[PATCH] libertas: clear current command on card removal

[PATCH] libertas: clear current command on card removal

[Dan Williams]

If certain commands were in-flight when the card was pulled or the
driver rmmod-ed, cleanup would block on the work queue stopping, but the
work queue was in turn blocked on the current command being canceled,
which didn't happen.  Fix that.

Signed-off-by: Dan Williams <dcbw@redhat.com>
---
John, this is 2.6.27 material if there's still time.

diff --git a/drivers/net/wireless/libertas/main.c b/drivers/net/wireless/libertas/main.c
index 2436634..73dc8c7 100644
--- a/drivers/net/wireless/libertas/main.c
+++ b/drivers/net/wireless/libertas/main.c
@@ -1205,7 +1205,13 @@ void lbs_remove_card(struct lbs_private *priv)
 	cancel_delayed_work_sync(&priv->scan_work);
 	cancel_delayed_work_sync(&priv->assoc_work);
 	cancel_work_sync(&priv->mcast_work);
+
+	/* worker thread destruction blocks on the in-flight command which
+	 * should have been cleared already in lbs_stop_card().
+	 */
+	lbs_deb_main("destroying worker thread\n");
 	destroy_workqueue(priv->work_thread);
+	lbs_deb_main("done destroying worker thread\n");
 
 	if (priv->psmode == LBS802_11POWERMODEMAX_PSP) {
 		priv->psmode = LBS802_11POWERMODECAM;
@@ -1323,14 +1329,26 @@ void lbs_stop_card(struct lbs_private *priv)
 		device_remove_file(&dev->dev, &dev_attr_lbs_rtap);
 	}
 
-	/* Flush pending command nodes */
+	/* Delete the timeout of the currently processing command */
 	del_timer_sync(&priv->command_timer);
+
+	/* Flush pending command nodes */
 	spin_lock_irqsave(&priv->driver_lock, flags);
+	lbs_deb_main("clearing pending commands\n");
 	list_for_each_entry(cmdnode, &priv->cmdpendingq, list) {
 		cmdnode->result = -ENOENT;
 		cmdnode->cmdwaitqwoken = 1;
 		wake_up_interruptible(&cmdnode->cmdwait_q);
 	}
+
+	/* Flush the command the card is currently processing */
+	if (priv->cur_cmd) {
+		lbs_deb_main("clearing current command\n");
+		priv->cur_cmd->result = -ENOENT;
+		priv->cur_cmd->cmdwaitqwoken = 1;
+		wake_up_interruptible(&priv->cur_cmd->cmdwait_q);
+	}
+	lbs_deb_main("done clearing commands\n");
 	spin_unlock_irqrestore(&priv->driver_lock, flags);
 
 	unregister_netdev(dev);

Re: [PATCH] libertas: clear current command on card removal

[John W. Linville]

On Mon, Sep 08, 2008 at 04:34:40PM -0400, Dan Williams wrote:
> If certain commands were in-flight when the card was pulled or the
> driver rmmod-ed, cleanup would block on the work queue stopping, but the
> work queue was in turn blocked on the current command being canceled,
> which didn't happen.  Fix that.
> 
> Signed-off-by: Dan Williams <dcbw@redhat.com>
> ---
> John, this is 2.6.27 material if there's still time.

I have to remind you of the new "regressions only" policy.  Is this a
problem that we introduced in old code during the 2.6.27 merge window?

John
-- 
John W. Linville
linville@tuxdriver.com

Re: [PATCH] libertas: clear current command on card removal

[Dan Williams]

On Mon, 2008-09-08 at 17:01 -0400, John W. Linville wrote:
> On Mon, Sep 08, 2008 at 04:34:40PM -0400, Dan Williams wrote:
> > If certain commands were in-flight when the card was pulled or the
> > driver rmmod-ed, cleanup would block on the work queue stopping, but the
> > work queue was in turn blocked on the current command being canceled,
> > which didn't happen.  Fix that.
> > 
> > Signed-off-by: Dan Williams <dcbw@redhat.com>
> > ---
> > John, this is 2.6.27 material if there's still time.
> 
> I have to remind you of the new "regressions only" policy.  Is this a
> problem that we introduced in old code during the 2.6.27 merge window?

True :)  It is not a regression from 2.6.26, but a fairly serious bug
that has existed for some time.  Thus, treat it as you like to avoid
getting flamed by Linus & Dave.

I'll post to stable after 2.6.27 comes out then.

Dan

Re: [stable] [PATCH] libertas: clear current command on card removal

[Greg KH]

On Mon, Sep 08, 2008 at 05:39:17PM -0400, Dan Williams wrote:
> On Mon, 2008-09-08 at 17:01 -0400, John W. Linville wrote:
> > On Mon, Sep 08, 2008 at 04:34:40PM -0400, Dan Williams wrote:
> > > If certain commands were in-flight when the card was pulled or the
> > > driver rmmod-ed, cleanup would block on the work queue stopping, but the
> > > work queue was in turn blocked on the current command being canceled,
> > > which didn't happen.  Fix that.
> > > 
> > > Signed-off-by: Dan Williams <dcbw@redhat.com>
> > > ---
> > > John, this is 2.6.27 material if there's still time.
> > 
> > I have to remind you of the new "regressions only" policy.  Is this a
> > problem that we introduced in old code during the 2.6.27 merge window?
> 
> True :)  It is not a regression from 2.6.26, but a fairly serious bug
> that has existed for some time.  Thus, treat it as you like to avoid
> getting flamed by Linus & Dave.
> 
> I'll post to stable after 2.6.27 comes out then.

No, post to stable when the patch goes into Linus's tree as that is a
requirement.

To do so automatically, just add:
	CC: stable <stable@kernel.org>
in the signed-off-by: area of the patch and we will be automatically
notified when it goes into Linus's tree and will pick it up then, no
extra work for you at all (and much easier for the stable-team as well.)

thanks,

greg k-h