[PATCH 1/6] arm: mach-k3: Add default ATF location for AM62/AM62a

[Bryan Brattlof]

On June 20, 2024 thus sayeth Wadim Egorov:
> 
> 
> Am 19.06.24 um 22:02 schrieb Andrew Davis:
> > On 6/19/24 1:20 PM, Nishanth Menon wrote:
> > > On 17:19-20240619, Dhruva Gole wrote:
> > > > Hi,
> > > > 
> > > > On Feb 14, 2024 at 10:30:04 -0600, Andrew Davis wrote:
> > > > > There is a default ATF load address that is used for devices that have
> > > > > ATF running in SRAM. For AM62 and AM62a, ATF runs from DRAM. Instead
> > > > > of having to override the address in every defconfig, make add a
> > > > > default for these ATF in DRAM devices.
> > > > > 
> > > > > Signed-off-by: Andrew Davis <afd at ti.com>
> > > > > ---
> > > > >   arch/arm/mach-k3/Kconfig               | 5 +++--
> > > > >   configs/am62ax_evm_a53_defconfig       | 1 -
> > > > >   configs/am62x_beagleplay_a53_defconfig | 1 -
> > > > >   configs/am62x_evm_a53_defconfig        | 1 -
> > > > >   configs/phycore_am62x_a53_defconfig    | 1 -
> > > > >   configs/verdin-am62_a53_defconfig      | 1 -
> > > > >   6 files changed, 3 insertions(+), 7 deletions(-)
> > > > > 
> > > > 
> > > > Beagleplay stops booting completely with latest U-boot so I did a little
> > > > bisect and it seems like reverting this patch helps.
> > > > 
> > > > I am not sure what other implications there are of reverting this so I
> > > > don't suggest immediately dropping it, however some hints around what's
> > > > missing on beagleplay vs other platforms would be helpful.
> > > > 
> > > > Because other platforms based off the same SoC don't seem to be
> > > > affected.
> > > 
> > > Grumble... I had to rediscover this in parallel as well - Thanks Dhruva.
> > > mkimage -l tispl.bin and comparing with kernel log of reserved mem
> > > had me completely confused.
> > > 
> > > a) Memory maps are already notorious to manage on complex heterogenous
> > >    systems. From beagle perspective, we have no need to go and monkey
> > >    with DT defined memory map and DT should be the default and modifying
> > 
> > Your DT source file is wrong, ATF is at 0x8000_0000 (or any address one
> > wants to put in K3_ATF_LOAD_ADDR). DT doesn't "define" hardware, it
> > "describes"
> > it. ATF's location is dynamic and configurable, it doesn't belong in DT.
> > 
> > You have two options, either go update your DT, then update it again every
> > time ATF moves. *OR* simply turn on OF_SYSTEM_SETUP and let U-Boot add the
> > correct location reserved memory node for you.
> > 
> > Then you can also drop out the reserved-node from the DT template file.
> > Only U-Boot knows where ATF is really placed in RAM, so U-Boot must add
> > this info to DT.
> 
> I think it is a good idea to remove the nodes from the device tree or at
> least mark them as "templates" and note that they are actually provided by
> the bootloader to make it less confusing.
> 

I have a patch lined up for this. My hope was we could remove all memory 
nodes completely from the kernel and have U-Boot fix these up. It's 
initialize DDR knows about capacity, inline ECC, and trowing all the 
different binaries into it. 

Our memory{} node fixup isn't ready yet unfortunately so I think just the 
reserved-memory{} node can be ripped out for now.

~Bryan

>
> > 
> > >    DT should be explicitly called out with a log (instead of done
> > >    "transparently") - so instead of CONFIG_K3_OPTEE_LOAD_ADDR default
> > >    being forced from u-boot, it should be made optional, where when
> > >    defined, it can overide the dt definition or some variant of that.
> > > 
> > > b) Looks like
> > >     https://lore.kernel.org/u-boot/20240214163009.983034-4-afd@ti.com/
> > >     missed beagleplay? and we were in for a surprise there - i dont want
> > 
> > This was the only miss, OF_SYSTEM_SETUP didn't get added to the beagleplay
> > config. Bryan just sent the fix for that now. I'll go look for a way to
> > make this common across the whole SoC family so we don't again miss any
> > new boards.
> > 
> > Andrew
> > 
> > >     to switch from default dts for beagleplay to something different just
> > >     because of simplicity for users to know exactly the carveouts and
> > >     with other s/w starting up on uC, dts is our "canonical truth".
> > >