Re: [RFC/T V2] b43: Fix Radio On/Off LED action

[Michael Buesch <mb@bu3sch.de>]

On Wednesday 28 November 2007 16:05:35 Larry Finger wrote:
> Michael Buesch wrote:
> > 
> > So it's a lock dependency between rfkill->mutex and wl->mutex?
> > So, now comes the question that really matters. Who is the caller
> > of rfkill_toggle_radio, in the case where it crashes?
> >
> 
> Here is the full dump. It looks to me as if b43_rfkill_soft_toggle() calls rfkill_handler(), which
> calls rfkill_toggle_radio() and it is calling b43_rfkill_soft_toggle(). I'd call that a loop.

I think it's a different bug. The backtrace seems corrupted.

Can you try this patch? There is some circular locking in rfkill.

Index: wireless-2.6/net/rfkill/rfkill.c
===================================================================
--- wireless-2.6.orig/net/rfkill/rfkill.c	2007-11-20 19:09:35.000000000 +0100
+++ wireless-2.6/net/rfkill/rfkill.c	2007-11-28 17:09:55.000000000 +0100
@@ -60,11 +60,7 @@ static void rfkill_led_trigger(struct rf
 static int rfkill_toggle_radio(struct rfkill *rfkill,
 				enum rfkill_state state)
 {
-	int retval;
-
-	retval = mutex_lock_interruptible(&rfkill->mutex);
-	if (retval)
-		return retval;
+	int retval = 0;
 
 	if (state != rfkill->state) {
 		retval = rfkill->toggle_radio(rfkill->data, state);
@@ -74,7 +70,6 @@ static int rfkill_toggle_radio(struct rf
 		}
 	}
 
-	mutex_unlock(&rfkill->mutex);
 	return retval;
 }
 
@@ -158,12 +153,13 @@ static ssize_t rfkill_state_store(struct
 	if (!capable(CAP_NET_ADMIN))
 		return -EPERM;
 
+	if (mutex_lock_interruptible(&rfkill->mutex))
+		return -ERESTARTSYS;
 	error = rfkill_toggle_radio(rfkill,
 			state ? RFKILL_STATE_ON : RFKILL_STATE_OFF);
-	if (error)
-		return error;
+	mutex_unlock(&rfkill->mutex);
 
-	return count;
+	return error ? error : count;
 }
 
 static ssize_t rfkill_claim_show(struct device *dev,

-- 
Greetings Michael.