CVE-2025-68171

In the Linux kernel, the following vulnerability has been resolved:<br /> <br /> x86/fpu: Ensure XFD state on signal delivery<br /> <br /> Sean reported [1] the following splat when running KVM tests:<br /> <br /> WARNING: CPU: 232 PID: 15391 at xfd_validate_state+0x65/0x70<br /> Call Trace:<br /> <br /> fpu__clear_user_states+0x9c/0x100<br /> arch_do_signal_or_restart+0x142/0x210<br /> exit_to_user_mode_loop+0x55/0x100<br /> do_syscall_64+0x205/0x2c0<br /> entry_SYSCALL_64_after_hwframe+0x4b/0x53<br /> <br /> Chao further identified [2] a reproducible scenario involving signal<br /> delivery: a non-AMX task is preempted by an AMX-enabled task which<br /> modifies the XFD MSR.<br /> <br /> When the non-AMX task resumes and reloads XSTATE with init values,<br /> a warning is triggered due to a mismatch between fpstate::xfd and the<br /> CPU&amp;#39;s current XFD state. fpu__clear_user_states() does not currently<br /> re-synchronize the XFD state after such preemption.<br /> <br /> Invoke xfd_update_state() which detects and corrects the mismatch if<br /> there is a dynamic feature.<br /> <br /> This also benefits the sigreturn path, as fpu__restore_sig() may call<br /> fpu__clear_user_states() when the sigframe is inaccessible.<br /> <br /> [ dhansen: minor changelog munging ]