"value":"In the Linux kernel, the following vulnerability has been resolved:\n\nx86/fpu: Drop fpregs lock before inheriting FPU permissions\n\nMike Galbraith reported the following against an old fork of preempt-rt\nbut the same issue also applies to the current preempt-rt tree.\n\n BUG: sleeping function called from invalid context at kernel/locking/spinlock_rt.c:46\n in_atomic(): 1, irqs_disabled(): 0, non_block: 0, pid: 1, name: systemd\n preempt_count: 1, expected: 0\n RCU nest depth: 0, expected: 0\n Preemption disabled at:\n fpu_clone\n CPU: 6 PID: 1 Comm: systemd Tainted: G E (unreleased)\n Call Trace:\n <TASK>\n dump_stack_lvl\n ? fpu_clone\n __might_resched\n rt_spin_lock\n fpu_clone\n ? copy_thread\n ? copy_process\n ? shmem_alloc_inode\n ? kmem_cache_alloc\n ? kernel_clone\n ? __do_sys_clone\n ? do_syscall_64\n ? __x64_sys_rt_sigprocmask\n ? syscall_exit_to_user_mode\n ? do_syscall_64\n ? syscall_exit_to_user_mode\n ? do_syscall_64\n ? syscall_exit_to_user_mode\n ? do_syscall_64\n ? exc_page_fault\n ? entry_SYSCALL_64_after_hwframe\n </TASK>\n\nMike says:\n\n The splat comes from fpu_inherit_perms() being called under fpregs_lock(),\n and us reaching the spin_lock_irq() therein due to fpu_state_size_dynamic()\n returning true despite static key __fpu_state_size_dynamic having never\n been enabled.\n\nMike's assessment looks correct. fpregs_lock on a PREEMPT_RT kernel disables\npreemption so calling spin_lock_irq() in fpu_inherit_perms() is unsafe. This\nproblem exists since commit\n\n 9e798e9aa14c (\"x86/fpu: Prepare fpu_clone() for dynamically enabled features\").\n\nEven though the original bug report should not have enabled the paths at\nall, the bug still exists.\n\nfpregs_lock is necessary when editing the FPU registers or a task's FP\nstate but it is not necessary for fpu_inherit_perms(). The only write\nof any FP state in fpu_inherit_perms() is for the new child which is\nnot running yet and cannot context switch or be borrowed by a kernel\nthread yet. Hence, fpregs_lock is not protecting anything in the new\nchild until clone() completes and can be dropped earlier. The siglock\nstill needs to be acquired by fpu_inherit_perms() as the read of the\nparent's permissions has to be serialised.\n\n [ bp: Cleanup splat. ]"