There is no any reason to do recalc_sigpending() after changing ->sighand.
To begin with, recalc_sigpending() does not take ->sighand into account.
This means we don't need to take newsighand->siglock while changing sighands.
rcu_assign_pointer() provides a necessary barrier, and if another process
reads the new ->sighand it should either take tasklist_lock or it should use
lock_task_sighand() which has a corresponding smp_read_barrier_depends().
Signed-off-by: Oleg Nesterov <[email protected]>
--- t/fs/exec.c~2_SWITCH 2007-08-18 18:19:41.000000000 +0400
+++ t/fs/exec.c 2007-08-18 18:26:48.000000000 +0400
@@ -945,12 +945,7 @@ no_thread_group:
write_lock_irq(&tasklist_lock);
spin_lock(&oldsighand->siglock);
- spin_lock_nested(&newsighand->siglock, SINGLE_DEPTH_NESTING);
-
rcu_assign_pointer(tsk->sighand, newsighand);
- recalc_sigpending();
-
- spin_unlock(&newsighand->siglock);
spin_unlock(&oldsighand->siglock);
write_unlock_irq(&tasklist_lock);
@@ -960,12 +955,11 @@ no_thread_group:
BUG_ON(!thread_group_leader(tsk));
return 0;
}
-
+
/*
* These functions flushes out all traces of the currently running executable
* so that a new one can be started
*/
-
static void flush_old_files(struct files_struct * files)
{
long j = -1;
> There is no any reason to do recalc_sigpending() after changing ->sighand.
> To begin with, recalc_sigpending() does not take ->sighand into account.
I agree. I think that call dates from before some other cleanups in that
code, when ->signal was changed there. At the time, it was the most
conservative change to leave recalc_sigpending in case the side effects it
used to have were desireable (we've since decided to change those anyway).
Thanks,
Roland