Hi Karl,

> In the read and write functions, isn't the "if (dev->useNANDECC)"
> unnecessary since the yaffs2 code only supports the MTD ECC?

I thought YAFFS2 was also capable of doing its own ECC. Anyway, I was
aiming to preserve functionality and only make things work with the
curernt MTD stack.

> And in the write function, you write the data and tags using separate mtd
> calls.  This means (at least for normal NAND) that the spare area is written
> twice, first with data ecc, and then with tags.  Wouldn't something like
> this be better, since write_oob() can write both in one go?  Note that tags
> is never NULL.

There definitely is rationale in your words. Though, as far as I
thought, tags might be NULL, but anyway tags were never written
without writing data.
I was trying to do writing both data and tags in one call but I wasn't
successful. Later I discovered that the problem was outside YAFFS and
submitted the patch fixing it linux-mtd, so now everything is ready to
do that in a better way (ie as you propose).

>
>   /* Write data and oob area, or only oob area */
>   yaffs_PackTags2 (&pt, tags);
>   ops.mode = MTD_OOB_AUTO;
>   ops.oobbuf = (void *)&pt;
>   ops.ooblen = sizeof(pt);
>   ops.len = data ? dev->nBytesPerChunk : sizeof(pt);
>   ops.ooboffs = 0;
>   ops.datbuf = (uint8_t *)data;
>   retval = mtd->write_oob (mtd, addr, &ops);

Well, a bit differently, but still... please find the updated patch attached.
Thanks for feedback!

Best regards,
   Vitaly