[U-Boot] Proposal to add NAND-boot support for Sunxi SPL

[Roy Spliet]

Hello Hans,

Re-sent to everybody instead of just you. Reply inline.

Op 25-05-15 om 22:39 schreef Hans de Goede:
> Hi,
>
> On 25-05-15 20:35, Hans de Goede wrote:
>> Hi,
>>
>> On 22-05-15 09:04, Roy Spliet wrote:
>>> Hello,
>>>
>>> For my set-up I made use of Boris Brezillon's sunxi-nand tree[1], or 
>>> actually I rebased his patches on top of 4.0rc7. This basically adds 
>>> support for NAND-chip partitioning, ECC and randomisation. Docs for 
>>> the DT specification in 
>>> Documentation/devicetree/bindings/mtd/sunxi-nand.txt , and an 
>>> example can be found in arch/arm/boot/dts/sun7i-a20-cubietruck.dts . 
>>> [2] lists the acceptable configuration options for the boot and 
>>> boot_rescue partitions, make sure to pick one of these (which should 
>>> be no problem for MLC-type nand). The ECC mode for these boot 
>>> partitions is called hw_syndrome.
>>>
>>> Assuming you now have a Linux set-up kernel based on this tree with 
>>> NAND support on an MMC, for U-boot what you should currently do is:
>>> 1) in include/configs/sunxi-common.h, adjust the parameters 
>>> <CONFIG_NAND_SUNXI_>PAGE_SIZE, ECC_STEP, ECC_STRENGTH to match your 
>>> NAND chip and DT configuration.
>>> 2) Build
>>> 3) Use your MMC to flash u-boot-sunxi-with-spl.bin onto NAND:
>>> # flash_erase /dev/mtd0
>>> # nandwrite -p /dev/mtd0 u-boot-sunxi-with-spl.bin
>>> 4) Reboot without the MMC card and see U-boot load
>>
>> Ok, it took me way longer then I wanted (see below) but I've this
>> working now. It is cool to see u-boot load from nand :)
>>>
>>> That should be all.
>>>
>>> @Alex: To answer your question specifically: It's likely that the 
>>> parameters in sunxi-common.h mentioned above might not match your 
>>> NAND-chip configuration in the Linux kernel. I can't tell you 
>>> precisely how to fetch these details from the 3.4 kernel, sorry. I 
>>> recall Daniel using 24-bit strength ECC with otherwise equal 
>>> parameters, but perhaps he can help you with this better than I can.
>>
>> Alex, could it be that you are writing the nand using
>> a (rebased) version of bbrezillon's sunxi-nand-next branch ?
>>
>> I started with that too because it is much newer and contains
>> various bug fixes, but it seems that it also contains a new
>> bug causing it to write the NAND in such a way that the BROM
>> and u-boot SPL code will not read it.
>>
>> I've just pushed a rebased version of the sunxi-nand branch of
>> Boris here:
>>
>> https://github.com/jwrdegoede/linux-sunxi/commits/sunxi-nand-experiment
>>
>> And that works for me, where as before I got the exact same errors
>> trying to fel load a nand enabled spl.
>>
>> I'm working on merging over all the changes from the sunxi-nand-next
>> branch onto my working sunxi-nand-experiment branch 1 by 1 until
>> I find the one which causes the breakage...
>
> Ok, so quick update the breakage was caused by this commit:
>
> https://github.com/bbrezillon/linux-sunxi/commit/7f7324bc6170a45742352070fb45170779a3611c 
>
>
> When it was rebased someone (Boris I guess) forgot to remove the
> "chip->read_buf(mtd, NULL, ecc->size);" line at line 1075 (after the 
> patch) and
> likewise the "chip->write_buf(mtd, buf + (i * ecc->size), ecc->size);" 
> line at
> line 1161. With these 2 lines removed the sunxi-nand-next branch from 
> Boris,
> rebased on 4.1-rc1 can write the nand boot parts and the brom / spl 
> can load
> the spl / resp. u-boot.bin from there (on a cubieboard2).
You're right... I did spot this, but assumed this was my own mistake in 
merging these patches with our 4.0RC7 tree. Sorry, could have saved you 
some trouble if I were sharper.
>
> I've also tried to get this code running on a cubieboard (non 2 so A10 
> rather then
> A20), the SPL loads fine there (indicating that the kernel bits work), 
> but then
> I get:
>
> U-Boot SPL 2015.07-rc1-00287-g050de86-dirty (May 25 2015 - 22:28:19)
> DRAM: 1024 MiB
> CPU: 1008000000Hz, AXI/AHB/APB: 3/2/2
> Nand initialised
> NAND timeout reading data
> NAND timeout reading data
> NAND timeout reading data
> NAND timeout reading data
> NAND timeout reading data
> NAND timeout reading data
> ...
Thanks for testing this. I don't own an A10 myself, so I haven't 
observed this behaviour.
First thing I would verify personally is whether all the clocks are 
properly configured. Could you make U-boot print &ccm->ahb_gate0 and 
&ccm->nand0_clk_cfg? For the first, bits 6 (DMA) and 13 (NAND) must be 
set, the second must have bit 31 set, bits 24 and 25 cleared (and 
otherwise, the accompanying PLL must be configured too... probably 
easier to use the OSC24M), and the divide ratios set to 0 (although a 
small divider, like 1, shouldn't be a problem either).
I am assuming these values should be correct, but only because BROM 
initialised part of it. I am aware of the fact that the SPL driver 
doesn't control the DMA gating reg and the NAND post-dividers. It might 
be a good idea to do so for as long as we don't have a full DMA driver, 
so I'll patch that up today.

Roy
>
> Which seems to indicate a problem with the SPL nand code on the A10. 
> I'll investigate
> this further tomorrow evening.
>
> A cleaned up version of my kernel work on this is available in my
> sunxi-wip branch.
>
> Regards,
>
> Hans


-- 


IMAGINE IT >> MAKE IT

Meet us online at Twitter <http://twitter.com/ultimaker>, Facebook 
<http://facebook.com/ultimaker>, Google+ <http://google.com/+Ultimaker>

www.ultimaker.com