On Tuesday 28 December 2010 06:38:17 Ross Younger wrote:
> * Boaz Ben-David <boaz.bd@wellsense-tech.com> wrote:
> > My problem is that when I try to write some files (even using "touch")
> > it seems yaffs is trying to retire all of the blocks on the partition.
>
> Most problems I've encountered of this nature have been caused by bugs in
> the MTD driver(s). The root causes have been things like the MTD driver
> misreporting to YAFFS that the write failed, the write actually failing
> due to a protocol or timing error, or YAFFS believing there to be an
> ECC mismatch somewhere along the line. Particularly check the ECC -
> sometimes you have to special-case the check for empty pages (if the
> ECC for an erased block - all 0xFF - is not itself all 0xFF).
>
> You may be able to get more of an insight into exactly what has gone
> wrong by turning on tracing and working backwards from the exact error
> reported.
>
> Ross
Ross is almost certainly right.
The nand oob/spare area is generally split into three areas:
* bad block marker - typically 2 bytes at offset 0
* ECC bytes.
* oobferee used by yaffs to store tags.
Eg there is a typical oob layout structure:
static struct nand_ecclayout nand_oob_64 = {
.eccbytes = 24,
.eccpos = {
40, 41, 42, 43, 44, 45, 46, 47,
48, 49, 50, 51, 52, 53, 54, 55,
56, 57, 58, 59, 60, 61, 62, 63},
.oobfree = {
{.offset = 2,
.length = 38}}
};
The structure starts at offset 2 because the first two bytes are used by the
bad block marker.
The yaffs tags reading/writing should be accessing oobfree.
Some drivers get this wrong and reads/writes that should be accessing the
oobfree areas are instead accessing some other area. A typical mistake is to
not apply the layout and just access at offset 0. This causes problems
because the tags and bad block marker get mashed together.
I suggest you work through the function calls to check that the right bytes
are being accessed.
--- Charles
(text/plain)