trusted-firmware-a-qcom, optee-os-qcom: split into per-platform recipes - #2910
Conversation
bf32dd5 to
eafdcda
Compare
eafdcda to
03410c6
Compare
03410c6 to
08a9319
Compare
cc11d14 to
c1ccd65
Compare
Royyan Zahir (royzah)
changed the title
c1ccd65 to
26c014b
Compare
578d9ca to
7f81cbd
Compare
Royyan Zahir (royzah)
changed the title
|
Added the OP-TEE split so the |
6d72129 to
f828fec
Compare
Dmitry Baryshkov (lumag)
left a comment
Dmitry Baryshkov (lumag)
left a comment
There was a problem hiding this comment.
I think you might want to split/squash the commits.
f828fec to
80bb05c
Compare
|
Restructured and rebased on master.
|
|
CI on the previous head caught one real failure: Fix is one line there: qualcomm-linux/meta-qcom-distro#422. Needs to merge before this can go green (CI tracks that repo at Rest of that run was clean: Dmitry Baryshkov (@lumag) Ricardo Salveti (@ricardosalveti) Sumit Garg (@b49020) Jose Quaresma (@quaresmajose) heads up, the rename touches that list. |
80bb05c to
27abd4f
Compare
|
Rebased on master, picks up the |
|
"This branch has conflicts that must be resolved" |
One recipe gated on MACHINE builds a single TF-A platform, so the platform, the FIP address and the qtiseclib blob all had to be keyed on the machine. Give each device its own recipe holding those, and keep the shared source and build rules in the include. The blob follows its recipe, so a platform without qtiseclib no longer claims its licence. FIP_ELF_ADDR has no sensible default, so refuse a device recipe that leaves it unset. Signed-off-by: Royyan Zahir <royzah@gmail.com>
optee-os-qcom picks its OP-TEE platform from MACHINE, so a build can only ever hold one, while the TF-A recipes are now per device. Give each OP-TEE platform its own recipe and point each TF-A recipe at the one it needs. The devkit and optee-test stay machine-scoped; a rootfs targets a single machine. Signed-off-by: Royyan Zahir <royzah@gmail.com>
…hine Now that both are per platform, nothing ties them to qcm6490 or qcs9100. Widen them so a machine like qcom-armv8a can build kodiak and lemans firmware side by side. Signed-off-by: Royyan Zahir <royzah@gmail.com>
27abd4f to
a5c633a
Compare
|
Rebased on master, conflicts resolved (picked up the OP-TEE 20260721 bump). Dmitry Baryshkov (@lumag) PTAL |
#422) qualcomm-linux/meta-qcom#2910 splits `trusted-firmware-a-qcom` per device, so the recipe carrying the qtiseclib blob becomes `trusted-firmware-a-qcom-rb3gen2` and stops matching the exception. The image licence gate then fails on rb3gen2-core-kit-open-fw. Add the new name; the old one stays until the split lands. Needed before meta-qcom#2910 can go green. CI there tracks this repo at `main` with no pin, so merging is enough.
|
Thanks Dmitry Baryshkov (@lumag)! Build is green (67/67) and the licence gate passes now that meta-qcom-distro#422 landed. Only red is the workflow security scan, which fails on other PRs too. Good to merge? |
Test Results 117 files 698 suites 7h 40m 11s ⏱️ Results for commit a5c633a. ♻️ This comment has been updated with latest results. |
|
Royyan Zahir (@royzah) it requires a second review. |
086b40a
into
qualcomm-linux:master
5 participants
qcom-armv8a cannot build kodiak and lemans firmware together: TF-A is gated on MACHINE and builds one platform, and optee-os-qcom picks its platform the same way.
Split both per platform, each TF-A recipe depending on its own OP-TEE, then widen the machine gate once nothing is tied to a single machine. Machines opt in through EXTRA_IMAGEDEPENDS as before.
Renames recipes, so layers with a trusted-firmware-a-qcom or optee-os-qcom bbappend or DEPENDS need updating.
Still machine-scoped on a multi-platform build: BL33 (
u-boot.bin) and the sharedoptee/deploy dir.Grew out of #2707.