{"id":"CVE-2026-89913","published":"2026-09-16T11:17:00.333","lastModified":"2026-09-17T10:17:04.810","description":"In the Linux kernel, the following vulnerability has been resolved:\n\nKVM: arm64: vgic-v3: take an LPI reference in vgic_v3_save_pending_tables\n\nvgic_v3_save_pending_tables() iterates dist->lpi_xa using xa_for_each()\nand dereferences the returned struct vgic_irq in the loop body without\nholding a reference on the LPI.\n\nThe xarray iterator only provides temporary RCU coverage while looking up\nthe current entry. That is not sufficient for this loop body, which reads\nfields from struct vgic_irq and performs guest memory accesses before the\niteration completes.\n\nA concurrent path can trigger this race: the irqfd cached injection path\n(vgic_its_inject_cached_translation) obtains a transient LPI reference\nvia vgic_its_check_cache() without holding kvm->lock, vcpu->mutex,\nconfig_lock, or its_lock. If guest ITS DISCARD then drops the cache and\nITE references under its_lock, the transient inject reference may become\nthe final one. When vgic_put_irq() drops it, the LPI is erased from\nlpi_xa and freed via kfree_rcu(). Meanwhile, vgic_v3_save_pending_tables()\nmay still hold a stale pointer obtained from the xarray iterator and\ndereference it after the RCU grace period completes.\n\nFix this by re-fetching each iterated LPI via vgic_get_irq(), which takes\na stable reference, and dropping it with vgic_put_irq() on all paths.\nThis matches the pattern already used by other lpi_xa iterators in the\nvgic ITS code.","cvssScore":8.8,"cvssSeverity":"HIGH","cvssVector":"CVSS:3.1/AV:L/AC:L/PR:L/UI:N/S:C/C:H/I:H/A:H","cwes":[],"vendors":[],"products":[],"references":[{"url":"https://git.kernel.org/stable/c/7631f95297560157d3a9283cb999e3be00103348","tags":[]},{"url":"https://git.kernel.org/stable/c/d3a2d20b7248ea67465af1cf79c79f90ff70926c","tags":[]},{"url":"https://git.kernel.org/stable/c/f5b8f203bfc07a5a257dff859e66d2c500f9f509","tags":[]}],"exploitRefs":[],"hasPoc":false,"ai":null}