On 12/01/2012 11:02:05 AM, Deltour, Stephane wrote:
I had a few boards with NAND related problems. In Linux a file was
written to a JFFS2 partition in NAND, but u-boot was unable to read
the
same file correctly from the JFFS2. This happened to be often the case
if the NAND had a few bad blocks in the JFFS2 partition.
Upon further examination it turned out Linux still was able to read
the
file 100% correctly from the partition, but u-boot wasn't.
I did get a lot of "read_nand_cache: error reading nand off ..."
messages in u-boot.
When debugging further it appears that
read_nand_cached(...) in /fs/jffs2/jffs2_1pass.c
throws away the data if nand_read (/drivers/ mtd/nand/nand_base.c)
doesn't return 0.
This happens in case of an uncorrectable ECC error (-EBADMSG) or a
correctable ECC error(-EUCLEAN).
(see nand_do_read_ops in /drivers/ mtd/nand/nand_base.c)
So this would mean that data gets thrown away in case of a correctable
ECC error, which doesn't seem the right thing to do and probably does
not match with how linux is behaving.
I patched the code to not throw away data with correctable ECC errors
and this seems to fix it for the two boards that had this issue.
If it matters: the correctable ECC errors appeared in good blocks (but
often close to a bad block).
So would my understanding be 'correct'?
Yes.
-Scott
_______________________________________________
U-Boot mailing list
U-Boot@lists.denx.de
http://lists.denx.de/mailman/listinfo/u-boot
