JIRA: https://issues.redhat.com/browse/RHEL-79906
ARK: cki-project/kernel-ark!3701

redhat/configs: automotive: disable VFIO_PLATFORM

Review did not reveal any user of vfio-platform and it is disabled in
RHEL currently. Inherit from RHEL until required otherwise.

Signed-off-by: Eric Chanudet [email protected]

Closing this as we no longer intend to do this on cs9, possibly on cs10.

JIRA: https://issues.redhat.com/browse/RHEL-109755

Upstream Status: RHEL-Only

This commit removes most module disables from the automotive config, for the modules that are already enabled in RHEL. It does not enable any module that is not coming from the RHEL config. It also does not aim to enable any major built-in features which are not modules and that were previously purposedly disabled to reduce validation scope for functional safety.

This is done by generating a comparison between the current RHEL config and the corresponding automotive config, iterating through modules that were enabled in RHEL, and removing the forced disable from the automotive config. A few outliers are module configs that select some other option which cascades into other things getting enabled, in that case the module remains disabled to avoid more things getting enabled.

The proposed benefits of this approach are:

• Reduce the delta between the automotive config and the RHEL config. The automotive kernel is claimed to be based on the RHEL kernel, but then it goes and diverges from RHEL in thousands of configs.
• Reduce maintenance burden. Often when some new module is enabled in RHEL, due to the vastly divergence of the automotive config, it is unclear if that module should remain enabled in the automotive config or not, and that increases the maintenance cost. At least for modules, with this we can maintain a policy of inheriting modules from RHEL by default.
• There has been a very loose policy of what can actually be enabled in rhe automotive config. The result so far is that the config is very arbitrary and things get slowly re-enabled as users hit missing modules. With this, we can set the expectation that modules are available for development if they were in RHEL.
• This aims to have no impact on the set of built-in options for the kernel and should not enable any major features that were disabled before due for a safety argumentation. So it should not impact that, but still reduce the divergence from the RHEL kernel.
• We do not have to actually ship these additional modules in any automotive images or include them in initramfs. We can control whether these get included in images later on. But there is still at least an option for users to include the modules if they have been built.

Signed-off-by: Erico Nunes [email protected]

JIRA: https://issues.redhat.com/browse/RHEL-78292

Sync perf codebase with upstream v6.18. Also, update the tools' UAPI to current RHEL-10 kernel include/*.

Signed-off-by: Michael Petlan [email protected]

Also update source to reference srcrev. Since we access the directory as downstream-dtbs-%{srcrev} use the %{srcrev} commit path so that the downloaded source will match it.

ernunes (6a47170d) at 17 Dec 13:38

ernunes (55b3db93) at 17 Dec 13:36

ernunes (206d6668) at 17 Dec 13:33

... and 1 more commit

Wouldn't it make sense for this to be in the same commit as the previous MR so that the automotive config is not broken until this lands?

JIRA: https://issues.redhat.com/browse/VROOM-35402

Vendor out-of-tree modules have dependencies on hidden/private headers. In order to avoid maintaining changes to the kernel to expose the required declarations, which would cause conflicts during subsystem upgrades and a maintenance burden down the line, create a separate package that will produce the private headers under /usr/src/kernel where the kernel-devel package already produce the public kernel headers and Kbuild infrastructure for out-of-tree modules.

Signed-off-by: Eric Chanudet [email protected]

JIRA: https://issues.redhat.com/browse/RHEL-132320

ARK-MR: cki-project/kernel-ark!4232

Upstream Status: RHEL only

The S32G dwmac driver relies on SCMI clock support. Currently, the missing configuration causes a probe failure. This patch resolves the issue.

Signed-off-by: Alessandro Carminati [email protected]

Keep up with CS10 and merge the recent changes bumping the kernel to 6.12.0-160.

Straight-forward merge, no conflict.

Signed-off-by: Eric Chanudet [email protected]

It still sounds like a bad idea to me and perhaps there was not much thought into it in the cs9 case. It seems a bit counter-productive to try to run tests with multiple expensive debug options at the same time. Is it possible to switch to passing the enable argument when running tests which will actually care about checking it?

JIRA: https://issues.redhat.com/browse/VROOM-33533

vfio-platform is currently disabled in Centos Stream 10 since https://issues.redhat.com/browse/RHEL-78945 from a recommendation by the virt team. If we re-enable it to support the Qualcomm architecture it would be against that recommendation. However, Qualcomm drivers for many devices in the ES21 device tree depend on vfio-platform.

Without it, it'd affect the Qualcomm scmi porting, e.g. the Qualcomm Graphics solution will not work. So, re-enable it for the automotive use case.

Signed-off-by: Enric Balletbo i Serra [email protected]

I agree that we probably want to keep the default CONFIG_DEBUG_KMEMLEAK_DEFAULT_OFF=y. It is also what RHEL does.

Since CONFIG_DEBUG_KMEMLEAK=y is enabled in RHEL debug kernel, it seems that it is inherited already? Perhaps we should just consider moving it to common/debug?

Merge Request Required Information

JIRA: https://issues.redhat.com/browse/RHEL-116573
JIRA: https://issues.redhat.com/browse/RHEL-123791
JIRA: https://issues.redhat.com/browse/RHEL-125486

Upstream Status: git://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git

CVE: CVE-2025-38594
CVE: CVE-2025-38676
CVE: CVE-2025-38688
CVE: CVE-2025-39739
CVE: CVE-2025-39961
CVE: CVE-2025-39966
CVE: CVE-2025-40058

Depends: !1384
Depends: !1507
Depends: !1577

Summary of Changes

Updates IOMMU and DMA Mapping API subsystems through the v6.17 merge window.

The development branch currently has 178 commits that correspond to the IOMMU and DMA Mapping API subsystems update.

Conflicts

Legend: ds is used for downstream and us is used for upstream commits.

• A merge conflict in drivers/iommu/io-pgtable-arm.c function arm_lpae_do_selftests() due to fa26198d30f3 ("iommu/io-pgtable-arm: dynamically allocate selftest device struct") and 3318f7b5cefb ("iommu/io-pgtable-arm: Add quirk to quiet WARN_ON()") solved upstream in 879b141b7cfa ("Merge branches 'fixes', 'apple/dart', 'arm/smmu/updates', 'arm/smmu/bindings', 'fsl/pamu', 'mediatek', 'renesas/ipmmu', 's390', 'intel/vt-d', 'amd/amd-vi' and 'core' into next")
  ds: 3c8d3807 ("iommu/io-pgtable-arm: Add quirk to quiet WARN_ON()")
  us: 3318f7b5cefb ("iommu/io-pgtable-arm: Add quirk to quiet WARN_ON()")

• Modify/delete conflict in drivers/iommu/riscv/iommu.c. The file does not exist in the downstream repository, but it is modified by upstream commit 4316ba4a5033 ("iommu/pages: Remove the order argument to iommu_free_pages()")
  ds: 43f97752 ("iommu/pages: Remove the order argument to iommu_free_pages()")
  us: 4316ba4a5033 ("iommu/pages: Remove the order argument to iommu_free_pages()")

• Modify/delete conflict in drivers/iommu/riscv/iommu.c. The file does not exist in the downstream repository, but it is modified by upstream commit 3e8e986ce8a0 ("iommu/pages: Remove iommu_free_page()"). drivers/iommu/rockchip-iommu.c was left unchanged (unsupported driver)
  ds: ba089f18 ("iommu/pages: Remove iommu_free_page()")
  us: 3e8e986ce8a0 ("iommu/pages: Remove iommu_free_page()")

• Modify/delete conflict in drivers/iommu/riscv/iommu.c. Currently the file does not exist in the downstream repository, but it is modified by upstream commit 5087f663c21e ("iommu/pages: Remove iommu_alloc_page_node()")
  ds: 95ce8e3c ("iommu/pages: Remove iommu_alloc_page_node()")
  us: 5087f663c21e ("iommu/pages: Remove iommu_alloc_page_node()")

• Modify/delete conflict in drivers/accel/amdxdna/aie2_pci.c. The file does not exist in the downstream repository, but it is modified by upstream commit 7c8896dd4a2a ("iommu: Remove IOMMU_DEV_FEAT_SVA")
  ds: d6130fea ("iommu: Remove IOMMU_DEV_FEAT_SVA")
  us: 7c8896dd4a2a ("iommu: Remove IOMMU_DEV_FEAT_SVA")

• Merge conflict in drivers/iommu/iommu.c due to downstream-only commit 806bdc79 ("iommu/arm-smmu: workaround DMA mode issues"). This specific downstream implementation has been left untouched, and the adjacent code next to it removed as per patch being applied.
  ds: dd3f1431 ("iommu: Remove iommu_dev_enable/disable_feature()")
  us: f984fb09e60e ("iommu: Remove iommu_dev_enable/disable_feature()")

• Merge conflict in drivers/iommu/Makefile due to upstream-only 5c0ebbd3c6c6 ("iommu/riscv: Add RISC-V IOMMU platform device driver") conflicting with the patch being applied.
  ds: ed7a0f94 ("iommu: make inclusion of intel directory conditional")
  us: ddcc66cfe83a ("iommu: make inclusion of intel directory conditional")

• Merge conflict in drivers/iommu/Makefile due to upstream-only 5c0ebbd3c6c6 ("iommu/riscv: Add RISC-V IOMMU platform device driver") conflicting with the patch being applied.
  ds: f27e9572 ("iommu: make inclusion of amd directory conditional")
  us: 85ef671f9727 ("iommu: make inclusion of amd directory conditional")

• Merge conflict in drivers/dma/idxd/init.c due to context difference in idxd_remove().
  ds: 61ce77af ("dmaengine: idxd: Remove unnecessary IOMMU_DEV_FEAT_IOPF")
  us: 853b01b5efd7 ("dmaengine: idxd: Remove unnecessary IOMMU_DEV_FEAT_IOPF")

• Modify/delete conflict in drivers/iommu/riscv/iommu.c. The file does not exist in the downstream repository, but it is modified by upstream commit.

• Contextual conflict in drivers/iommu/virtio-iommu.c related to the addition of a 'const' type qualifier to viommu_ops (now removed) in the patch being backported and the fact that 72b6f7cd89cea ("iommu/virtio: Make instance lookup robust") had already been backported.
  ds: 3cd4fb5f ("iommu: Remove ops.pgsize_bitmap from drivers that don't use it")
  us: 8901812485de ("iommu: Remove ops.pgsize_bitmap from drivers that don't use it")

• Files in drivers/iommu/ related to currently unsupported downstream drivers have been reverted to their pre-patch state: exynos-iommu.c, mtk_iommu_v1.c, omap-iommu.c, rockchip-iommu.c, sprd-iommu.c, and sun50i-iommu.c.
  ds: 3fdead74 ("iommu: Remove iommu_ops pgsize_bitmap from simple drivers")
  us: cf39047e460e ("iommu: Remove iommu_ops pgsize_bitmap from simple drivers")

• Merge conflict in include/linux/iommu.h related to the documentation of pgsize_bitmap. Upstream commit 792ea7b6cafa ("iommu: Remove ops->pgsize_bitmap") already backported as downstream 5d2a8e11 removed the pgsize_bitmap entry while the base used for creating the patch still had pgsize_bitmap documented.
  ds: 9b96afd4 ("iommu: Deprecate viommu_alloc op")
  us: f842ea208e43 ("iommu: Deprecate viommu_alloc op")

Omitted

Omitted-fix: d41d75fe1b75 ("crypto: qat - fix DMA direction for compression on GEN2 devices")
Part of QAT subsystem update.

The following upstream commits have been omitted

• 01c13a1d0e347 ("iommu: make inclusion of riscv directory conditional")
  Currently RISCV is not supported downstream

• e436576b02315 ("iommu: make inclusion of arm/arm-smmu-v3 directory conditional")
  Reverted by upstream commit 9d49da438819 Revert "iommu: make inclusion of arm/arm-smmu-v3 directory conditional"

• 41cb08555c41 ("treewide, timers: Rename from_timer() to timer_container_of()")
  Backported in !1425

• 968e300068071 ("x86/cpuid: Set <asm/cpuid/api.h> as the main CPUID header")

• 46a7418a3aa6 ("iommu/apple-dart: Drop default ARCH_APPLE in Kconfig")
  Currently not supported downstream

• 2c223f7239f3 ("of: reserved_mem: Restructure call site for dma_contiguous_early_fixup()")
  Fixes 8a6e02d0c00e ("of: reserved_mem: Restructure how the reserved memory regions are processed"), currently not available downstream

• KVM Updates (see !1640):

1da19c5ce0533 ("iommu/amd: KVM: SVM: Delete now-unused cached/previous GA tag fields")
c4cdbaf9d81c8 ("iommu/amd: KVM: SVM: Use pi_desc_addr to derive ga_root_ptr")
95d50ebe6df80 ("iommu/amd: KVM: SVM: Pass NULL @vcpu_info to indicate "not guest mode"")
53527ea1b7022 ("iommu: KVM: Split "struct vcpu_data" into separate AMD vs. Intel structs")
3be405e89f3da ("iommu/amd: Document which IRTE fields amd_iommu_update_ga() can modify")
08d9ccdd1a5c7 ("iommu/amd: KVM: SVM: Infer IsRun from validity of pCPU destination")
0b2b541fa3cd8 ("iommu/amd: Factor out helper for manipulating IRTE GA/CPU info")
f965255dc5033 ("iommu/amd: KVM: SVM: Set pCPU info in IRTE when setting vCPU affinity")
6df262f915abc ("iommu/amd: KVM: SVM: Add IRTE metadata to affined vCPU's list if AVIC is inhibited")
a23480fe21de6 ("iommu/amd: WARN if KVM calls GA IRTE helpers without virtual APIC support")
b9e53f9ff4a88 ("iommu/amd: KVM: SVM: Allow KVM to control need for GA log interrupts")

• 212c439bdd8f ("iommu/arm: Add BBM Level 2 smmu feature")
  Depends on support to feature not yet available downstream (see !1662):

3df6979d222b8 ("arm64: mm: split linear mapping if BBML2 unsupported on secondary CPUs")
a166563e7ec37 ("arm64: mm: support large block mapping when rodata=full")
83bbd6be7d171 ("arm64/mm: Elide tlbi in contpte_convert() under BBML2")
212c439bdd8f9 ("iommu/arm: Add BBM Level 2 smmu feature")
5aa4b625762e5 ("arm64: Add BBM Level 2 cpu feature")

• 75952c497550 ("iommu/vt-d: Use pci_is_display()")
  Backported in !1492

• 9ffaf5229055 ("iommu/s390: Make attach succeed when the device was surprise removed")
  Backported in !1477

• b3506e9bcc77 ("iommu/s390: Fix memory corruption when using identity domain")
  Backported in !1477

RHEL-only

CONFIG_X86_POSTED_MSI=y.
File moved from redhat/configs/rhel/generic/CONFIG_X86_POSTED_MSI to redhat/configs/common/generic/x86/CONFIG_X86_POSTED_MSI.

Owners file update in documentation!985

Signed-off-by: Eder Zulian [email protected]

JIRA: https://issues.redhat.com/browse/RHEL-122494
Upstream-status: RHEL-only
ARK-MR: cki-project/kernel-ark!4184

CONFIG_ARCH_BCM was disabled accidentally when it was introduced as a new option, and it gates everything related to BCM2835.

Signed-off-by: Radu Rendec [email protected]