vcs: hook sysfs devices into object lifetime instead of "binding"
During bootup performance tracing I noticed many occurrences of
vca* device creation and removal, leading to the usual userspace
uevent processing, which are, in this case, rather pointless.
A simple test showing the kernel timing (not including all the
work userspace has to do), gives us these numbers:
$ time for i in `seq 1000`; do echo a > /dev/tty2; done
real 0m1.142s
user 0m0.015s
sys 0m0.540s
If we move the hook for the vcs* driver core devices from the
tty "binding" to the vc allocation/deallocation, which is what
the vcs* devices represent, we get the following numbers:
$ time for i in `seq 1000`; do echo a > /dev/tty2; done
real 0m0.152s
user 0m0.030s
sys 0m0.072s
Cc: Alan Cox <alan@lxorguk.ukuu.org.uk>
Signed-off-by: Kay Sievers <kay.sievers@vrfy.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
diff --git a/drivers/char/vt.c b/drivers/char/vt.c
index 7900bd6..2c1d133 100644
--- a/drivers/char/vt.c
+++ b/drivers/char/vt.c
@@ -778,6 +778,7 @@
}
vc->vc_kmalloced = 1;
vc_init(vc, vc->vc_rows, vc->vc_cols, 1);
+ vcs_make_sysfs(currcons);
atomic_notifier_call_chain(&vt_notifier_list, VT_ALLOCATE, ¶m);
}
return 0;
@@ -987,7 +988,9 @@
if (vc_cons_allocated(currcons)) {
struct vc_data *vc = vc_cons[currcons].d;
struct vt_notifier_param param = { .vc = vc };
+
atomic_notifier_call_chain(&vt_notifier_list, VT_DEALLOCATE, ¶m);
+ vcs_remove_sysfs(currcons);
vc->vc_sw->con_deinit(vc);
put_pid(vc->vt_pid);
module_put(vc->vc_sw->owner);
@@ -2775,7 +2778,6 @@
tty->termios->c_iflag |= IUTF8;
else
tty->termios->c_iflag &= ~IUTF8;
- vcs_make_sysfs(tty);
release_console_sem();
return ret;
}
@@ -2795,7 +2797,6 @@
BUG_ON(vc == NULL);
acquire_console_sem();
vc->vc_tty = NULL;
- vcs_remove_sysfs(tty);
release_console_sem();
tty_shutdown(tty);
}