2013-02-25 16:09:44

by Larry Finger

Subject: [PATCH] b43: Fix lockdep splat on module unload

On unload, b43 produces a lockdep warning that can be summarized in the
following way:

======================================================
[ INFO: possible circular locking dependency detected ]
3.8.0-wl+ #117 Not tainted
-------------------------------------------------------
modprobe/5557 is trying to acquire lock:
((&wl->firmware_load)){+.+.+.}, at: [<ffffffff81062160>] flush_work+0x0/0x2a0

but task is already holding lock:
(rtnl_mutex){+.+.+.}, at: [<ffffffff813bd7d2>] rtnl_lock+0x12/0x20

which lock already depends on the new lock.
[ INFO: possible circular locking dependency detected ]
======================================================

The full output is available at http://lkml.indiana.edu/hypermail/linux/kernel/1302.3/00060.html.
To summarize, commit 6b6fa58 added a 'cancel_work_sync(&wl->firmware_load)'
call in the wrong place.

The fix is to move the cancel_work_sync() call to b43_bcma_remove() and
b43_ssb_remove(). Thanks to Johannes Berg and Michael Buesch for help in
diagnosing the log output.

Signed-off-by: Larry Finger <[email protected]>
Cc: Stable <[email protected]> [V3.5+]
---
drivers/net/wireless/b43/main.c | 3 ++-
1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/drivers/net/wireless/b43/main.c b/drivers/net/wireless/b43/main.c
index 806e34c..0568273 100644
--- a/drivers/net/wireless/b43/main.c
+++ b/drivers/net/wireless/b43/main.c
@@ -4214,7 +4214,6 @@ redo:
mutex_unlock(&wl->mutex);
cancel_delayed_work_sync(&dev->periodic_work);
cancel_work_sync(&wl->tx_work);
- cancel_work_sync(&wl->firmware_load);
mutex_lock(&wl->mutex);
dev = wl->current_dev;
if (!dev || b43_status(dev) < B43_STAT_STARTED) {
@@ -5434,6 +5433,7 @@ static void b43_bcma_remove(struct bcma_device *core)
/* We must cancel any work here before unregistering from ieee80211,
* as the ieee80211 unreg will destroy the workqueue. */
cancel_work_sync(&wldev->restart_work);
+ cancel_work_sync(&wl->firmware_load);

B43_WARN_ON(!wl);
if (!wldev->fw.ucode.data)
@@ -5510,6 +5510,7 @@ static void b43_ssb_remove(struct ssb_device *sdev)
/* We must cancel any work here before unregistering from ieee80211,
* as the ieee80211 unreg will destroy the workqueue. */
cancel_work_sync(&wldev->restart_work);
+ cancel_work_sync(&wl->firmware_load);

B43_WARN_ON(!wl);
if (!wldev->fw.ucode.data)
--
1.8.1.4

2013-02-25 20:48:07

by David Miller

Subject: Re: [PATCH] b43: Fix lockdep splat on module unload

From: Larry Finger <[email protected]>
Date: Mon, 25 Feb 2013 10:09:24 -0600

> On unload, b43 produces a lockdep warning that can be summarized in the
> following way:
>
> ======================================================
> [ INFO: possible circular locking dependency detected ]
> 3.8.0-wl+ #117 Not tainted
> -------------------------------------------------------
> modprobe/5557 is trying to acquire lock:
> ((&wl->firmware_load)){+.+.+.}, at: [<ffffffff81062160>] flush_work+0x0/0x2a0
>
> but task is already holding lock:
> (rtnl_mutex){+.+.+.}, at: [<ffffffff813bd7d2>] rtnl_lock+0x12/0x20
>
> which lock already depends on the new lock.
> [ INFO: possible circular locking dependency detected ]
> ======================================================
>
> The full output is available at http://lkml.indiana.edu/hypermail/linux/kernel/1302.3/00060.html.
> To summarize, commit 6b6fa58 added a 'cancel_work_sync(&wl->firmware_load)'
> call in the wrong place.
>
> The fix is to move the cancel_work_sync() call to b43_bcma_remove() and
> b43_ssb_remove(). Thanks to Johannes Berg and Michael Buesch for help in
> diagnosing the log output.
>
> Signed-off-by: Larry Finger <[email protected]>

Applied.