n_gsm.c : Use of asynchronous hangup

BZ: 19302

In Linux 3.0 a new "big tty mutex" have been implemented.
This prevent tty_vhangup to be nested checking task ID (in tty_lock).
We need to launch the hangup in another workqueue which has a different task ID.
This is done by an asynchronous hangup.

flush_work is not necessary as this is done by tty layer before releasing the tty.

Change-Id: Ifdb1a4a942861e84ab948e5c978d9e493d70297a
Signed-off-by: Vincent Pillet <vincentx.pillet@intel.com>
Reviewed-on: http://android.intel.com:8080/32192
Reviewed-by: Stoltz-douchet, OlivierX <olivierx.stoltz-douchet@intel.com>
Reviewed-by: Guerard, Christophe <christophe.guerard@intel.com>
Tested-by: Guerard, Christophe <christophe.guerard@intel.com>
Reviewed-by: Predon, Frederic <frederic.predon@intel.com>
Reviewed-by: Lebsir, SamiX <samix.lebsir@intel.com>
Tested-by: Lebsir, SamiX <samix.lebsir@intel.com>
Reviewed-by: buildbot <buildbot@intel.com>
Tested-by: buildbot <buildbot@intel.com>

diff --git a/drivers/tty/n_gsm.c b/drivers/tty/n_gsm.c
index fe85e32..55c9db7 100644 (file)
--- a/drivers/tty/n_gsm.c
+++ b/drivers/tty/n_gsm.c
@@ -78,9 +78,6 @@ module_param(debug, int, 0600);
 #define MAX_MTU 32768 /* In specification 3GPP TS 27.010, 5.7.2 */
 #define        GSM_NET_TX_TIMEOUT (HZ*10)
 
-/* Workarounds */
-#undef  WA_VHANGUP_CLOCK
-
 /**
  *     struct gsm_mux_net      -       network interface
  *     @struct gsm_dlci* dlci
@@ -1700,28 +1697,14 @@ static void gsm_dlci_release(struct gsm_dlci *dlci)
 {
        struct tty_struct *tty = tty_port_tty_get(&dlci->port);
        if (tty) {
-#ifdef WA_VHANGUP_CLOCK
                /*
-                * Calling tty_vhangup from here
-                * can tickle a circular lock problem  on some
-                * versions of the kernel where TIOCSETD calls
-                * tty_set_ldisc and takes tty->ldisc_mutex, and while
-                * holding that mutex removes the old ldisc (this code)
-                * then applies the new ldisc so this code is called
-                * with the mutex held. tty_vhangup synchronously calls
-                * tty_ldisc_hangup which takes the mutex again.
-                * From perusing the code, newer versions of the kernel
-                * do not appear to have this issue but I dion't know what
-                * version ift was fixed in.
-                *
-                * we'll workaround this by just calling tty_hangup
-                * and waiting a bit
+                * In Linux 3.0 a new "big tty mutex" have been implemented.
+                * This prevent tty_vhangup to be nested checking task ID
+                * (in tty_lock). We need to launch the hangup in another
+                * workqueue which has a different task ID.
+                * This is done by an asynchronous hangup.
                 */
                tty_hangup(tty);
-               msleep(250);
-#else /* WA_VHANGUP_CLOCK */
-               tty_vhangup(tty);
-#endif /* WA_VHANGUP_CLOCK */
                tty_kref_put(tty);
        }
        kref_put(&dlci->ref, gsm_dlci_free);