linux/meta-freescale.git/recipes-devtools/qemu, branch master

qemu-qoriq: Use cp --no-preserve=ownership in do_install_ptest per oelint

2026-09-01T20:57:16+00:00

The recipe carries its own do_install_ptest (overriding the one in qemu-qoriq.inc) that copies the built tests tree into the ptest package with plain cp -rL, which oelint.task.nocopy flags because cp inherits the invoking user's ownership. This is a recursive copy that must dereference symlinks (-L), which install(1) cannot do, so add --no-preserve=ownership, the alternative the rule documents, and keep the -r/-L flags. This do_install_ptest is a verbatim fork of oe-core's qemu.inc, whose copy line is still plain "cp -rL ${B}/tests ${D}${PTEST_PATH}". This change diverges from that source on purpose to satisfy lint; a future re-sync from oe-core should carry the flag forward rather than drop it. Validated with buildhistory: rebuilt qemu-qoriq for MACHINE=ls1046ardb with and without this change; the qemu-qoriq-ptest package is byte-identical (file list, size, dependencies unchanged), because packaging runs under pseudo and records the files as root-owned regardless. Signed-off-by: Luciano Dittgen

qemu-qoriq: Use cp --no-preserve=ownership in do_install_ptest per oelint

2026-09-01T20:57:16+00:00

do_install_ptest copies the built tests tree into the ptest package with plain cp -rL, which oelint.task.nocopy flags because cp inherits the invoking user's ownership. This is a recursive copy that must dereference symlinks (-L), which install(1) cannot do, so add --no-preserve=ownership, the alternative the rule documents, and keep the -r/-L flags. This do_install_ptest is a verbatim fork of oe-core's qemu.inc, whose copy line is still plain "cp -rL ${B}/tests ${D}${PTEST_PATH}". This change diverges from that source on purpose to satisfy lint; a future re-sync from oe-core should carry the flag forward rather than drop it. Validated with buildhistory: rebuilt qemu-qoriq for MACHINE=ls1046ardb with and without this change; the qemu-qoriq-ptest package is byte-identical (file list, size, dependencies unchanged), because packaging runs under pseudo and records the files as root-owned regardless. Signed-off-by: Luciano Dittgen

Merge pull request #2645 from lucianogdittgen/oelint-vars-inconspaces

2026-08-25T16:24:43+00:00

Clear all oelint.vars.inconspaces findings

qemu-qoriq: Move separator space out of conditional EXTRA_OECONF append

2026-08-25T14:51:47+00:00

The mipsarcho32 EXTRA_OECONF:append kept its separator space inside ${@bb.utils.contains(...)} (' --disable-capstone'), so the :append value did not begin with a literal space and tripped oelint.vars.inconspaces. Move the space outside the expansion, matching the EXTRA_OECONF:append line directly above it. When the multilib branch is taken the appended token is identical; otherwise the append adds only whitespace, which is inert in EXTRA_OECONF. Validated with buildhistory: baseline and candidate builds of qemu-qoriq (ls1046ardb), forcing do_package. The mipsarcho32 override is not active on this build, so the only buildhistory delta is metadata-revs; the qemu-qoriq packages are byte-identical. Signed-off-by: Luciano Dittgen

qemu-qoriq: Document INSANE_SKIP arch exception for oelint

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

QEMU ships firmware and BIOS blobs built for guest architectures other than the build target, so the arch QA check flags them by design and the INSANE_SKIP is required. Add an inline # nooelint directive with the semantic reason. No effective metadata changes. Signed-off-by: Luciano Dittgen

qemu-qoriq: Drop LIC_FILES_CHKSUM duplicated from the include

2026-08-17T20:59:56+00:00

qemu-qoriq_4.2.bb set LIC_FILES_CHKSUM to a value byte-identical to the one already provided by qemu-qoriq.inc, which the recipe requires on the line above it. The local assignment therefore only re-stated the include and tripped oelint.var.override without changing anything. Remove the duplicate and rely on the include. This is value-preserving: the effective LIC_FILES_CHKSUM is unchanged. Validated with a real build in a resolved qoriq environment (MACHINE=ls1046ardb, sharing the existing download/sstate caches): qemu-qoriq was built at the parent commit and at this commit and their Yocto buildhistory compared. The only delta is metadata-revs (the layer commit hash); packages, versions, runtime dependencies and file lists are unchanged. Behaviour-preserving. Signed-off-by: Luciano Dittgen

qemu-qoriq: Drop the fossil SRC_URI and its orphaned checksum

2026-08-07T12:29:59+00:00

qemu-qoriq_4.2.bb requires this include at line 3 and then replaces SRC_URI wholesale at line 12 with a gitsm:// fetch of the NXP fork plus a pinned SRCREV. Everything the include declared for the old tarball build has been unreachable ever since. That the block is dead is not an inference: of the 25 file:// entries it lists, only three exist on disk -- powerpc_rom.bin, run-ptest and 0002-Add-subpackage-ptest-which-runs-all-unit-test-cases-.patch, which are exactly the three the .bb re-lists. The other 22, including the whole CVE-2020-* series, have no corresponding file. Were this SRC_URI ever selected, do_fetch would fail outright. SRC_URI[sha256sum] is removed with it. A plain SRC_URI assignment does not clear varflags, so the tarball checksum survived the .bb's override and was left describing a URL that is no longer fetched. The two halves were measured separately, because they do not cost the same: dead SRC_URI block removed all 19 task signatures identical SRC_URI[sha256sum] removed 18 of 19 signatures change The block removal is a true no-op. The checksum removal takes the varflag out of do_fetch's basehash and that cascades downstream, so this commit does invalidate sstate for qemu-qoriq and forces a rebuild. It does not change what gets built: with the change applied, do_fetch, do_unpack and do_patch all succeed, and the assembled patched source tree is byte-identical to the baseline -- 66851 files excluding .git, sha256 1666963241b444dcae50a92f6785c291477bf0eacaad53055cad34ad1e81ae45 on both sides, from a clean rebuild on each. bitbake -e on ls1046ardb shows the expanded SRC_URI, SRCREV, FILESPATH, S, B and WORKDIR unchanged; the only delta in the recipe data is the varflag itself. No buildhistory comparison was run, so packaged output is not measured here. The claim made is narrower and is what the evidence supports: every input to do_configure and beyond is identical, so the forced rebuild should reproduce the same output. A layer-wide oelint-adv 9.10.1 rescan clears two findings and introduces none: oelint.vars.srcurimutualex at the checksum, and an oelint.var.override for SRC_URI that only existed because both files set it. bitbake -p parses 2222 entries with 0 errors. Signed-off-by: Luciano Dittgen

qemu-qoriq: Drop the dead class-native and class-nativesdk overrides

2026-08-07T12:29:58+00:00

qemu-qoriq_4.2.bb sets BBCLASSEXTEND = "" and neither file inherits native or nativesdk, so these class-specific overrides -- carried over from the upstream qemu recipe this one forked -- can never be selected. Removed: qemu-qoriq.inc do_configure:prepend:class-native() qemu-qoriq_4.2.bb EXTRA_OECONF:append:class-nativesdk qemu-qoriq_4.2.bb do_install:append:class-nativesdk() qemu-qoriq_4.2.bb PACKAGECONFIG:class-nativesdk The empty BBCLASSEXTEND is deliberate and load-bearing, so the other remedy oelint.vars.inappclassoverride documents -- declaring BBCLASSEXTEND or inheriting the class -- is not available here. The anonymous python in qemu-qoriq_4.2.bb appends BPN with "-qoriq" stripped to PROVIDES, so a native or nativesdk variant would PROVIDE qemu-native and collide with the recipe of that name in openembedded-core. Validated with bitbake -e on ls1046ardb before and after. The expanded OVERRIDES contains class-target and neither class-native nor class-nativesdk, and the only difference between the two captures is the disappearance of the dead variable itself; EXTRA_OECONF, PACKAGECONFIG, do_configure and do_install are byte-identical. bitbake -p parses 2222 entries with 0 errors. A layer-wide oelint-adv 9.10.1 rescan clears 4 oelint.vars.inappclassoverride findings and introduces none. It also clears one oelint.spaces.lineend finding at qemu-qoriq_4.2.bb:37, an analyzer artifact of the inline python inside the removed do_install block. Signed-off-by: Luciano Dittgen

recipes: Document the remaining custom tasks per oelint

2026-08-03T19:13:30+00:00

Set doc strings on the custom functions oelint-adv reports via 'oelint.task.docstrings': the three i.MX boot firmware deploy helpers, the libubootenv uboot-config fixup handler, the qemu-qoriq wrapper generator, the testfloat SoftFloat unpack step, the linux-imx device tree compatibility handler and the imx-mp4-parser QA-skip hook. imx-mp4-parser reuses the wording already used for the identical hook in imx-parser, so the two prebuilt-binary recipes read the same. opencv and libcamera are deliberately left alone. Both recipes carry their upstream recipe as a marked verbatim block ("meta-openembedded copy"), and the two functions oelint flags -- do_unpack_extra and do_package_recalculate_ipa_signatures -- sit inside those blocks, where meta-openembedded sets no doc string. Annotating them would break the diff against upstream for no functional gain, so those two findings stay visible. bitbake never reads the 'doc' flag when computing task checksums -- bb/data.py build_dependencies() only consults vardeps, vardepvalue, vardepsexclude, exports, postfuncs, prefuncs, lineno and filename -- so these additions cannot change a signature or invalidate sstate. Verified by re-linting each file. Signed-off-by: Luciano Dittgen

qemu-qoriq: Install the ptest Makefile fragment instead of copying it

2026-08-03T19:13:00+00:00

Both do_install_ptest implementations placed tests/Makefile.include into ${PTEST_PATH}/tests with a bare "cp", so the installed mode came from the source tree filtered through the building user's umask. The file is plain ptest data, so give it an explicit 0644. The surrounding "cp -rL ${B}/tests" is deliberately left alone: it copies a whole directory tree and dereferences symlinks, which install cannot express. The .inc and the 4.2 recipe carry the same do_install_ptest body, so both are updated together to keep them in sync. Resolves two oelint.task.nocopy findings. Validation: - Built qemu-qoriq on a QorIQ e5500 target with ptest in DISTRO_FEATURES, as a bound baseline/candidate pair with do_install forced - Buildhistory and a full ${D}/${PKGDEST} snapshot are identical on both sides; tests/Makefile.include is 0644 either way, so the recipe now states the mode the source tree was already producing - Reaching do_install_ptest needs the ptest feature plus local QA workarounds, so this is not a configuration the layer ships Signed-off-by: Luciano Dittgen