How to use u-boot-beaglebone port?

[Paul Mather]

On Mar 17, 2015, at 2:34 PM, Ian Lepore <ian at freebsd.org> wrote:

> On Tue, 2015-03-17 at 14:21 -0400, Paul Mather wrote:
>> On Mar 17, 2015, at 11:59 AM, Ian Lepore <ian at freebsd.org> wrote:
>> 
>>> On Tue, 2015-03-17 at 09:55 -0400, Paul Mather wrote:
>>>> On Mar 16, 2015, at 3:39 PM, Ian Lepore <ian at FreeBSD.org> wrote:
>>>> 
>>>>> On Sun, 2015-03-15 at 19:57 -0700, Tim Kientzle wrote:
>>>>>>> On Mar 12, 2015, at 5:59 PM, Paul Mather <paul at gromit.dlib.vt.edu> wrote:
>>>>>>> 
>>>>>>> Has anyone successfully used the sysutils/u-boot-beaglebone port?
>>>>>>> 
>>>>>>> I managed to build [1] and install it today.  I tried to install it to the SD card FAT partition, as per the README, and the result was an unbootable system.
>>>>>>> 
>>>>>>> When I copied the u-boot.img file as u-boot.img (rather than the bb-uboot.img as suggested in the README), I got it to start up to the "U-Boot#" prompt.
>>>>>> 
>>>>>> 
>>>>>> Apparently, no one ever patched the port to use bb-uboot.img and bb-ubldr
>>>>>> as the name.
>>>>>> 
>>>>>> I did this in Crochet when I was experimenting with having multiple
>>>>>> U-Boots on a single SD card image.  That experiment was to try
>>>>>> to see what would be required to build single images that booted on
>>>>>> multiple different devices.
>>>>>> 
>>>>> 
>>>>> When I created the u-boot-beaglebone port I specifically removed that
>>>>> bb- prefix stuff, because there will never be a unified image that runs
>>>>> on both rpi and beaglebone [*].  I had hoped someone would update
>>>>> crochet to use the new ports and this is one of the minor changes that
>>>>> would be needed on the crochet side.
>>>>> 
>>>>> -- Ian
>>>>> 
>>>>> [*] Because armv6 != armv7 in this case.  While armv6 is synonymous with
>>>>> armv7 for most purposes in freebsd, the rpi is the exception to that in
>>>>> that it really IS armv6, and that leads to the kernel being built with
>>>>> different cache maintenance routines that don't work on armv7.
>>>> 
>>>> 
>>>> Does the sysutils/u-boot-beaglebone boot the BeagleBone Black for you?  As I reported earlier in the start to this thread, I can't get it to boot the system for me.
>>>> 
>>>> I've copied MLO, u-boot.img, and /boot/ubldr to the FAT partition, but I just get to where U-Boot loads ubldr and then pauses before starting over again in a loop.
>>>> 
>>>> Are there some other files that need to be copied to the FAT partition, or are those three files, plus the defaults compiled into u-boot.img sufficient to boot the BeagleBone Black from SD card?
>>>> 
>>>> Cheers,
>>>> 
>>>> Paul.
>>> 
>>> Yep, it works for me on BBW and BBB.  The only time I've seen a totally
>>> silent lockup like that is when the loadaddr variable in the uboot env
>>> didn't match the UBLDR_LOADADDR value when ubldr was compiled.  For BB,
>>> those values are usually 0x88000000, iirc.  If you do a "readelf -a
>>> ubldr" on your build system you should see a line like
>>> 
>>> Entry point address:               0x88000074
>>> 
>>> and whatever it is should be your uboot loadaddr + 0x74.
>> 
>> My current /boot/ubldr entry point address appears to be 0x1000074, which seems to be derived from the default set in /usr/src/sys/boot/arm/uboot/Makefile.
>> 
>> What would I need to put in uEnv.txt to get such a ubldr to boot via the sysutils/u-boot-beaglebone port files?  Would I just need to have the single line "loadaddr=1000000" in uEnv.txt, or would I have to reproduce the whole environment embedded into u-boot.img?  (Do the settings in uEnv.txt replace entirely those in u-boot.img?)
>> 
>>> You should only need MLO, u-boot.img, and ubldr on the fat partition.
>>> (There is an optional uEnv.txt that can be there, but it's not required
>>> to boot.)
>> 
>> It looks like your supposition above is correct and a mis-matching loadaddr variable is likely to blame.
>> 
>> Is a loadaddr of 0x1000000 correct for a BBB?
>> 
>> Cheers,
>> 
>> Paul.
> 
> The address is a physical ram address, so you can't just make up any
> number -- there has to be actual ram at that address on the board, and
> the address must not conflict with where u-boot itself is loaded and
> where the kernel will be loaded.
> 
> I think the BB ram starts at 0x80000000, so 0x10000000 won't work.  The
> right fix would be to recompile ubldr with UBLDR_LOADADDR set to
> 0x88000000.

What is the best place to set this?

My current /boot/ubldr is the product of a native build.  I looked through the source code and it seems the only place that sets this is /usr/src/sys/boot/arm/uboot/Makefile, where we have this:

	# Address at which ubldr will be loaded.
	# This varies for different boards and SOCs.
	UBLDR_LOADADDR?=        0x1000000


The "?=" makes me think this is just a fallback default to stop the build from breaking and that UBLDR_LOADADDR needs to be set accordingly for each different ARM system.  If that is so, and we know what the UBLDR_LOADADDR should be for the BBB (or at least that the default won't work because there's no RAM there on the BBB), then why doesn't -CURRENT set a value such that a working ubldr is built?  (I'd prefer ubldr not to be built at all than a non-working version be built.)

I presume I could set this in /etc/make.conf on my BBB.  Could I put it in my BBB kernel config file (which seems a good place for it)?

Many thanks for the help and information.  I plan to rebuild /boot/ubldr with a UBLDR_LOADADDR of 0x88000000.

Cheers,

Paul.