Andy Green wrote: > So U-boot used soft ECC and its data was incompatible with the hard ECC > used in the kernel?
Yup. > Presumably if we farm out ECC to the hardware it is > going to be the same in U-Boot and kernel. Yes. > Trying to understand this so > it doesn't trip us up in kboot world. Yet another problem that simply can't happen since the same code is used for all these operations :-) However, I think we should do ECC error correction also in the kboot loader, because NAND performance is only guaranteed for correctable errors (i.e., single-bits), but not for the raw memory. (I guess this makes you like NAND yet a little less ? ;-) By the way, for "kboot" userspace, it's interesting to follow what the PS3 Linux project is doing since they've been using kboot for a while. It turns out that they've abandoned kboot and now have something called petitboot: http://ozlabs.org/~jk/projects/petitboot/ Petitboot makes the build process easier (kboot basically builds its own little Linux distribution, with all the problems associated with doing that) and it provides a neat little GUI. Given that the shell features of kboot wouldn't be terribly useful in everyday use on our devices anyway (Petitboot has a shell escape, so no functinality is lost), I think we should try to make Petitboot work as our main "kboot" interface. (That is, in non-fastpath mode.) - Werner
