[PATCHv3 09/16] ARM: mvebu: implement suspend/resume support for Armada XP

[Andrew Lunn]

On Sat, Nov 22, 2014 at 09:56:48PM +0100, Thomas Petazzoni wrote:
> Dear Andrew Lunn,
> 
> On Fri, 21 Nov 2014 18:20:31 +0100, Andrew Lunn wrote:
> > > > Hi Thomas
> > > > 
> > > > Is this a well defined mechanism supported by mainline uboot, barebox
> > > > etc. Or is it some Marvell extension to their uboot?
> > > 
> > > As far as I know, it is a Marvell extension to their "binary header",
> > > so it's done even before U-Boot starts.
> > 
> > Ah, O.K. So i miss understood what you meant by bootloader. It is the
> > first stage boot loader as part of the BootROM that does this.
> > 
> > If i've got this wrong, probably others will as well. So could you
> > make the comments talk about BootROM bootloader?
> 
> No, it is not the BootROM that does this. It is the binary header,
> which is something different.

O.K, so this definitely needs documenting somewhere, since it is not
obvious. I'm not sure if it is sufficient to make this part of the
commit log, or if it should be documented in Documentation/power.
 
> Essentially, the boot process is the following:
> 
>  * BootROM runs. This one cannot be changed, it is set in stone in the
>    chip.
> 
>  * It locates a bootable image from some media, and looks at its
>    header. Its header can contain several sections, some describing
>    some (address, value) pairs to write some values to some registers
>    (this was used on Kirkwood to configure the memory controller, for
>    example). Some other sections can contain binary code that will get
>    executed on the SoC, this is what is called a "binary header". It
>    gets executed before U-Boot starts, but it's really a piece of
>    software that can be changed.
> 
>  * Then it loads and runs U-Boot.
>
> 
> The "binary header" is currently the piece of code that we extract
> using the kwbimage tool from existing U-Boot images to build working
> Barebox images, since we haven't yet written the equivalent code in
> Barebox land.

Is it a binary blob in the uboot source tree?
 
> So, the specification of where the "resume informations" is located and
> how this information is organized is purely defined by software that
> can potentially be changed, not by the BootROM. Though that if we
> decided to use a different protocol, we would basically have a
> suspend/resume in the kernel that would not work with any Marvell
> platform  that uses the default bootloader.

It also makes mainline suspend/resume dependent on Marvell not
changing the bootloader.

	 Andrew