I'm working on getting the libertas_tf driver on 2.6.31 working on an
XO-1 laptopt with a USB connected device and running into the following
situation during a suspend:
Upon initiating a suspend, lbtf_op_remove_interface() gets called via
the wiphy_suspend() path:
[ 142.576332] [<cf946376>] lbtf_op_remove_interface+0x17/0x7c [libertas_tf]
[ 142.583333] [<c06c260f>] __ieee80211_suspend+0x1c3/0x200
[ 142.588776] [<c06bb9d6>] ieee80211_suspend+0x15/0x17
[ 142.593953] [<c06a3807>] wiphy_suspend+0x38/0x48
[ 142.598695] [<c0432371>] ? down+0x2b/0x2f
[ 142.602932] [<c05b693a>] dpm_suspend_start+0x214/0x32a
[ 142.608203] [<c04403a1>] suspend_devices_and_enter+0x38/0x165
[ 142.614153] [<c0440596>] enter_state+0xc8/0x114
[ 142.618800] [<c043fde4>] state_store+0x98/0xad
[ 142.623460] [<c043fd4c>] ? state_store+0x0/0xad
[ 142.628111] [<c056abd1>] kobj_attr_store+0x16/0x22
[ 142.633105] [<c04c22d7>] sysfs_write_file+0xc0/0xeb
[ 142.638107] [<c0485432>] vfs_write+0x8a/0x117
[ 142.642683] [<c04c2217>] ? sysfs_write_file+0x0/0xeb
[ 142.647771] [<c0485556>] sys_write+0x3b/0x60
[ 142.652397] [<c04033d4>] sysenter_do_call+0x12/0x26
Later on, the USB device is removed from the tree and
lbtf_op_remove_interface() gets called again and this causes
an OOOPS:
[ 143.209556] [<cf946376>] lbtf_op_remove_interface+0x17/0x7c
[libertas_tf]
[ 143.216547] [<c06b8a50>] ieee80211_stop+0x373/0x3db
[ 143.221547] [<c06cb2b6>] ? _spin_unlock_bh+0x1a/0x1c
[ 143.226736] [<c0626627>] dev_close+0x6c/0x88
[ 143.231128] [<c06266bb>] rollback_registered+0x78/0x204
[ 143.236560] [<c0626879>] unregister_netdevice+0x32/0x4d
[ 143.241901] [<c06b832c>] ieee80211_remove_interfaces+0x68/0x77
[ 143.247942] [<c06ae5fa>] ieee80211_unregister_hw+0x38/0xc3
[ 143.253564] [<cf9461a5>] lbtf_remove_card+0x31/0x3e [libertas_tf]
[ 143.259891] [<cf981966>] if_usb_disconnect+0x20/0x3e [libertas_tf_usb]
[ 143.266701] [<ce758b32>] usb_unbind_interface+0x4b/0xbe [usbcore]
[ 143.272999] [<c05b3ddf>] __device_release_driver+0x47/0x7f
[ 143.278601] [<c05b3eb4>] device_release_driver+0x18/0x23
[ 143.284236] [<ce758cf2>] usb_driver_release_interface+0x33/0x59 [usbcore]
[ 143.291347] [<ce758d83>] usb_forced_unbind_intf+0x13/0x1a [usbcore]
[ 143.297827] [<ce758dca>] do_unbind_rebind+0x40/0x5b [usbcore]
[ 143.303890] [<ce758e72>] usb_external_suspend_device+0x11/0x3e [usbcore]
[ 143.310908] [<ce758ed2>] usb_suspend+0x33/0x35 [usbcore]
[ 143.316427] [<ce7500fb>] usb_dev_suspend+0xd/0xf [usbcore]
[ 143.322130] [<c05b658a>] pm_op+0x21/0x5b
[ 143.326173] [<c05b6960>] dpm_suspend_start+0x23a/0x32a
[ 143.331534] [<c04403a1>] suspend_devices_and_enter+0x38/0x165
[ 143.337403] [<c0440596>] enter_state+0xc8/0x114
[ 143.342148] [<c043fde4>] state_store+0x98/0xad
[ 143.346710] [<c043fd4c>] ? state_store+0x0/0xad
[ 143.351455] [<c056abd1>] kobj_attr_store+0x16/0x22
[ 143.356365] [<c04c22d7>] sysfs_write_file+0xc0/0xeb
[ 143.361467] [<c0485432>] vfs_write+0x8a/0x117
[ 143.365938] [<c04c2217>] ? sysfs_write_file+0x0/0xeb
[ 143.371107] [<c0485556>] sys_write+0x3b/0x60
[ 143.375498] [<c04033d4>] sysenter_do_call+0x12/0x26
Following is the code for lbtf_op_remove_interface(). Basically we set
priv->vif to NULL the first time around and then we crash on trying
to access priv->vif->type the second time around. Easy "fix" is a NULL
check in lbtf_op_remove_interface() to simply return to caller
if !priv->vif, but I don't think the remove_interface() function
should be called twice.
static void lbtf_op_remove_interface(struct ieee80211_hw *hw,
struct ieee80211_if_init_conf *conf)
{
struct lbtf_private *priv = hw->priv;
if (priv->vif->type == NL80211_IFTYPE_AP ||
priv->vif->type == NL80211_IFTYPE_MESH_POINT)
lbtf_beacon_ctrl(priv, 0, 0);
lbtf_set_mode(priv, LBTF_PASSIVE_MODE);
lbtf_set_bssid(priv, 0, NULL);
priv->vif = NULL;
}
~Deepak
On Tue, 2010-02-09 at 12:24 -0800, Deepak Saxena wrote:
> I'm working on getting the libertas_tf driver on 2.6.31 working on an
> XO-1 laptopt with a USB connected device and running into the
> following
> situation during a suspend:
>
> Upon initiating a suspend, lbtf_op_remove_interface() gets called via
> the wiphy_suspend() path:
> Later on, the USB device is removed from the tree and
> lbtf_op_remove_interface() gets called again and this causes
> an OOOPS:
> Following is the code for lbtf_op_remove_interface(). Basically we
> set
> priv->vif to NULL the first time around and then we crash on trying
> to access priv->vif->type the second time around. Easy "fix" is a NULL
>
> check in lbtf_op_remove_interface() to simply return to caller
> if !priv->vif, but I don't think the remove_interface() function
> should be called twice.
Indeed. mac80211's removal logic should cope with interfaces that are
suspended.
johannes