using a 
technique similar to flash_detect_cfi() but instead of checking for 
combinations of Q R Y etc., I was trying to see there was a change in the 
first 16 bytes of flash. Well that seemed workable until I actually tried 
the modified code :( The probe this way failed miserably with Intel flash 
responding to 0x90 in either in upper or lower byte even in 16-bit mode!

Thus, if the flash is not detected by CFI method the portwidth and chipwidth 
will have to be set manually for even jedec ids to be read properly. If the 
jedec probe worked, the jedec ids could be used to lookup a static list for 
proper configuration of info structure.

I have not given up yet. I would like to start over and take a simpler 
approach. Here is the idea:

The CFI driver will do CFI probe. If the probe is not successful and board 
configuration has defined a macro, CFI driver will call a board specific 
function to setup the flash_info_t for that bank.

I will pass the bank number to the function as argument. If the function 
returns non-zero, I will assume that board has setup the info structure 
successfully (hopefully) and move on. This still avoids driver code 
duplication by only passing the responsibility of setting up the info 
structure to the board.

A variation of this could be the callback function obtains the portwidth, 
chipwidth (and possibly the command set) for the specific bank and do 
flash_read_jedec_ids to get the ids and lookup the values from a statically 
compiled list like I originally intended. This list would be compiled in 
compiled only if this feature is enabled, say CFG_FLASH_CFI_JEDEC. Instead 
of a callback we could use the flash_base address list and augment the array 
to an array of structures which include base, portwidth and chipwidth but 
that would require a change in all boards that uses CFI driver. A callback 
to the board supplied function is the least invasive way in my opinion.

In either method, to cut the compiled image size a bit if there is 
definitely no cfi compliant flash on any bank we can omit the cfi detection 
and other cfi related setup code using a macro like CFG_FLASH_CFI_DISABLE 
(or something like that).

I would like to hear opinions regarding which choices looks more attractive...

Tolunay