s3c244x hardware ecc

Andy Green andy at openmoko.com
Sun Jul 27 09:18:55 CEST 2008

Hash: SHA1

Somebody in the thread at some point said:
| Hey,
| I had a look at using hardware ECC for the s3c244x. I'm in a state that I
| patched uboot to write a ECC for dfu flashed stuff that the kernel
likes to
| read it and is happy.

Wah, no wonder you need a drink.  Nice work.

| What is left:
| 	- Flag day? Provide a upgrade path?
| 	- hardware_ecc=1 into the cmdline and then enable it in the kernel

We can recover from this with NOR U-Boot if it has some issue, but that
is a bit scary for customers.  Likewise U-Boot env changes are scary for

We need to confirm "operation is not degraded" and then just transition
to it as the default.  Particularly in U-Boot, we just need the one way.
~ I think it means sticking these on andy branch with hard ECC enabled
for a little while to get comfortable with them, but right now I am
blocked with other problems on andy branch and will take these in a day
or two.

BTW on the A5 I use it continues to pull 8MBytes of NAND when booting
from there, I guess it is in its environment, hopefully this got reduced
~ to the 2MBytes actually necessary on shipping devices.

| 	- Check if the ECC is useful at all (have to look at the result of the
| hardware register)
| 	- Is the ECC config useful? We have main memory ECC0 and ECC1 but
only write
| in so small chunks that we fill up ECC0? How does this relate to the
2K page
| 	- I have to learn more about OOB, ECC and NAND. And figure out why we
do the
| ECC over 256 bytes for LP...
| 	- Find out why it is not twice as fast?
| 	- Wonder why we only have 8-bit flash and if it is matters?

That's what Samsung gave us, obviously it impacts performance but
equally obviously can't do anything about it.

- -Andy
Version: GnuPG v1.4.9 (GNU/Linux)
Comment: Using GnuPG with Fedora - http://enigmail.mozdev.org


More information about the openmoko-kernel mailing list