Repository navigation
Rebase edera/mainline onto master @ 666539eecdf4 - #115
Closed
edera-rebase-app[bot] wants to merge 10000 commits into
Closed
edera-rebase-app[bot] wants to merge 10000 commits into
edera-rebase-app[bot] wants to merge 10000 commits into
Conversation
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Nightly rebase of
edera/mainline(7ad1695c933c) ontomaster(666539eecdf4), driven by Claude. Not pushed automatically because the checker found drift and a build failure and Claude's report asks for a decision. Run log.Important
Do not use the merge button: this branch replaces
edera/mainlinerather than merging into it. Approve the pull request to land it; the land workflow force-pushesedera/mainlineto the approved head, pinned to7ad1695c933c, and keeps a backup.Independent checks
Builds: failure (x86_64 vmlinux on the Edera host config and arm64 vmlinux, both with the Edera features built in; per-build results in the run).
The build failed and Claude's repair round (success) left no new branch; this is the unrepaired rebase.
7ad1695c933cbd59e1cfc08d864c07de0c7d666539eecdf4Commits whose hunk context moved (own +/- lines identical)
ea62ac9f1b0f iommu/xen: Add Xen PV-IOMMU driver→564fa331b3d7d8221bf28f22 x86/CPU/AMD: avoid printing reset reasons on Xen domU→644ee3e585a7bd1c0dba57c3 sysctl: wire up PaX softmode setting→9841c6cb480d849ed61740ea use maximum supported mmap rnd entropy by default→5a5b2e82cb645b1ff5bbc8cb 9p/xen: carry the 9P session across a restore→cfa61581f94785e1da02b838 9p/xen: end foreign access on the grant refs we granted, not the intf page's→35903485bd72Commits lost, added, or whose own +/- lines changed
f492e4d0cb26 9p/xen: reconnect to the backend on resume365523c9c55c 9p/xen: reconnect to the backend on resume3f7ca89e3d70 xen/gntdev: don't copy to user space under priv->lock (#79)5209f7143a73 xen/gntdev: don't copy to user space under priv->lock (#79)Drift (changed lines: upstream delta vs. actual change)
Claude's report
Summary
I rebased
edera/mainline(126 non-merge commits, old tip7ad1695c933c) from merge base864c07de0c7d(v7.2-rc5-26646-g864c07de0c7d) onto666539eecdf4(v7.2-rc5-43417-g666539eecdf4), a merge-window range of 15798 upstream commits. The result isrebase-resultatbd59e1cfc08d. All 126 downstream commits were replayed in order and none were dropped. Four commits conflicted and were resolved by hand: sysctl, arch/Kconfig, 9p/xen resume, and xen/gntdev. The x86_64 and arm64vmlinuxbuilds both pass. I expectverify-upstream-rebase.shto exit 1 (drift), not 0. Every drift line and every changed commit it lists comes from one of the four resolutions below; I ran it and checked this.Needs a decision
UNSURE:
xen/gntdev: don't copy to user space under priv->lock (#79)versus upstream7620a997e485 xen/gntdev: fix deadlock between map ioctl and mmap(Cc: stable). Both commits fix the same lockdep/deadlock bug ingntdev_ioctl_map_grant_ref()by droppingpriv->lockbeforecopy_to_user(). They differ only on the-EFAULTpath.gntdev_find_map_index(op.index, op.count), and unlinks and puts whatever it finds.refcount_inc(&map->users)before unlocking. On failure it unlinks the map only if that exact pointer is still onpriv->maps, then drops both references. On success it drops the extra reference.The two readings:
copy_to_userunderpriv->lock; re-lock before touching the list; a concurrent unmap is tolerated). It also cannot unlink a different map that reused the same index after a concurrent unmap and remap, which upstream's index lookup can.I committed (b). The final
drivers/xen/gntdev.cis byte-identical to the file on the old downstream tip, so this commit shows as changed in the checker and upstream's three-EFAULT-path lines appear as drift. If you prefer upstream's code, drop the downstream commit.e79780b9d1b6 xen/gntdev: reset the copy batch's writeable flagdoes not depend on it.UNSURE:
x86/CPU/AMD: avoid printing reset reasons on Xen domU(oldd8221bf28f22) is now probably dead code. Upstreamea51d1a4c31f x86/CPU/AMD: Don't print reset reason in VMs(Cc: stable) makesprint_s5_reset_status_mmio()return early whenX86_FEATURE_HYPERVISORis set (arch/x86/kernel/cpu/amd.c:1373). The downstream check (XENPV && !xen_initial_domain(),amd.c:1377) runs after it. The downstream commit replayed cleanly and I kept it unchanged. The two readings:#include <xen/xen.h>. Drop it in a follow-up.One behaviour note: upstream also suppresses the message on a Xen PV dom0 whenever dom0 sees the CPUID hypervisor bit, which I expect it normally does. The downstream patch had deliberately left dom0 printing. Nothing to do unless dom0 must keep printing the reset reason.
UNSURE: Upstream
228e21764b38 Drivers: hv: Run VMBus kexec cleanup from syscore shutdownchanges the kexec path. The VMBus kexec teardown (vmbus_initiate_unload, SynIC cpuhp removal) now runs fromhv_vmbus_syscore_shutdownwhenever a kexec is in progress, regardless of hypervisor type. Before, it ran only throughhv_machine_shutdown, which the downstream Xen PV dom0 nested-on-Hyper-V path never installed.hyperv_remove_xen_vmbus_irq().No code was changed. Decide whether the Xen VIRQ bridge should also be torn down there.
Conflicts resolved
sysctl: wire up PaX softmode setting,kernel/sysctl.c. Collided with443bea2f71c5 sysctl: consolidate CONFIG_SYSCTL into a single block in kernel/sysctl.c(andacd19f2c56f2), which movedsysctl_subsys_tableandsysctl_init_bases()into the single#ifdef CONFIG_SYSCTLblock. I re-added the downstreampax_table(underCONFIG_OPENPAX_SOFTMODE) and itsregister_sysctl_init("kernel/pax", pax_table)at the new location. The added lines are identical to the original commit. Only the position moved, so the drift is just the moved hunk.use maximum supported mmap rnd entropy by default,arch/Kconfig. Collided with8cdae75cebe0 arch/Kconfig: fix dead conditions by removing dead options, which removedARCH_MMAP_RND_BITS_DEFAULT/ARCH_MMAP_RND_COMPAT_BITS_DEFAULTand theirdefault ... if ...lines. I kept upstream's removal and applied only the downstream change,default ARCH_MMAP_RND_{,COMPAT_}BITS_MIN→_MAX. No arch references the removed symbols. The commit's own +/- lines are unchanged; only the hunk context moved.9p/xen: reconnect to the backend on resume,net/9p/trans_xen.c. Collided with two upstream commits:b3d07d64832c 9p/xen: Use array for data rings:ringsis now an embeddedrings[XEN_9PFS_NUM_RINGS]array, not akzalloc'd pointer.7b6f2e3c29fb 9p/xen: drain response work after unbinding IRQ(Fixes, Cc: stable):cancel_work_sync()now runs afterunbind_from_irqhandler().What I kept:
ring_order,rings_sem,connected) next to it.xen_9pfs_front_init(), the downstreamring_order/init_rwsem()setup, without the separate allocation.xen_9pfs_front_free_rings(), upstream's unbind-then-cancel ordering.xen_9pfs_front_free(), an unconditionalxen_9pfs_front_free_rings()with nokfree(priv->rings).xen_9pfs_front_disconnect()still schedules and flushes ring work before freeing the rings, and the freeing then unbinds before cancelling, so upstream's guarantee holds on the resume path too. Later downstream commits (9p/xen: grant each ring page by its own frame,...end foreign access on the grant refs we granted...) replayed cleanly on top. In the finalfree_ringsI confirmed the ordering is unbind →cancel_work_sync→ end grants → free.9p/xen: carry the 9P session across a restore,net/9p/trans_xen.c. Collided with two upstream commits:9b4cf60c6b73 9p/xen: validate the response size from the backend(Fixes, Cc: stable), inp9_xen_response().b3d07d64832c, in the alloc/free paths, again.Upstream's
h.size < sizeof(h) || h.size > queuedcheck is placed before the downstream control-reply branch (ring == priv->rings && h.tag == P9_NOTAG && ctl_pending), so a control reply is also size-checked before it is used. A rejected control reply ends the work, and the control waiter falls back to its existingwait_for_completion_timeout(), so nothing hangs.ring == priv->ringsstill means ring 0 with the array. The alloc/free hunks were fitted to the array as in the previous commit, withINIT_LIST_HEAD(&priv->session),init_completion(&priv->ctl_done)andxen_9pfs_front_session_free()kept. The commit's own +/- lines are unchanged; only the context moved.xen/gntdev: don't copy to user space under priv->lock (#79),drivers/xen/gntdev.c. Collided with7620a997e485 xen/gntdev: fix deadlock between map ioctl and mmap. I kept the downstream variant; see the first UNSURE item above.Dropped downstream commits
None.
Build fixes
None. The branch came back from the workflow's independent build as failed, but the failure could not be reproduced and was not caused by the branch; see Builds. The branch was left unchanged at
bd59e1cfc08d.There is one latent issue I did not change, because it already exists on the old tip
7ad1695c933cand the CI configs do not reach it.drivers/hv/vmbus_drv.c:1548and:3069callhyperv_remove_xen_vmbus_irq().include/asm-generic/mshyperv.hdeclareshyperv_setup_xen_vmbus_irq()and its stub but nothyperv_remove_xen_vmbus_irq(). An arm64 build withCONFIG_HYPERV_VMBUS=ywould therefore fail with an implicit declaration. The arm64 CI config has# CONFIG_HYPERV is not set. The fix belongs in the downstream commit that added the asm-generic declarations (drivers: hv: tear down the Xen VMBus interrupt bridge on unwind).Upstream commits that meet the series
Fixes (Fixes:/stable/CVE) are marked [fix]. "Intact" means I read the final tree and the fix's checks and ordering are present, and the downstream code added nearby is covered by them.
net/9p, fs/9p, include/net/9p:
4d183e6f1532 9p/xen: fix refcount leak in p9_xen_response() on wrong tag[fix]: intact;p9_req_put()on the wrong-tag path is present.9b4cf60c6b73 9p/xen: validate the response size from the backend[fix]: intact, and it now also guards the downstream control-reply branch (conflict above).7b6f2e3c29fb 9p/xen: drain response work after unbinding IRQ[fix]: intact inxen_9pfs_front_free_rings(), which teardown, resume-disconnect and reconnect-error all use.b3d07d64832c 9p/xen: Use array for data rings: adopted. No leftoverpriv->ringsNULL checks orkfree.241d46901226 9p: reject RERROR/RLERROR with zero errcode: intact inp9_check_errors(). The downstream control-request parser intrans_xen.calready rejectsecode == 0andecode > MAX_ERRNO, so the two agree.2acd800de882 9p: fix caches_show() out-of-bounds write[fix],7fb9bd365905 9p: fix privport option setting wrong RDMA field[fix]: intact infs/9p/v9fs.c; no overlap with the downstream shared-frontend code.9eb9b583ea35 9p: remove duplicated assignment in v9fs_fill_super(),8bb0279930cc 9p: skip intermediate directory entry name copy in p9dirent_read(),33625427a382 net/9p: add vsock transport,5852cdf0006f net/9p: fix typos in comments.drivers/xen:
7620a997e485 xen/gntdev: fix deadlock between map ioctl and mmap[fix]: the deadlock fix is intact through the downstream variant (UNSURE above).2b66f2cc7532 Xen/gnttab: improve error handling in gnttab_dma_{alloc,free}_pages()[fix],4ebb3b8d129c xen: don't allow gnttab_dma_[alloc|free]_pages() in PV guest,704f812cdab8 xen/grant-table: annotate on-stack work in gnttab_unmap_refs_sync()[fix],6eecb67b69af xen/grant-table: stop setting PG_private on pages for grant mapping,0af298e62e5c treewide: refresh kmalloc_obj() conversions: all intact. The downstreamgnttab_alloc_pages_node()(whichgnttab_alloc_pages()now wraps) never sets PG_private, and nothing underdrivers/xen,arch/x86/xenorinclude/xenusesPagePrivate.Xen backends/frontends:
a02926f7c89b xen-blkback: Avoid UAF when reading I/O statistics[fix]: intact.rings_lockcovers ring alloc, disconnect and the stats reader, and the downstream per-ring kthread/IRQ node placement works only on rings published under it.bfddc420afe8 xen-netback: defer xenvif_mcast_addr_list_free()[fix]: intact. The free happens only inbackend_disconnect()aftersynchronize_net(), and the downstream per-queue kthread/IRQ placement does not reintroduce the earlier free.0af298e62e5c(xen-blkfront.c,hv/channel.c): no interaction.x86:
ea51d1a4c31f x86/CPU/AMD: Don't print reset reason in VMs[fix]: intact; makes the downstream domU check redundant (UNSURE above).cf27412f19c9 x86/bugs: Mitigate TSA on AMD Zen5: no interaction with the downstream hunk.Hyper-V:
2690d4156b51 iommu/hyperv: Add para-virtualized IOMMU support for Hyper-V guest:hv_iommu_initis hooked only inhyperv_init(), not in the downstreamhyperv_init_nested_on_xen()path, so the nested-on-Xen dom0 does not turn on the Hyper-V pvIOMMU.118bdac10546 Drivers: hv: Add logical device ID registry for vPCI devices: the registry is set up inhv_common_init(), which the nested-on-Xen init also calls. It does not overlap the downstream vPCI MSI routing.228e21764b38 Drivers: hv: Run VMBus kexec cleanup from syscore shutdown: see UNSURE above.56fb075b77f8 mshv: Redesign hypervisor memory deposit logic,dd57772e36f7 Drivers: hv: vmbus: Check ACPI companion before use,d0ec95a8a4e7 hv_netvsc: Advertise the SR-IOV capability for CoCo VMs: no interaction. Nothing in the series uses removed helpers.IOMMU:
1ae140d0f272 iommu: Make iommu-dma helpers private: the downstreamxen-iommu.ccalls no iommu-dma helpers and already includesiommu-priv.h.dma-iommu.cnow includesiommu-priv.htwice (once from upstream, once from downstreamiommu/dma: Don't reserve PCI windows under a paravirtual IOMMU). That is harmless because the header has an include guard, but a follow-up could drop the downstream line.fdd81f15fc43 iommu/dma: Clean up redundant MSI cookie check,1c8f6308e557 iommu/dma: Catch scatterlist length overflows,641a9b9cc697 dma-mapping: use exact allocation for DMA pages,95d2fec40c48 dma-mapping: rename dma_opt_mapping_size(): intact. The downstreamxen_iommu_manages_iova()check iniova_reserve_iommu_regions()sits correctly.0bf031ba8c49 iommu: Add Broadcom BCM2712 IOMMU driver,42eed9bb9ca4 iommu: apple-dart: ...(Kconfig/Makefile): context only.mm (OpenPaX):
b1ee87802111,0193e8336802,4c2dec1dce36,168ed9c5d5ec,c386d799f423,769417e410dd,13fc99f46399,8146c6478224,ef694e2d2101,68cb098eadda,7135a05838e8,aa09311c40f6.dc0b17144d62,14a5c975b138,38564710f443,6a9ef02c60bf,72a993449cb0(removesVM_SPECIAL),44debd840ec9,c911e7ebebf5(renamesis_vm_hugetlb_page),94c86564538e,88d77002e1a1,99fd6bf43d5d,61e012ab9a1c,9201d0f5b9d5,d99c4a302c5a,7fdbc26b69c0,b7b4d8f37a4f,4852f3421456,6335c786299f,9cae38dc11da,86a779c53204,f85eb0656c40,c2d4ccfbba3f,e59c9a971a9f,fe73aa910ea5,4b019e056f6c,47e725702b0b,bbe5d3b3d402,69d38c792c53.All of these are intact in the final tree. The downstream PaX flags are a separate
mm_structfield accessed with bitops and never write VMA flags directly, so the atomic-VMA-flag fixes are unaffected. The downstream MPROTECT check (map_deny_write_exec) is still called frommmap_regionbefore the mmap hooks and from mprotect.VM_SPECIAL,is_vm_hugetlb_pageandfile_doesnt_need_getappear nowhere in the tree.Other:
0f4a9e682f7f xattr: always invoke cap_convert_nscap() for security.capability writes[fix]: intact. The downstream change only addspax_getxattr.278b2abf660a init/bootconfig: pass init arguments in documented order[fix],51e766bcd35e[fix],b82ed0e3a1da[fix],958c01049e0a,aabe6e1b36ce,677f83795fdf,2e83e2ce34ff: intact. The downstreampax_softmode=is a plain__setuphandler and is unaffected.443bea2f71c5,acd19f2c56f2(conflict above);242bac52294e,d08ab285b096,2bfbb09b505c: the downstreamproc_dointvecon anintis still valid.8cdae75cebe0(conflict above);35e340a47888 mm: userland pgtable freeing is RCU-safe now, remove leftover bits: context only.38828d44f31e,799203afd298,e3e7a85730a4,f000a06da2a5,28d5df7b7274,084100a1b579,10823ced2f18,8ebd2eb87f50,681f3b68f435,537ec2b15fdd,5728d3523f12,abaa0835b4f3,264d8fd2794f) touch unrelated entries. The downstreampax_softmode=entry applied cleanly.Builds
The workflow's first independent build of
bd59e1cfc08dwas reported as failed. In the repair round I found:build-x86_64.log), and it passed: it ends withkernel-build: x86_64 vmlinux built.teefrom the first line ofkernel-build.sh, so the arm64 job must have failed before the build step ran (checkout, bundle fetch, artifact download or toolchain install), not in the kernel build. The cause is in the Actions run, not on the branch.In the repair round I rebuilt both arches from fresh object directories at
bd59e1cfc08d, with no changes to the branch:kernel-build.sh, Edera host config from edera-dev/linux-kernel-oci@530e23bbf906;olddefconfigonly warned about five unrelated=msymbols that are invalid as modules).aarch64-linux-gnu-, the same toolchain the workflow installs; defconfig plus downstream features;CONFIG_HYPERVis not set, see the latent issue under Build fixes).verify-upstream-rebase.shgives the same result as before (exit 1, with the 2 changed commits and the drift explained under Conflicts resolved). There is no new drift.