1 From: =?UTF-8?q?Rafa=C5=82=20Mi=C5=82ecki?= <zajec5@gmail.com>
2 Date: Thu, 9 Jul 2015 00:07:59 +0200
3 Subject: [PATCH] brcmfmac: workaround bug with some inconsistent BSSes state
5 Content-Type: text/plain; charset=UTF-8
6 Content-Transfer-Encoding: 8bit
8 Signed-off-by: Rafał Miłecki <zajec5@gmail.com>
11 --- a/drivers/net/wireless/broadcom/brcm80211/brcmfmac/cfg80211.c
12 +++ b/drivers/net/wireless/broadcom/brcm80211/brcmfmac/cfg80211.c
13 @@ -639,8 +639,36 @@ static struct wireless_dev *brcmf_cfg802
14 struct brcmf_cfg80211_info *cfg = wiphy_to_cfg(wiphy);
15 struct brcmf_pub *drvr = cfg->pub;
16 struct wireless_dev *wdev;
17 + struct net_device *dev;
21 + * There is a bug with in-firmware BSS management. When adding virtual
22 + * interface brcmfmac first tells firmware to create new BSS and then
23 + * it creates new struct net_device.
25 + * If creating/registering netdev(ice) fails, BSS remains in some bugged
26 + * state. It conflicts with existing BSSes by overtaking their auth
29 + * It results in one BSS (addresss X) sending beacons and another BSS
30 + * (address Y) replying to authentication requests. This makes interface
33 + * To workaround this bug we may try to guess if register_netdev(ice)
34 + * will fail. The most obvious case is using interface name that already
35 + * exists. This is actually quite likely with brcmfmac & some user space
36 + * scripts as brcmfmac doesn't allow deleting virtual interfaces.
37 + * So this bug can be triggered even by something trivial like:
38 + * iw dev wlan0 delete
39 + * iw phy phy0 interface add wlan0 type __ap
41 + dev = dev_get_by_name(&init_net, name);
44 + return ERR_PTR(-ENFILE);
47 brcmf_dbg(TRACE, "enter: %s type %d\n", name, type);
48 err = brcmf_vif_add_validate(wiphy_to_cfg(wiphy), type);