On Wed, Feb 29, 2012 at 03:00:49PM -0600, Scott Wood wrote:
On 02/29/2012 06:12 AM, Orjan Friberg wrote:
For the beagleboard, ecc.size is not explicitly set when doing 'nandecc sw'. If it's not set for the NAND_ECC_SOFT case in nand_scan_tail, it's set to 256 bytes.
When doing 'nandecc hw', ecc.size is set to 512 bytes. Hence, when changing back to 'nandecc sw' ecc.size remains at 512 bytes and suddenly the format has changed.
It seems the current nandecc command needs to set this explicitly, but also needs to be augmented to be able to select the newly added 4/8-bit BCH ECC.
Yes, that looks like a bug in omap_nand_switch_ecc().
Correct. And wrt using the new 4/8/16-bit BCH library, that's what I was starting on when I found the nand dump issue :)
But it also seems like nandecc selection should be more generic than for omap3 (currently it lives in arch/arm/cpu/armv7/omap3/board.c).
ECC mode is normally not something that you want to be runtime switchable, as changing it usually changes the on-flash format. It also requires driver cooperation -- the actual implementation (as opposed to the command-line wrapper) is in drivers/mtd/nand/omap_gpmc.c.
There is unfortunately a long-running saga here and what we have now is what we have for a number of existing platforms. Moving forward, things are better and there's an argument at least for not making ECC mode switching a run-time command.