orinoco: take the driver lock in the rx tasklet
authorDavid Kilroy <kilroyd@googlemail.com>
Wed, 7 Jan 2009 00:23:55 +0000 (00:23 +0000)
committerJohn W. Linville <linville@tuxdriver.com>
Mon, 12 Jan 2009 19:24:51 +0000 (14:24 -0500)
Fix the warning reproduced below.

We add to rx_list in interrupt context and remove elements in tasklet
context. While removing elements we need to prevent the interrupt
modifying the list.

Note that "orinoco: Process bulk of receive interrupt in a tasklet" did not
preserve locking semantics on what is now orinoco_rx.

This patch reinstates the locking semantics and ensures it covers
rx_list as well. This leads to additional cleanup required in
free_orinocodev.

[89479.105038] WARNING: at lib/list_debug.c:30 __list_add+0x8f/0xa0()
[89479.105058] list_add corruption. prev->next should be next (dddb3568), but was cbc28978. (prev=dddb3568).
[89479.106002] Pid: 15746, comm: X Not tainted 2.6.28-1avb #26
[89479.106020] Call Trace:
[89479.106062]  [<c011d3b0>] warn_slowpath+0x60/0x80
[89479.106104]  [<c01073d0>] ? native_sched_clock+0x20/0x70
[89479.106194]  [<c013d825>] ? lock_release_holdtime+0x35/0x200
[89479.106218]  [<c018d9f0>] ? __slab_alloc+0x550/0x560
[89479.106254]  [<c02f9c9d>] ? _spin_unlock+0x1d/0x20
[89479.106270]  [<c018d9f0>] ? __slab_alloc+0x550/0x560
[89479.106302]  [<c01ff2a7>] ? delay_tsc+0x17/0x24
[89479.106319]  [<c01ff221>] ? __const_udelay+0x21/0x30
[89479.106376]  [<dfa8b1e2>] ? hermes_bap_seek+0x112/0x1e0 [hermes]
[89479.106396]  [<c013d7eb>] ? trace_hardirqs_off+0xb/0x10
[89479.106418]  [<c018e307>] ? __kmalloc_track_caller+0xb7/0x110
[89479.106448]  [<c028eefc>] ? dev_alloc_skb+0x1c/0x30
[89479.106465]  [<c028eefc>] ? dev_alloc_skb+0x1c/0x30
[89479.106482]  [<c020e13f>] __list_add+0x8f/0xa0
[89479.106551]  [<dfd0fcae>] orinoco_interrupt+0xcae/0x16c0 [orinoco]
[89479.106574]  [<c013b0e3>] ? tick_dev_program_event+0x33/0xb0
[89479.106594]  [<c01073d0>] ? native_sched_clock+0x20/0x70
[89479.106613]  [<c013d825>] ? lock_release_holdtime+0x35/0x200
[89479.106662]  [<c013d7eb>] ? trace_hardirqs_off+0xb/0x10
[89479.106892]  [<dfe7faa7>] ? usb_hcd_irq+0x97/0xa0 [usbcore]
[89479.106926]  [<c015ba79>] handle_IRQ_event+0x29/0x60
[89479.106947]  [<c015cf89>] handle_level_irq+0x69/0xe0
[89479.106963]  [<c015cf20>] ? handle_level_irq+0x0/0xe0
[89479.106977]  <IRQ>  [<c02ca933>] ? tcp_v4_rcv+0x633/0x6e0
[89479.107025]  [<c0103f0c>] ? common_interrupt+0x28/0x30
[89479.107057]  [<c02a0000>] ? sk_run_filter+0x320/0x7a0
[89479.107078]  [<c020e041>] ? list_del+0x21/0x90
[89479.107106]  [<dfd0d24e>] ? orinoco_rx_isr_tasklet+0x2ce/0x480 [orinoco]
[89479.107131]  [<c01402e0>] ? __lock_acquire+0x160/0x1650
[89479.107151]  [<c01073d0>] ? native_sched_clock+0x20/0x70
[89479.107169]  [<c013d825>] ? lock_release_holdtime+0x35/0x200
[89479.107200]  [<c012249a>] ? irq_enter+0xa/0x60
[89479.107217]  [<c0104e52>] ? do_IRQ+0xd2/0x130
[89479.107518]  [<c010342c>] ? restore_nocheck_notrace+0x0/0xe
[89479.107542]  [<c0122830>] ? __do_softirq+0x0/0x110
[89479.107561]  [<c013f7b4>] ? trace_hardirqs_on_caller+0x74/0x140
[89479.107583]  [<c01ff678>] ? trace_hardirqs_on_thunk+0xc/0x10
[89479.107602]  [<c0122087>] ? tasklet_action+0x27/0x90
[89479.107620]  [<c013f7b4>] ? trace_hardirqs_on_caller+0x74/0x140
[89479.107638]  [<c01220a3>] ? tasklet_action+0x43/0x90
[89479.107655]  [<c012289f>] ? __do_softirq+0x6f/0x110
[89479.107674]  [<c0122830>] ? __do_softirq+0x0/0x110
[89479.107685]  <IRQ>  [<c015cf20>] ? handle_level_irq+0x0/0xe0
[89479.107715]  [<c012246d>] ? irq_exit+0x5d/0x80
[89479.107732]  [<c0104e52>] ? do_IRQ+0xd2/0x130
[89479.107747]  [<c0103337>] ? sysenter_exit+0xf/0x16
[89479.107765]  [<c013f83d>] ? trace_hardirqs_on_caller+0xfd/0x140
[89479.107782]  [<c0103f0c>] ? common_interrupt+0x28/0x30
[89479.107797] ---[ end trace a1fc0a52df4a729d ]---

Reported-by: Andrey Borzenkov <arvidjaar@mail.ru>
Signed-off-by: David Kilroy <kilroyd@googlemail.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
drivers/net/wireless/orinoco/orinoco.c

index bc84e27..c3bb85e 100644 (file)
@@ -1610,6 +1610,16 @@ static void orinoco_rx_isr_tasklet(unsigned long data)
        struct orinoco_rx_data *rx_data, *temp;
        struct hermes_rx_descriptor *desc;
        struct sk_buff *skb;
+       unsigned long flags;
+
+       /* orinoco_rx requires the driver lock, and we also need to
+        * protect priv->rx_list, so just hold the lock over the
+        * lot.
+        *
+        * If orinoco_lock fails, we've unplugged the card. In this
+        * case just abort. */
+       if (orinoco_lock(priv, &flags) != 0)
+               return;
 
        /* extract desc and skb from queue */
        list_for_each_entry_safe(rx_data, temp, &priv->rx_list, list) {
@@ -1622,6 +1632,8 @@ static void orinoco_rx_isr_tasklet(unsigned long data)
 
                kfree(desc);
        }
+
+       orinoco_unlock(priv, &flags);
 }
 
 /********************************************************************/
@@ -3645,12 +3657,22 @@ struct net_device
 void free_orinocodev(struct net_device *dev)
 {
        struct orinoco_private *priv = netdev_priv(dev);
+       struct orinoco_rx_data *rx_data, *temp;
 
-       /* No need to empty priv->rx_list: if the tasklet is scheduled
-        * when we call tasklet_kill it will run one final time,
-        * emptying the list */
+       /* If the tasklet is scheduled when we call tasklet_kill it
+        * will run one final time. However the tasklet will only
+        * drain priv->rx_list if the hw is still available. */
        tasklet_kill(&priv->rx_tasklet);
 
+       /* Explicitly drain priv->rx_list */
+       list_for_each_entry_safe(rx_data, temp, &priv->rx_list, list) {
+               list_del(&rx_data->list);
+
+               dev_kfree_skb(rx_data->skb);
+               kfree(rx_data->desc);
+               kfree(rx_data);
+       }
+
        unregister_pm_notifier(&priv->pm_notifier);
        orinoco_uncache_fw(priv);