On 02/08/11 10:19, Wolfgang Denk wrote:
Dear Richard Retanubun,
In message4D515D06.7020209@RuggedCom.com you wrote:
If by "how NOR flash in handled" you mean cmd_flash.c::flash_sect_roundb() then yes, I do want to do this but I need to know what number to round it to, no?
The code needs to know that, you don't ;-)
In the context of cmd_flash, it has access to flash_info_t which contains the sector_size information. Also cfi_flash.c also provides a function called flash_sector_size() that is called by other components that needs to know. If I have misunderstood what you mean, sorry, please clarify.
You are right.
<sensing a wavering in your resolve> ;-)
Thus we come to my original question, in the same way that cfi_flash.c provides flash_sector_size();
I propose we add spi_flash.c::spi_flash_sector_size() which returns the same information so that other code can validate (or round up if requested) the length of the operation to the nearest sector size.
right now (unless I misunderstood), there is no easy way to get the spi flash sector_size, no?
would this be acceptable?
To have SPI flash do this, for example; stmicro.c::stmicro_erase() function will auto round up to the nearest sector_size. Which is indeed the simpler way. Right now, most drivers/mtd/spi/*.c just does this check:
<quoted_code> sector_size = stm->params->page_size * stm->params->pages_per_sector;
if (offset % sector_size || len % sector_size) { debug("SF: Erase offset/length not multiple of sector size\n"); return -1; }
</quoted_code>
This is ok if an explicit size is given - you will see the same behaviour on NOR flash when trying something like
erase 40000000 40002345
Understood. So the spi flash driver erase function itself should only validate the args and not auto round up. This means that the decision of [warn or auto-round-up] should be handled by other code that parses the user's commands that is operating on spi flash.
To do this, they need a way to ask the spi flash driver what the sector size to round up to is.
I am worried about unintentionally erasing more than what the caller requested. Would it be better to round up manually and then calls the erase?
No. If the caller uses the "+<size>" notation he explictly requests to round up. He is supposed to know what he is doing.
Being able to use "+${filesize}" allows for many powerful solutions to standard tasks that without this would result in cumbersome manipulation of the size - which would then probably break if you issue a new revision of your hardware with different flash chips fit.
This is a cool convention. Should we add this to "sf erase command?" if it is not there already. if so, what command (or better yet, git commit) shows how to add handling of "+N"
Thanks!
- Richard