Re: [PATCH RFC] ARM: dove: build multiple pbl images for DT based boards

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



On 02/09/2014 09:36 AM, Sascha Hauer wrote:
On Sat, Feb 08, 2014 at 02:27:36PM +0100, Sebastian Hesselbarth wrote:
[...]
What I want to achieve is to have a common pbl entry function for all
DT based boards of one SoC. For now, let's just talk about Marvell Dove
here, but on the long run, I'd like to have it also for the other MVEBU
SoCs. Right now, I guess, each SoC will still have its own "DT board"
but it could also end up in one board for all MVEBU SoCs..
[...]
Now, for having one board file for a set of selectable dtbs, you would
need a single entry function name and also a single symbol to access
the linked dtb at runtime. In the example for Dove below, I have chosen
start_dove_dt for the entry function and __linked_dtb_start for the
required dtb.

It's not quite clear to me what you want to archieve or maybe better why
you want to archieve it. Right now there are board specific entry
functions, with 'board' meaning everything that you can't detect during
runtime. So if want to support multiple boards and can detect during
runtime on which one you are on then you can use a single entry
function, detect the board and pass the correct dtb to
barebox_arm_entry. If you can't detect the board then you need different
images, one per board. An image is defined by its entry function. Why do
you want to make the entry function common per SoC? I mean if there's
some SoC specific setup that every board of a SoC type needs, then you
can factor it out to a common function and call mvebu_setup_something()
in each entry function or make a mvebu_barebox_entry like we already
have.
I think the entry functions are short and easy enough to duplicate them
per board. Also it gives you the chance to do board specific quirks very
early without having to #ifdef common code.

Well, the idea was clearly driven from me being too lazy to copy the
board code for every supported board we have. For example, there are
three different CuBox revisions around, ES, Production 1GB, and 2GB.
You can detect the 2GB version at runtime from the mem region registers,
but you cannot tell ES and 1GB apart.

Also, currently you cannot really select a different dtb for e.g. the
CuBox because the board code depends on __dtb_dove_cubox_start.
Choosing a different dtb e.g. dove-cubox-1gb.dts will break the board
code because of the missing symbol.

With one single board code and selectable dtbs, you would link only
one dtb at link time and get another layer of differentiation for free
without bloating the code with multiple dtbs. I admit, "real" board
specific quirks would have to be dealt with somehow.

Maybe the idea is non-sense, but I though giving the user a list of
currently supported boards from which he can select from and get
boot images for every board selected would be nicer than spreading
the same board code over and over again in arm/boards/.

Sebastian

_______________________________________________
barebox mailing list
barebox@xxxxxxxxxxxxxxxxxxx
http://lists.infradead.org/mailman/listinfo/barebox




[Index of Archives]     [Linux Embedded]     [Linux USB Devel]     [Linux Audio Users]     [Yosemite News]     [Linux Kernel]     [Linux SCSI]     [XFree86]

  Powered by Linux