perf_event_comm(tsk, exec);
 }
 
+/*
+ * Calling this is the point of no return. None of the failures will be
+ * seen by userspace since either the process is already taking a fatal
+ * signal (via de_thread() or coredump), or will have SEGV raised
+ * (after exec_mmap()) by search_binary_handlers (see below).
+ */
 int flush_old_exec(struct linux_binprm * bprm)
 {
        int retval;
        if (retval)
                goto out;
 
-       bprm->mm = NULL;                /* We're using it now */
+       /*
+        * After clearing bprm->mm (to mark that current is using the
+        * prepared mm now), we have nothing left of the original
+        * process. If anything from here on returns an error, the check
+        * in search_binary_handler() will SEGV current.
+        */
+       bprm->mm = NULL;
 
        set_fs(USER_DS);
        current->flags &= ~(PF_RANDOMIZE | PF_FORKNOEXEC | PF_KTHREAD |
 {
        arch_pick_mmap_layout(current->mm);
 
-       /* This is the point of no return */
        current->sas_ss_sp = current->sas_ss_size = 0;
 
        if (uid_eq(current_euid(), current_uid()) && gid_eq(current_egid(), current_gid()))
         */
        current->mm->task_size = TASK_SIZE;
 
-       /* install the new credentials */
        if (!uid_eq(bprm->cred->uid, current_euid()) ||
            !gid_eq(bprm->cred->gid, current_egid())) {
                current->pdeath_signal = 0;