Musings on the PLX kernel sources.

[telzey]

I'm not sure where you're coming from there

The 3MB region is comprised of 24 erase blocks. Even if you allow for 10% of bad blocks (i.e. 3), then you've got 21 blocks * 128KB = 2752512 bytes of space. For comparison, UBIFS is only allowing for 1% bad blocks.

The 2KB page size is irrelevant in this case since we're talking about a linear memory image and not a filesystem like UBIFS.

Am I missing something?

And speaking of NAND ... did you ever try the new NAND stage1 that I sent you? I've been using it for a while now without problem, and it would be nice for people to be able to completely fix their NAND problems after mistakenly flashing Jeff's Pogoplug-v2 u-boot on a Pogoplug-v3.

[WarheadsSE]

1) We need to erase before writing, and have to work around their image locations and ours, and I've allowed some buffer. I can re-do my calculations later.
2) Actually, in all this hubub @ work and with my back.. I've actually let that slip my mind, sorry.

[telzey]

Your kernels and the original kernels are all aligned to erase-block boundaries, so that's not an issue.

You may be allowing for 6 bad blocks in your calculations. You can certainly make a case for that many, but it does seem a little excessive.

OTOH you've got to deal with the potential of irate users which, as you've seen from my posts, is a responsibility that I'm strenuously trying to avoid myself

No problem with the stage1 ... it's not exactly a high priority for anyone since we don't have a new u-boot to test yet.

[WarheadsSE]

Actually, for the most part my u-boot on GitHub will work. It will boot from NAND/SATA/network. We need to get ext2 & fatload supported. Also, somehow weasel USB drivers into it.

[telzey]

Ah, I didn't think that you were considering it for a NAND installation yet, but rather just for SATA booting (which will ignore the NAND anyway).

Once the extra stuff is in there, it'll be a must-have!

[WarheadsSE]

Yeah, it will work for either one, I just haven't messed with locations in NAND :p

[ftcodes]

Regarding DMA,

There are a lot of specific glue code things that need to be right on multi-core systems. In fact, you can take any architecture.

The ARM11MPCore has a specific unit called SCU which keeps the cache coherency between CPUs right.
But DMA is not addressed by the SCU unit and therefore needs specific code to deal with those memory blocks.

I am assuming that the generic ARM 11 MPCore is generally sufficiently tested but I am also assuming that some PLX specific code managed to sneak through unnoticed.

That is why I decided on redoing the first step that got the code into the 3.1 kernel by taking the 3.3.2 vanilla and merging the changes step by step to get passed any sneaked in things that should not be there.
My idea is to keep as much of the generic ARM11MPCore support code as possible from vanilla and only having the startup specific glue code for it.

[Philoo]

@ telzey :

as mentionned somewhere in this thread, 2.6.31.14 is not totally stable wrt SATA. I have set up a web proxy (actually an arch'ed apt-cacher-ng) for my updates and when many or bigger packages are already cached when I pacman -Syu the pogoplug pro I get sata write errors.
This does not happen if using a direct connection (MUCH slower than my local Gb/s) or a common /var/cache/pacman (much less disk IO).

in anycase your patched 2.6.31.14 is really a GOOD job.

[telzey]

I haven't had any problems myself during testing, but that certainly doesn't mean that they don't exist!

I was a little worried about using the CE patches since CE never supported the SATA connection, but I'd been told that they were stable with 2.6.31.6, so it was worth a try.

I do have a 2.6.31.14 patchset based upon the Silverstone 2.6.31.14 kernel patches if you'd like to try it, but I've personally lost patience with PLX.

With the recent spate of E02s shipping in P21 boxes, and the D-Link DNS-325 2-bay raid box going on sale for $109, I've decided to abandon the OX820 and jump to the kirkwood platform instead

[Philoo]

I hear you, alas I'm too stuborn for my own good.
I'd like to give a shot at the silverstone patches. Where can I get them (not sure I'll have the patience to get the source from silverstone and make the diff myself).