[edera-rebase-app]

Nightly rebase of edera/mainline (017f96b23bf1) onto master (17bb0e2f6c80), driven by Claude. Not pushed automatically because the checker found drift. Run log.

Important

Do not use the merge button: this branch replaces edera/mainline rather than merging into it. Approve the pull request to land it; the land workflow force-pushes edera/mainline to the approved head, pinned to 017f96b23bf1, and keeps a backup.

Independent checks

Builds: success (x86_64 and arm64 vmlinux with the Edera features built in; per-build results in the run).

Old tip	017f96b23bf1
New tip	e6149d463658
Old upstream base	864c07de0c7d
New upstream base	17bb0e2f6c80

• New branch is based on the upstream tip.
• Downstream series is linear.
• All 94 downstream commits replayed; 2 changed only in hunk context (listed below).
• Downstream drift: the tree changed by more than the upstream delta.

Commits whose hunk context moved (own +/- lines identical)

• ea62ac9f1b0f iommu/xen: Add Xen PV-IOMMU driver → 69077f43772f
• bd1c0dba57c3 sysctl: wire up PaX softmode setting → 8d6b01d0b725

Drift (changed lines: upstream delta vs. actual change)

--- upstream-lines
+++ actual-lines
@@ -222987,6 +222987,18 @@
 +	return ret;
 +}
 +
++#ifdef CONFIG_OPENPAX_SOFTMODE
++static const struct ctl_table pax_table[] = {
++	{
++		.procname       = "softmode",
++		.data           = &pax_softmode,
++		.maxlen         = sizeof(int),
++		.mode           = 0600,
++		.proc_handler   = proc_dointvec,
++	},
++};
++#endif
++
 +static const struct ctl_table sysctl_subsys_table[] = {
 +	{
 +		.procname	= "sysctl_writes_strict",
@@ -223034,6 +223046,9 @@
 +int __init sysctl_init_bases(void)
 +{
 +	register_sysctl_init("kernel", sysctl_subsys_table);
++#ifdef CONFIG_OPENPAX_SOFTMODE
++	register_sysctl_init("kernel/pax", pax_table);
++#endif
 +
 +	return 0;
 +}
@@ -223067,6 +223082,18 @@
 -	mutex_unlock(&static_key_mutex);
 -	return ret;
 +	return -ENOSYS;
+-#ifdef CONFIG_OPENPAX_SOFTMODE
+-static const struct ctl_table pax_table[] = {
+-	{
+-		.procname       = "softmode",
+-		.data           = &pax_softmode,
+-		.maxlen         = sizeof(int),
+-		.mode           = 0600,
+-		.proc_handler   = proc_dointvec,
+-	},
+-};
+-#endif
+-
 -static const struct ctl_table sysctl_subsys_table[] = {
 -#ifdef CONFIG_SYSCTL
 -	{
@@ -223116,6 +223143,9 @@
 -int __init sysctl_init_bases(void)
 -{
 -	register_sysctl_init("kernel", sysctl_subsys_table);
+-#ifdef CONFIG_OPENPAX_SOFTMODE
+-	register_sysctl_init("kernel/pax", pax_table);
+-#endif
 -
 -	return 0;
 -}

Claude's report

Summary

Rebased edera/mainline (017f96b23bf1) from merge base 864c07de0c7d (v7.2-rc5-26646-g864c07de0c7d) onto upstream 17bb0e2f6c80 (v7.2-rc5-31263-g17bb0e2f6c80, "Merge tag 'pci-v7.4-changes'"). That covers 4215 upstream non-merge commits from the v7.4 merge window. All 94 downstream commits were replayed onto the new tip e6149d463658 on rebase-result. One conflict was resolved and no commits were dropped. verify-upstream-rebase.sh passes on base, linearity and commit count (94/94; 2 commits differ only in hunk context). It reports drift in kernel/sysctl.c. That drift comes entirely from the one conflict resolution below: the PaX sysctl table moved along with the upstream code it sits next to. So I expect the checker to flag drift, and this to go to review as a PR.

Needs a decision

None.

Conflicts resolved

• sysctl: wire up PaX softmode setting (bd1c0dba57c3 → 8d6b01d0b725), kernel/sysctl.c. It collided with upstream 443bea2f71c5 ("sysctl: consolidate CONFIG_SYSCTL into a single block in kernel/sysctl.c"), building on acd19f2c56f2. Upstream moved sysctl_subsys_table[] and sysctl_init_bases() up into the single #ifdef CONFIG_SYSCTL block (after proc_do_static_key()), dropped the inner #ifdef CONFIG_SYSCTL around sysctl_writes_strict, and added a proc_do_static_key() stub in the #else branch. I took upstream's layout unchanged and placed the downstream additions at the new location: the #ifdef CONFIG_OPENPAX_SOFTMODE pax_table[] straight before sysctl_subsys_table[], and the register_sysctl_init("kernel/pax", pax_table) call straight after the "kernel" registration in sysctl_init_bases(). The commit's own +/- lines are byte-identical to the original. Only their position changed. That is the four drift hunks the checker lists for kernel/sysctl.c. The only behavioural difference is that the PaX table is now compiled only under CONFIG_SYSCTL. That matches upstream, where sysctl_init_bases() itself now exists only under CONFIG_SYSCTL, and register_sysctl_init() has nothing to register into without it.

Dropped downstream commits

None.

Build fixes

None.

Upstream commits that meet the series

Upstream commits that touch a file the series also touches (18 shared files):

• 1c8f6308e557 iommu/dma: Catch scatterlist length overflows: fix (Fixes: 809eac54cdd6, Cc: stable). The downstream commit iommu/dma: Don't reserve PCI windows under a paravirtual IOMMU also edits drivers/iommu/dma-iommu.c, but only in iova_reserve_iommu_regions(). In the final tree, both the overflows_type(s_length, s->length) check and the check_add_overflow(prev->length, pad_len, ...) check are present in iommu_dma_map_sg() with their -EOVERFLOW → out_restore_sg error paths. The fix is intact.
• fdd81f15fc43 iommu/dma: Clean up redundant MSI cookie check: iommu_dma_sw_msi() now relies on the core dispatching by cookie_type. The Xen PV-IOMMU uses core-managed paging/identity domains and no custom cookie, so it is unaffected.
• 1ae140d0f272 iommu: Make iommu-dma helpers private: moves declarations into iommu-priv.h. xen-iommu.c uses none of them; it already includes iommu-priv.h for xen_iommu_manages_iova(). Both hunks merged cleanly.
• 2690d4156b51 iommu/hyperv: Add para-virtualized IOMMU support for Hyper-V guest: adds HYPERV_PVIOMMU (default HYPERV, so it is now built in alongside the Xen PV-IOMMU) and sets x86_init.iommu.iommu_init = hv_iommu_init in hyperv_init(). The downstream nested-on-Xen path brings Hyper-V up through hyperv_init_nested_on_xen(), not hyperv_init(), so hv_iommu_init is never installed in that mode. There is no other caller. I see no collision. hv_is_hyperv_initialized(), which downstream makes return true when nested on Xen, is checked inside hv_iommu_init(), but it can only be reached through hyperv_init().
• 118bdac10546 Drivers: hv: Add logical device ID registry for vPCI devices: pci-hyperv now registers a per-domain prefix in an rhashtable at probe time. That table is initialised in hv_common_init(), which hyperv_init_nested_on_xen() also calls, so vPCI probe in nested-on-Xen mode finds it initialised. The retarget path's device_id change is a pure refactor next to the downstream hv_msi_get_int_vector() change.
• d0ec95a8a4e7 hv_netvsc: Advertise the SR-IOV capability for CoCo VMs: removes hv_is_isolation_supported() gates and only affects CoCo VMs. The downstream DMA-mapping change for nested-on-Xen transmit pages is elsewhere in netvsc.c.
• 6eecb67b69af xen/grant-table: stop setting PG_private on pages for grant mapping: gnttab_pages_set_private()/clear_private() no longer set or test PG_private. The downstream gnttab_alloc_pages_node() (with gnttab_alloc_pages() now a wrapper around it) still goes through these helpers. No downstream code tests PG_private or uses page->private in grant-table.c, unpopulated-alloc.c or balloon.c. No collision.
• bbe5d3b3d402 treewide: remove PagePrivate() and PG_private from comments and docs: comments only.
• e59c9a971a9f mm/vma: make MAP_PRIVATE-mapped /dev/zero mappings truly anonymous and 4b019e056f6c mm: implement file_is_dev_zero() to uniquely identify /dev/zero: these touch mm/vma.h and assert_sane_pgoff(). The downstream mman: enforce MPROTECT when enabled for a task only changes map_deny_write_exec(), and no downstream code depends on vm_file/vma_is_anonymous().
• 47e725702b0b mm: make per-VMA locks available universally, 35e340a47888 mm: userland pgtable freeing is RCU-safe now, remove leftover bits, 69d38c792c53 mm: Add copy_remote_mm_str(): touch mm_types.h, mm/util.c and tools/testing/vma/include/dup.h next to the OpenPaX pax_flags field, the brk randomisation and the exec arg-block randomisation. All merged cleanly, and the downstream code uses none of the changed interfaces.
• 958c01049e0a bootconfig: Skip printing early params to cmdline from bootconfig, 677f83795fdf bootconfig: Move BOOTCONFIG_FOOTER_SIZE to include/linux/bootconfig.h: pax_softmode= is registered with __setup(), not early_param(), so bootconfig still applies it.
• 443bea2f71c5, acd19f2c56f2 (sysctl CONFIG_SYSCTL consolidation): see the conflict above. 242bac52294e sysctl: Disallow partial updates for erroneous sysctl vectors, d08ab285b096 sysctl: Reject uint arrays before calling the general proc_vec, 2bfbb09b505c sysctl: Split data conversion and file position handling: change vector/uint handling inside proc_dointvec and friends. kernel.pax.softmode is a single int handled by proc_dointvec, so its behaviour does not change.
• 641a9b9cc697 dma-mapping: use exact allocation for DMA pages, 95d2fec40c48 dma-mapping: rename dma_opt_mapping_size(): touch dma-iommu.c. No downstream code calls the renamed function or the changed allocation path.
• 0bf031ba8c49 iommu: Add Broadcom BCM2712 IOMMU driver, 42eed9bb9ca4 iommu: apple-dart: Support specifying the DMA aperture in the DT: Kconfig/Makefile neighbours of the Xen PV-IOMMU entries only.
• 28d5df7b7274 KVM: split CONFIG_KVM_PRIVATE_MEMATTR_VM out of KVM_VM_MEMORY_ATTRIBUTES, 537ec2b15fdd KVM: Let userspace disable per-VM mem attributes, enable per-gmem attributes, 681f3b68f435 KVM: guest_memfd: Fix the in-place conversion documentation (fix, documentation only), 264d8fd2794f bpf, keys: Add a bpf keyring for program signature validation, abaa0835b4f3 bpf: Refuse caller-supplied keyrings when the bpf one is active: touch arch/Kconfig or kernel-parameters.txt away from the downstream ARCH_MMAP_RND_BITS default change and the pax_softmode= doc entry. No interaction.

Builds

BUILDS_PENDING

[edera-rebase-app]

Superseded by #56.