On Tue, 7 Feb 2023 19:21:29 +0100 Florent Revest <revest@xxxxxxxxxxxx> wrote: > @@ -5445,6 +5445,7 @@ __modify_ftrace_direct(struct ftrace_ops *ops, unsigned long addr) > /* Enable the tmp_ops to have the same functions as the direct ops */ > ftrace_ops_init(&tmp_ops); > tmp_ops.func_hash = ops->func_hash; > + tmp_ops.direct_call = addr; > > err = register_ftrace_function_nolock(&tmp_ops); > if (err) > @@ -5466,6 +5467,7 @@ __modify_ftrace_direct(struct ftrace_ops *ops, unsigned long addr) > entry->direct = addr; > } > } > + WRITE_ONCE(ops->direct_call, addr); I'm curious about the use of WRITE_ONCE(). It should not go outside the mutex barrier. -- Steve > > mutex_unlock(&ftrace_lock); >
