On 13/12/2022 15:31, Quentin Schulz wrote:
Hi,
On 12/13/22 15:14, Neil Armstrong wrote:
On 13/12/2022 14:45, Quentin Schulz wrote:
Hi Tom,
On 12/13/22 14:37, Tom Rini wrote:
This reverts commits 51c5c28af59c ("cmd: pxe: use strdup to copy config"), a5dacef7380e ("cmd: pxe: support INITRD and FDT selection with FIT") and f723c2778cf8 ("cmd: pxe: reorder kernel treatment in label_boot").
As reported by Quentin Schulz, this introduces a regression with previously generated and working extlinux.conf files.
The situation is much complex than that:
- Since the commit d5ba6188dfbf ("cmd: pxe_utils: Check fdtcontroladdr in label_boot"),\
all the extlinux.conf with only "KERNEL /fitImage" don't work anymore. Those are generated by Poky/OE WIC bootimg-partition bootloader partition generator.
- With the changeset introduced by Patrick, this doesn't permit booting
those images, but it solves the issue by explicitly specifying we want to get the dtb from the fitImage with "FDT /fitImage", which is smart.
- If we start generating extlinux.conf with "FDT /fitImage", those won't
boot on previous u-boot versions (as I reported in the second link shared by Quentin).
But honestly, as of today we need to patch by reverting d5ba6188dfbf to use extlinux.conf with fitImage, so I'll vote to keep this solution and go forward and implement the change in the Poky/OE bootimg-partition bootloader partition generator with perhaps a check on the u-boot version.
Yocto could do this, though I'm not entirely sure they'll like it either. But you have plenty other distributions, and I imagine it is perfectly possible it does not have knowledge of the bootloader being used to boot the distribution.
I'm voting against because currently I see d5ba6188dfbf as a regression and distros are free to handle them how they wish, by e.g. reverting this patch only (which I have done for our boards for example). I'd expect Yocto to have a local patch reverting this in their next release, where U-Boot would be bumped to something containing this commit, for example.
By merging the patchset, we limit other possible implementations by adding one more logic we should then maintain for backward compatibility.
Ack, you're right.
Neil
Cheers, Quentin