On Thu, Nov 18, 2010 at 4:20 PM, Scott Wood scottwood@freescale.com wrote:
On Thu, 18 Nov 2010 16:13:52 -0800 Steve Sakoman steve@sakoman.com wrote:
The code generating both errors is in the nand_do_read_ops function in nand_base.c:
if (mtd->ecc_stats.failed - stats.failed) return -EBADMSG;
return mtd->ecc_stats.corrected - stats.corrected ? -EUCLEAN : 0; }
I understand that the -EUCLEAN error indicates a correctable ECC error. What does the -EBADMSG error indicate?
An uncorrectable ECC error (or other failure).
This condition doesn't seem to bother the linux driver, but u-boot doesn't like it at all!
Check whether the ECC layout and code is the same for this driver in both U-Boot and Linux.
Well, the mystery is solved.
The strange behavior was a combination of the -EUCLEAN issue in u-boot and the following bizarre bug that crept into the Linux OMAP NAND driver in 2.6.26:
http://article.gmane.org/gmane.linux.ports.arm.omap/46545
I will submit a patch to deal with the u-boot issue tomorrow, and it seems that a fix is already queued for Linux 2.6.37.
Thanks to Scott Wood for helping with the -EUCLEAN issue and Scott Ellis for noticing that what might be the same issue was being discussed on both the u-boot and linux lists today.
Steve