On 1/4/24 1:52 AM, Manorit Chawdhry wrote:
Hi Andrew, Simon,
On 08:58-20240102, Andrew Davis wrote:
On 12/31/23 6:48 AM, Simon Glass wrote:
Hi,
On Wed, Nov 29, 2023 at 7:48 PM Simon Glass <s...@chromium.org> wrote:
Hi Andrew,
On Mon, 6 Nov 2023 at 11:05, Andrew Davis <a...@ti.com> wrote:
On 11/6/23 11:47 AM, Simon Glass wrote:
Hi Andrew,
On Mon, 6 Nov 2023 at 10:27, Andrew Davis <a...@ti.com> wrote:
On 11/6/23 9:31 AM, Tom Rini wrote:
On Mon, Nov 06, 2023 at 11:23:51AM +0530, Manorit Chawdhry wrote:
Hi Simon,
On 11:22-20231005, Simon Glass wrote:
Hi Nishanth,
On Thu, 5 Oct 2023 at 11:16, Nishanth Menon <n...@ti.com> wrote:
On 12:10-20231005, Nishanth Menon wrote:
On 12:36-20231005, Tom Rini wrote:
On Thu, Oct 05, 2023 at 09:19:48AM -0500, Andrew Davis wrote:
On 10/4/23 8:54 AM, Nishanth Menon wrote:
On 08:48-20231004, Andrew Davis wrote:
On 10/4/23 8:23 AM, Roger Quadros wrote:
ti_mmc is not a valid boot_target for standard boot flow so
Is there some way to make it into a valid boot_target? Otherwise
how do we use uEnv.txt files, or boot from FIT images with overlays?
envboot takes care of uEnv.txt file (see
https://lore.kernel.org/all/20231004132324.44198-3-rog...@kernel.org/)
Early remote proc loading and FIT image is a question for stdboot itself.
If stdboot is missing these features then we shouldn't switch until it
has them. I'm all for switching to this, but only if it is complete.
Depends on what you mean? Did you mean an option to run scripts
(exists) or an option to do what TI needs done, via
boot/bootmeth_something.c ? If the latter, someone from TI needs to
figure out what that should be and do (but plumbing-wise everything it
needs should exist).
Andrew is generalizing here (on the wrong patch though).
On am62x platforms, there is nothing regressing with this series. The
challenge is early remote_proc loading which is done for J7* platforms.
How that is initiated as part of bootmethods is something of a gap.
The other gap has been support for uEnv.txt -> which we can workaround
at the moment by using CONFIG_BOOTCOMMAND="run envboot; bootflow scan
-lb" in defconfig (This series from Roger already does that - hence I am
saying that Andrew is complaining on the wrong series).
Ideally, we should just have CONFIG_BOOTCOMMAND="bootflow scan-lb" and
uEnv.txt remoteproc loads and the various standard bootmethods should
"just work".
I forgot to add: FIT image authenticated boot flow. That is really what
ti_mmc distroboot method was trying to solve.
Maybe Simon or someone know how the stdboot flow handles authenticated
kernel image and dtb boot flow with FIT image?
Yes you can use FIT configuration verification and things should work as normal.
Could you give any reference documentation for this?
I suspect you should start with doc/usage/fit/beaglebone_vboot.rst
From that doc:
```
Boot the board using the commands below::
setenv bootargs console=ttyO0,115200n8 quiet root=/dev/mmcblk0p2 ro
rootfstype=ext4 rootwait
ext2load mmc 0:2 82000000 /boot/image.fit
bootm 82000000
```
This is very much not stdboot :( This doc has some good and current info on
building
a secure FIT image, but much of it is showing its age. Stdboot is still missing
* ability to limit boot mode search to only one image (FIT)
What does this mean? Can you please be more specific or give an example?
Sure, for instance with secure boot we only want to search for FIT images
and if for each media this fails we do not want to fall back to other
image types or boot modes (like UART boot or net boot which would bypass
the signature checks).
We could have something like:
bootstd {
image-types = "fit", "uimage";
}
to limit the supported types.
Something similar is needed for searching for EFI compatible boot across
each enabled boot media first, before trying other methods on each media.
Basically breadth-first search on each media type not depth-first.
How about:
bootstd {
scan-order = "bootmeth,bootdev"; // instead of default
"bootdev,bootmeth";
};
This along with the image-types above seem reasonable. Might also help with
our EFI search order issue (we need to search all boot media for EFI parts
first before trying all the other types of boot on each media, basically
breadth-first search vs depth for EFI). We should be able to describe that
pattern here if we think this through.
This seems okay and will solve the problem of the FIT Image issues and
fallback that we were having with stdboot.
* fetching and applying FIT overlay/config strings
Re overlays, is this the 'extension' command? What are config strings?
When we have DT overlays in our FIT image, we build a string to pass to
bootm to apply all the selected overlays[0]. If there is now another
mechanism for this please let me know (building this string today is messy).
https://source.denx.de/u-boot/u-boot/-/blob/master/include/env/ti/ti_common.env?ref_type=heads#L18
My proposal for this is to implement extensions in FIT. This would
have a new extensions nodes, so you can specify what extensions are
available for each FIT configuration.
configurations {
conf-1 {
compatible = ...
extensions = "ext1", "ext-2";
};
extensions {
ext-1 {
fdto = "fdt-1"; // fdt overlay for this 'cape'
compatible = "vendor,combined-device1";
extensions = "ext3";
};
ext-2 {
fdto = "fdto-2"; // fdt overlay for this 'cape'
compatible = "vendor,device2";
};
ext-3 {
fdto = "fdto-3";
compatible = "vendor,device3";
};
So FIT configurations now have a list of supported extensions. The
extensions are hierarchical so that you can have ext1 which can
optionally have ext-2 as well. This allows boards which share a lot of
'capes' to be specified only ones.
Extensions actually present are declared by a sysinfo driver for the
board, with new methods:
get_compat() - determine the compatible strings for the current platform
get_ext() - get a list of compatible strings for extensions which are
actually present
The extension compatible strings are used to select the correct things
from the FIT, apply the overlays and produce the final DT.
This is something that is still concerning me, so previously we used to
have two methods of applying overlays, one was through FIT Image where
we have the overlays embedded and in another we used to use non FIT
boot where the overlays were present in the filesystem only and we used
to apply them, this might solve the problem with FIT Image booting overlays but
normal kernel boot and dtb with overlays is still something that I don't
know would be possible or not with this. Is there any way to do that?
If we have a list of compatible strings and extensions, then do we need
the extensions nodes? Couldn't each extension string simply be the name
of the overlay ftdo-x node to apply?
* remote proc loading
What command can do this?
https://source.denx.de/u-boot/u-boot/-/blob/master/include/env/ti/ti_common.env?ref_type=heads#L28
This could be a new property in bootstd:
bootstd: {
probe-before = <&phandle of remoteproc>, ...
};
I think this can also work, the firmware names we can get from the dt
nodes and I think modifying the names to signed firmwares that we used
to do in the scripts can be handled in the driver itself.
We need to also think about where the firmware blobs themselves come
from in this case. In fact we are having that discussion in a different
thread right now[0].
My suggestion would be to have the firmware inside the FIT. This is
already supported to some degree for normal firmware (TFA/OP-TEE/etc.).
This way drivers do not need to deal with all the complexity of finding
and loading blobs (they would have to replicate all the logic of stdboot
to find the blobs correctly in all cases if the blobs are not in the
same FIT that stdboot finds for us).
Andrew, do you see anything else in this that cannot work?
* uEnv.txt processing
What command can do this?
run envboot;
https://source.denx.de/u-boot/u-boot/-/blob/master/include/env/ti/mmc.env?ref_type=heads#L20
This could be a new bootmeth which looks for uenv.txt on available
devices. It might be better to put the env in a FIT or something with
a checksum.
We use non-FIT boot as well for our non HS platforms, I think the new
bootmeth might be helpful as you mentioned, would have to look at this
more as I thought bootmeth is supposed to boot a platform at the end
only but it seems like here we'll be combining this bootmeth ( that'll
just import the env ) and then other bootmeth will actually boot Linux.
As long as bootmeths can perform an action then continue with other
bootmeths that could work. This might require complex logic though.
For instance the uenv.txt bootmeth would only need run on some subset
of available bootdevs, and it should scan all those devs for uenv.txt
first, before the Linux finding bootmeths start.
Andrew
[0] https://lore.kernel.org/all/64a18d0a-294d-4e11-a867-4191e99f2...@ti.com/
Regards,
Manorit
Last two can be solved by adding to CONFIG_BOOTCOMMAND before calling`bootflow
scan`
as suggested above. Until we figure out the first two I don't see us ready to
switch.
I'm happy to help figure this out.
Thanks, we are eager to standardize our boot flows and stdboot seems like the
right way forward, very happy to have your guidance on these last few items :)
I'm just checking whether there is any feedback on the above? I sent
the FIT part as a separate proposal.
Manorit,
Does the above all work for you?
Andrew
Regards,
Simon
