On Thu, May 23, 2024 at 04:09:38PM -0500, Jon Humphreys wrote:
Tom Rini trini@konsulko.com writes:
On Wed, May 22, 2024 at 11:12:35PM -0500, Jon Humphreys wrote:
Tom Rini trini@konsulko.com writes:
On Tue, May 21, 2024 at 09:20:26PM -0500, Jon Humphreys wrote:
Tom Rini trini@konsulko.com writes:
On Fri, Apr 19, 2024 at 04:28:16PM -0500, Jonathan Humphreys wrote:
> Fill in the BeaglePlay's capsule GUID properties of the base binman capsule > nodes. > > Signed-off-by: Jonathan Humphreys j-humphreys@ti.com > --- > arch/arm/dts/k3-am625-beagleplay-u-boot.dtsi | 27 ++++++++++++++++++++ > arch/arm/dts/k3-am625-r5-beagleplay.dts | 15 +++++++++++ > 2 files changed, 42 insertions(+)
This series introduces failure to build in CI, and it's a little tricky to replicate locally, depending on your environment. You first need to NOT have BINMAN_INDIRS set and instead be using fake binaries. Second, it seems python version dependent perhaps? I don't see this on my host, but by:
- Using the CI container
- Setting up a virtualenv inside of it
- pip install -r tools/buildman/requirements.txt
I get: $ ./tools/buildman/buildman --keep-outputs --reproducible-builds -dvel --force-build -PEWM --output /tmp/am62x_beagleplay_r5 --board am62x_beagleplay_r5 Building current source for 1 boards (1 thread, 12 jobs per thread) arm: + am62x_beagleplay_r5 +(am62x_beagleplay_r5) Image 'tiboot3-am62x-gp-evm.bin' is missing optional external blobs but is still functional: ti-fs-gp.bin +(am62x_beagleplay_r5) +(am62x_beagleplay_r5) /binman/tiboot3-am62x-gp-evm.bin/ti-fs-gp.bin (ti-sysfw/ti-fs-firmware-am62x-gp.bin): +(am62x_beagleplay_r5) Missing blob +(am62x_beagleplay_r5) binman: object of type 'NoneType' has no len() +(am62x_beagleplay_r5) make[1]: *** [Makefile:1126: .binman_stamp] Error 1 +(am62x_beagleplay_r5) make: *** [Makefile:177: sub-make] Error 2 0 0 1 /1 am62x_beagleplay_r5
Tom, this is failing in the CI container because the container is missing the mkeficapsule tool.
To solve this, we just need to add it to the CI container.
My understanding of binman's handling of missing bintools is that it should gracefully continue with fake data, so that buildman can successfully test out builds for boards even when you don't have all the required bintools. If I have that correct, I can also create a patch to properly handle this when using mkeficapsule. But I want to verify this is the desired behavior, since mkeficapsule isn't a unique or vendor specific tool, so shouldn't we require it as part of the U-Boot build environment and err out if it is missing?
Perhaps it's a binman issue since we build mkeficapsule or at least should be? Neha?
Never mind - I figured it out.
The mkeficapsule tools is built by u-boot if TOOLS_MKEFICAPSULE config is set. I didn't explicitly set it because it is implied by EFI_CAPSULE_ON_DISK config. But this is only set for the A core u-boot, as that is what would apply the capsules. SPL running on the R cores does not need this. But that then means that the R core u-boot builds don't have TOOLS_MKEFICAPSULE set and if that is all that is being built (as in the case of buildman), the mkeficapsule tool isn't built and so is missing.
So I need to explicitly set TOOLS_MKEFICAPSULE in the R core defconfigs. I'll repost the patches.
Interesting. My next thought here is that whatever symbol is allowing for "make a capsule" should be select'ing TOOLS_MKEFICAPSULE and so the current logic is a bit flawed. I'm just not sure off-hand where it should be instead, do you have some ideas now? Thanks.
There is no config that indicates that capsules will be generated for the resulting binary. The only thing I can think of is to scan the board's DTB for the presence of a capsule binman node, and then set the TOOLS_MKEFICAPSULE config. But this seems very complicated to do at build time.
Yeah, that's the wrong direction I think. We should perhaps just make TOOLS_MKEFICAPSULE default y if EFI_LOADER instead.