linux/meta-freescale.git/recipes-kernel, branch g2d-rename

linux-imx: Update to 6.18.20 (lf-6.18.y)

2026-09-17T20:51:28+00:00

Bump SRCREV to the latest lf-6.18.y tip and LINUX_VERSION to 6.18.20, bumping LOCALVERSION for the ABI change. Drop the ipu uapi integer types patch. Upstream now includes in include/uapi/linux/ipu.h and no longer defines uint32_t, uint16_t and uint8_t as macros, which is what the patch was carried for. The header still defines u32, u8 and __u32 as macros, so including after it remains unwise, but the standard integer type names are no longer affected. Also install the imx_drm.h uapi header via linux-imx-headers. It is needed by the imx-dpu-g2d recipe, which builds from source and includes . Signed-off-by: Stefan Eichenberger

linux-imx: Document the DELTA_KERNEL_DEFCONFIG migration guard

2026-09-10T17:12:25+00:00

The guard exists to abort the parse when a machine still sets the removed variable, so nothing reads a value and a task would run far too late. It tests whether the key is defined at all, which BitBake offers no declarative syntax for, and no declarative form raises an error either. Signed-off-by: Luciano Dittgen

linux-qoriq: Suppress vars.pkgspecific on the kernel-image override

2026-09-04T12:51:40+00:00

kernel.bbclass declares kernel-image in PACKAGES and sets its FILES to "", so this include is the only thing that adds zImage: the override is load-bearing. oelint-adv scans this layer alone and never parses that class, and neither PACKAGES_DYNAMIC nor the constant DB can declare a static package name to it. Hard-coding kernel-image would defeat KERNEL_PACKAGE_NAME, so the suppression is the honest form. Signed-off-by: Luciano Dittgen

linux-qoriq: Drop the dead RDEPENDS assignment for kernel-base

2026-09-04T12:51:40+00:00

kernel.bbclass sets no RDEPENDS for kernel-base; it links base to image through RRECOMMENDS since oe-core 1c90b27d2c. So this line assigned empty to an already-empty variable and never had the effect its comment claimed. Commit 5171b4b5 fixed it once, and the LF-6.12.20_2.0.0 resync reverted it. Making kernel-base stop pulling kernel-image needs RRECOMMENDS, which is a behaviour change and is not made here. Validated on MACHINE=ls1046ardb, DISTRO=fslc-framebuffer: do_package forced, and the buildhistory packages tree is byte-identical. Signed-off-by: Luciano Dittgen

kernel-module-imx-gpu-viv: Checksum the license header in the source

2026-09-03T14:52:26+00:00

SRC_URI fetches only the drivers/mxc/gpu-viv subpath of linux-imx, which carries no standalone license file, so checksum the header instead: 117 of its sources carry the Vivante dual MIT/GPL header. Unlike the generic text, this detects an upstream license change, which is what LIC_FILES_CHKSUM is for. Validated on MACHINE=imx8mp-ddr4-evk: do_populate_lic re-ran and passed, so the range and md5 resolve; the buildhistory packages tree is identical. Signed-off-by: Luciano Dittgen

kernel-module-isp-vvcam: Drop the variable from the license path

2026-09-03T14:31:36+00:00

LIC_FILES_CHKSUM wrote the path as ${S}/../LICENSE. S expands to ${UNPACKDIR}/${BP}/vvcam/v4l2, so the expanded path still carries variables that a standalone analyzer cannot resolve. A file:// path is already relative to S, so drop the prefix. The file and the md5 do not change. Validated on MACHINE=imx8mp-ddr4-evk: do_populate_lic re-ran and passed, so the path still resolves; the buildhistory packages tree is identical. Signed-off-by: Luciano Dittgen

linux-qoriq: Suppress vars.specific on kernel-class package names

2026-09-01T13:39:24+00:00

This include sets FILES:${KERNEL_PACKAGE_NAME}-image and RDEPENDS:${KERNEL_PACKAGE_NAME}-base. Both ${KERNEL_PACKAGE_NAME}-image and ${KERNEL_PACKAGE_NAME}-base are packages that oe-core kernel.bbclass places in PACKAGES; this include inherits that class (inherit kernel at the top). oelint-adv parses the include standalone, without oe-core's kernel.bbclass on its analyzer path, so it cannot see those PACKAGES and reports both assignments as oelint.vars.specific ("set specific to a package not known from PACKAGES"). This is a false positive from missing analyzer context, not a recipe defect: the assignments are correct kernel packaging -- they extend the image package's FILES and clear the base package's RDEPENDS. No structural fix applies. Re-declaring PACKAGES or PACKAGES_DYNAMIC in the include would duplicate the class and misplace package-split ownership, and the oelint.vars.specific rule has no constant-DB hook for registering known package names. Add an inline # nooelint: oelint.vars.specific to each assignment, with a comment explaining the kernel.bbclass provenance. Fixing it once on the shared include clears it for both consumers (linux-qoriq_6.12.bb and linux-fslc-qoriq_5.4.bb). Suppression-only change (comments and nooelint annotations only); the effective BitBake metadata is unchanged, so build and buildhistory validation were skipped. Signed-off-by: Luciano Dittgen

kernel-module-ar: Document INSANE_SKIP ldflags exception for oelint

2026-08-24T21:20:36+00:00

The userspace tools are linked by the module Makefile via CROSS_COMPILE and do not consume the OE LDFLAGS, so the ldflags QA check cannot pass and the INSANE_SKIP is required. Add an inline # nooelint directive with the semantic reason. No effective metadata changes. Signed-off-by: Luciano Dittgen

linux-qoriq: Document INSANE_SKIP buildpaths exception for oelint

2026-08-24T21:20:36+00:00

The kernel -src package ships generated build files (Makefiles, configs) that reference build paths, so the buildpaths QA check cannot pass and the INSANE_SKIP is required. Add an inline # nooelint directive with the semantic reason. No effective metadata changes. Signed-off-by: Luciano Dittgen

linux-fslc-qoriq: Drop SUMMARY/DESCRIPTION shadowed by the include

2026-08-18T20:26:37+00:00

linux-fslc-qoriq_5.4.bb set SUMMARY and DESCRIPTION above "require linux-qoriq.inc". The include sets both variables too, and since the require runs after these assignments, the include's values win and the recipe's own SUMMARY/DESCRIPTION never reach the package. oelint.var.override reported them as overridden. Remove the two dead assignments. This is value-preserving: the effective SUMMARY and DESCRIPTION already come from linux-qoriq.inc and are unchanged. HOMEPAGE is kept -- the include does not set it, so that assignment is live. Validated in a resolved qoriq environment (MACHINE=ls1046ardb, with linux-fslc-qoriq pinned as the kernel provider): SUMMARY and DESCRIPTION resolve to the linux-qoriq.inc values ("Linux Kernel for NXP QorIQ platforms" / "Linux kernel for NXP QorIQ and Layerscape processors, based on the NXP lf (Layerscape factory) kernel tree.") both before and after, and every task signature of the recipe is identical across the parent and this commit, so no generated output can move. (do_compile of this 5.4 kernel fails under the host's gcc 16 with -Werror=enum-int-mismatch, for reasons unrelated to and independent of this change and identical on both sides; the equivalence is therefore established from the resolved variables and the full set of task signatures.) Behaviour-preserving. Signed-off-by: Luciano Dittgen