{ "schema_version": "1.6.1", "id": "RPI-CVE-2025-68171", "modified": "2026-10-09T11:39:44Z", "upstream": [ "CVE-2025-68171" ], "published": "2025-12-16T13:43:51Z", "summary": "x86/fpu: Ensure XFD state on signal delivery", "details": "In the Linux kernel, the following vulnerability has been resolved:\n\nx86/fpu: Ensure XFD state on signal delivery\n\nSean reported [1] the following splat when running KVM tests:\n\n WARNING: CPU: 232 PID: 15391 at xfd_validate_state+0x65/0x70\n Call Trace:\n \n fpu__clear_user_states+0x9c/0x100\n arch_do_signal_or_restart+0x142/0x210\n exit_to_user_mode_loop+0x55/0x100\n do_syscall_64+0x205/0x2c0\n entry_SYSCALL_64_after_hwframe+0x4b/0x53\n\nChao further identified [2] a reproducible scenario involving signal\ndelivery: a non-AMX task is preempted by an AMX-enabled task which\nmodifies the XFD MSR.\n\nWhen the non-AMX task resumes and reloads XSTATE with init values,\na warning is triggered due to a mismatch between fpstate::xfd and the\nCPU's current XFD state. fpu__clear_user_states() does not currently\nre-synchronize the XFD state after such preemption.\n\nInvoke xfd_update_state() which detects and corrects the mismatch if\nthere is a dynamic feature.\n\nThis also benefits the sigreturn path, as fpu__restore_sig() may call\nfpu__clear_user_states() when the sigframe is inaccessible.\n\n[ dhansen: minor changelog munging ]", "severity": [ { "type": "CVSS_V3", "score": "CVSS:3.1/AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H" } ], "affected": [ { "package": { "ecosystem": "Raspberry Pi OS:12", "name": "linux" }, "versions": [ "1:6.1.47-1+rpt4" ], "ecosystem_specific": { "binaries": [ { "binary_name": "linux-image-6.1.0-rpi3-rpi-2712", "binary_version": "1:6.1.47-1+rpt4" }, { "binary_name": "linux-image-6.1.0-rpi3-rpi-v6", "binary_version": "1:6.1.47-1+rpt4" }, { "binary_name": "linux-image-6.1.0-rpi3-rpi-v7", "binary_version": "1:6.1.47-1+rpt4" }, { "binary_name": "linux-image-6.1.0-rpi3-rpi-v7l", "binary_version": "1:6.1.47-1+rpt4" }, { "binary_name": "linux-image-6.1.0-rpi3-rpi-v8", "binary_version": "1:6.1.47-1+rpt4" } ] }, "database_specific": { "per_version": { "1:6.1.47-1+rpt4": { "linux_commit": "655fc658a15ae7a6f37103754adb39ba52a9a14e", "upstream_version": "6.1.47", "assessment": "commit ancestry (strak rule)", "upstream_fixed": "6.1.159", "compiled_filter": false } } } } ], "references": [ { "type": "ADVISORY", "url": "https://www.cve.org/CVERecord?id=CVE-2025-68171" }, { "type": "FIX", "url": "https://git.kernel.org/stable/c/eefbfb722042fc9210d2e0ac2b063fd1abf51895" }, { "type": "FIX", "url": "https://git.kernel.org/stable/c/1811c610653c0cd21cc9add14595b7cffaeca511" }, { "type": "FIX", "url": "https://git.kernel.org/stable/c/5b2619b488f1d08b960c43c6468dd0759e8b3035" }, { "type": "FIX", "url": "https://git.kernel.org/stable/c/3f735419c4b43cde42e6d408db39137b82474e31" }, { "type": "FIX", "url": "https://git.kernel.org/stable/c/388eff894d6bc5f921e9bfff0e4b0ab2684a96e9" } ], "database_specific": { "source": "https://git.kernel.org/pub/scm/linux/security/vulns.git", "upstream_modified": "2026-08-05T12:19:11Z", "severity": { "source": "kernel CNA", "cvss_v3_score": 7.8, "cvss_v3_severity": "High", "scenario": "AV:L - The vulnerable code is reached only through local task-context paths \u2014 signal delivery from `arch_do_signal_or_restart()` and the `rt_sigreturn` syscall \u2014 requiring the attacker to execute code on the machine. No network or adjacent-network reachability exists.\nAC:L - The attacker controls both sides of the race: it runs its own AMX-enabled and non-AMX threads, pins them to a chosen CPU, floods signals, and can force a scheduling point in the exact window between `copy_fpstate_to_sigframe()` and `fpu__clear_user_states()` by faulting on userfaultfd-backed sigframe pages. AMX/XFD-capable Xeons (Sapphire Rapids and later) are mainstream cloud/datacenter hardware, so the required configuration is a normal deployment rather than a rare precondition.\nPR:L - Any unprivileged local user can deliver signals to its own threads, call `arch_prctl(ARCH_REQ_XCOMP_PERM, XFEATURE_XTILEDATA)`, set CPU affinity, and invoke `rt_sigreturn` with an inaccessible sigframe \u2014 there is no capability, namespace, or LSM check anywhere on the path.\nUI:N - The attacker triggers the entire sequence from its own processes via signals and scheduling; no action by any other user is required.\nS:U - The XFD desync and its consequences stay within the kernel's own security authority on the host \u2014 there is no guest-to-host escape, IOMMU bypass, or other crossing of a security-authority boundary.\nC:H - The stale XFD causes the XRSTOR that is specifically responsible for scrubbing dynamic-feature state left by the previous task to skip XTILEDATA, and `fpregs_mark_activate()` suppresses the later corrective restore, so up to 8 KiB of another process's AMX tile register content survives into the attacker's context and can be dumped with a userspace `XSAVE` after re-enabling XFD via #NM; the debug WARN additionally discloses kernel addresses.\nI:H - The same primitive works in reverse \u2014 attacker-chosen tile register content is left un-scrubbed and becomes the operating state of a victim AMX process, silently corrupting its computation \u2014 and the desync also lets a task use AMX with no `ARCH_REQ_XCOMP_PERM` grant while its fpstate has no XTILEDATA storage, bypassing the dynamic-xstate permission model and silently discarding FPU state across context switches.\nA:H - The mismatch trips `WARN_ON_ONCE()` in `xfd_validate_state()` from both `os_xrstor()` and the subsequent `os_xsave()` (the reported splat), which is a fatal panic on systems booted with `panic_on_warn=1`, and leaves the FPU state machine persistently inconsistent with the hardware." } } }