The solution for NAND/SMC on S3C2410 problems

[Selmeci, Tamas]

> > [can't mount root fs (cramfs) from SMC, complaining about ECC read errors]
> 
> Wait, you are trying to store a kernel and cramfs filesystem on raw
> NAND flash? Chances are, they are each sitting on at least one bad
> sector. You need an FTL or YAFFS.

Yesterday I've found a working solution.

Firstly, I had to disable both FTL and NFTL support. When these were switched on, the kernel worked not better at all. I've spent 10-12 hours with examining MTD (NAND and S3C2410-specific) codes, drawn some figures, made notes and it turned out what was the problem: 

Mmy old (64MB, Samsung) SMC was in an absolutely bad condition, Windows could detect and use it with FAT, but the many bad blocks on them prevented vivi to create an adequate partition table etc... after obtaining a new SMC, everything went better. 

Vivi was recompiled without BonFS support. As it turned out, BonFS is absolutely unneeded for NAND/SMC, I don't known why Embest implemented in 2.4.18-rmk7... furthermore, I still don't know what BonFS is good for...

With vivi, I simply created 4 MTD partitions on the SMC (vivi, param, kernel and root).
In the board-specific code (arch/arm/mach-s3c2410/mach-....), I had to add some NAND-related structures describing the NAND-set present on the board, timing settings and the partition table layout etc...

After adding this, it seemed that MTD code for S3C2410 has some bugs in calculating flash timings/clocks, so I fixed that (s3c2410_nand_get_clk_rate?). After this, the card was successfully identified by the kernel, and the initial block scan (to build up bad block table) runned perfectly. I figured out that MTD registered the partitions as /dev/mtdblock0,1,2,3, so my root was /dev/mtdblock3.

For some reason, with correct root= kernel parameter, the kernel complained about read errors (ECC), when it tried to mount the root partition. With hardware ECC, there were less, with SW ECC, there were more ECC-error messages... It seems, that there are still bugs or misconfigurations in the ECC-calculation algorithms in S3C2410, because after manually disabling ECC-check, everything worked well, the root partition was mounted!!! (I saw some selections for ECC-calculations, maybe I'm gonna try them)

Now I'm making kernel menu extensions, slightly modifying codes, dealing with BBT, and trying to find a solution for that ECC-problem. (I've found a boot log on the internet with 2.6.11 on an S3C2410, and ECC was disabled there too, so it seems to me that there are problems with ECC-support). Currently my root partition is cramfs, but JFFS2 will be also tried.

Maybe I should try using FTL or YAFFS, because it ensures that bad-blocks are separated, if I'm right...?

I'm very happy, because I could make my SMC work with partitions! :)

Bye
--
Tamas Selmeci