On 13/01/2014 12:19, Dmitriy B. wrote:
---------- Forwarded message ----------
From: *Dmitriy B.* <[email protected] <mailto:[email protected]>>
Date: 2014/1/9
Subject: Re: [linux-sunxi] [RFC PATCH 0/9] mtd: nand: add sunxi NAND
Flash Controller support
To: [email protected] <mailto:[email protected]>
2014/1/8 Boris BREZILLON <[email protected]
<mailto:[email protected]>>
Hello,
This series add the sunxi NFC support with up to 8 NAND chip
connected.
I'm still in the early stages drivers development and some key
features are
missing, but it's usable (I tested it on the cubietruck board).
Hello Boris,
Thanks for working on NAND controller support, I was thinking about
doing the non-DMA version of Yuq's driver for mainline testing, but
didn't have time to do it.
First, I want to ask about performance, did you have time to make some
measurements for NAND-flavored file systems like UBIFS/etc. ?
I haven't done any perf tests yet (and don't plan to do it in a near
future).
But any help is welcome.
Info about this will be really interesting to know, how it will
perform using the soft_bch and no DMA available.
Second, did you encounter any problems with badblock scan and oob
areas usage?
No, but I haven't lot of tests. Again: any help is welcome. :)
When I was poking around the Yuq's driver I found couple bugs there
that I could not avoid, for example, badblock bug was that on the end
of any NAND chip 4 or 5 blocks were always non-readable after write
and standard badblock detection didn't detect that right, I was forced
to manually shrink my MTD area so it wont use those blocks at the end.
This might come from the way you store ECC bytes.
You can store then in:
1)OOB area
or
2)after each ECC block size
if you choose the 2nd solution, you will end up writing on the OOB bad
block marker when you write the last page.
And third, what about the SPL area and possibilities of pure NAND
boot? Yuq's driver had the nand1k things to work this around.
I already answered this in my last mail :).
Here's what's missing:
- HW ECC support
- DMA support
- HW randomization support
- many more improvements
Any plans on implementing these? Especially, HW ECC/HW random? If I
remember correctly, random was broken in Yuq's driver, even though he
tried to use it exactly like in the allwinner's sources.
Ditto
And last question, do you plan to implement community 3.4 kernel support?
No.
If your version has no bugs with badblocks/oob it could be used as an
alternative to what we have on 3.4 (the ugly block layer).
It's up to. You can backport it ;).
Thanks again for working on this subject, hope we will soon see a full
NAND MTD boot on our sunxi devices.
Thanks to you (for all your work on reverse engineering the NFC IP) :).
Best Regards,
Boris
Best Regards,
Dmitriy Beykun
--
You received this message because you are subscribed to the Google Groups
"linux-sunxi" group.
To unsubscribe from this group and stop receiving emails from it, send an email
to [email protected].
For more options, visit https://groups.google.com/groups/opt_out.
