On Tuesday 09 August 2005 00:26, Ludovic Guilhamat wrote:
> Hi,
>
> I found some  other clues about my problem, but I didn't fully manage it
> yet...
>
> Brief summary : I want to use yaffs on linux-2.4.22 kernel with uClinux
> (m68k coldfire), and with the MTD of the 2.4.22 kernel. (I attempted to
> backport a newer MTD, but without success for the moment). My Flash is a
> Samsung 16Mb Nand.

Part numbers please. Different parts have different properties.


>
> So, my problem occurs when I erase a file on my 2Mb partition. More
> precisely, it occurs when I want to copy/create some other files. Hence,
> I think it's the garbage collect process that is the source of my
> problem on my system. 

In almost all cases, problems that arise during gc are cause by one of:
1) flash geometry issue. The block size you have specified in yaffs or mtd 
does not match the physical hardware.
2) Some issue with ECC or low-level NAND access functions.



> Here is the log: 
>
> * after erasing the file :
>
> yaffs locking
> ___________ really free is 3033, pending 1001, nFree is 4034
> yaffs unlocking
>
> * When I want to recreate a file after an unmount / mount :
>
> nand_read_ecc: from = 0x00909e00, len = 512
> nand_read_oob: from = 0x00909e00, len = 16
> nand_write_oob: to = 0x00908000, len = 16
> nand_write_oob: Failed write verify, page 0x00004840     <----- This
> corresponds to the (falsely) damaged marked blocks that I see when I
> nand_write_oob: to = 0x00908200, len = 16
> nand_write_oob: Failed write verify, page 0x00004841     <----- want to
> format my partition.

You need to turn off write verification in the mtd and do that in YAFFS 
instead.

>
> * Finally, here is the interesting part of 'cat /proc/yaffs' :
>
> nPageWrites........ 10
> nPageReads......... 181
> nBlockErasures..... 0
> nGCCopies.......... 0
> garbageCollections. 3     <----- It seems that at each garbage
> collection, 8Kb are used on my partition (is it normal ?). Then,
> passiveGCs......... 3          <----- I find 104Kb used, instead of 80Kb...
> nRetriedWrites..... 0
> nRetireBlocks...... 3

Retired blocks == blocks that YAFFS has decided went bad and then marked bad. 
This is most likely caused by a mismatch between YAFFS and mtd, most likely 
in either the ECC or similar area.

-- Charles