Hi Marek,

I've checked today's linux-next with the updated commit 27bc93583e35 
("cfg80211: avoid holding the RTNL when calling the driver") and there 
is still an issue there, but at least it doesn't cause a deadlock:

cfg80211: Loading compiled-in X.509 certificates for regulatory database
Bluetooth: vendor=0x2df, device=0x912a, class=255, fn=2
cfg80211: Loaded X.509 cert 'sforshee: 00b28ddf47aef9cea7'
Bluetooth: FW download over, size 533976 bytes
btmrvl_sdio mmc3:0001:2: sdio device tree data not available
mwifiex_sdio mmc3:0001:1: WLAN FW already running! Skip FW dnld
mwifiex_sdio mmc3:0001:1: WLAN FW is active
mwifiex_sdio mmc3:0001:1: CMD_RESP: cmd 0x242 error, result=0x2
mwifiex_sdio mmc3:0001:1: mwifiex_process_cmdresp: cmd 0x242 failed 
during       initialization
------------[ cut here ]------------
WARNING: CPU: 0 PID: 5 at net/wireless/core.c:1336 
cfg80211_register_netdevice+0xa4/0x198 [cfg80211]

Yeah, umm, brown paper bag style bug.

I meant to _move_ that line down, but somehow managed to _copy_ it down.
Need to just remove it since rdev is not even initialized at that point.

I've updated my tree to include this:

diff --git a/net/wireless/core.c b/net/wireless/core.c
index 5e8b523dc645..200cd9f5fd5f 100644
--- a/net/wireless/core.c
+++ b/net/wireless/core.c

@@ -1333,7 +1333,6 @@ int cfg80211_register_netdevice(struct net_device *dev)
        int ret;
 
        ASSERT_RTNL();
-       lockdep_assert_held(&rdev->wiphy.mtx);
 
        if (WARN_ON(!wdev))
                return -EINVAL;

johannes