[RFC] mac80211: don't assume driver has been attached on registration

[RFC] mac80211: don't assume driver has been attached on registration

[Luis R. Rodriguez]

mac80211's ieee80211_register_hw() is often called within the
probe path so it cannot assume the device's driver structure
has been attached yet so to create a workqueue instead of
using driver->name use the wiphy's phy%d name.

This should fix sporadic oopses found when we race to beat the
driver pointer setting. Not even sure how this was working properly.

http://www.kerneloops.org/search.php?search=ieee80211_register_hw

Signed-off-by: Luis R. Rodriguez <lrodriguez@atheros.com>
---

Tested with iwlagn. If you are running into this oops please test.

 net/mac80211/main.c |    5 ++---
 1 files changed, 2 insertions(+), 3 deletions(-)

diff --git a/net/mac80211/main.c b/net/mac80211/main.c
index d631dc9..cec9b6d 100644
--- a/net/mac80211/main.c
+++ b/net/mac80211/main.c
@@ -722,7 +722,6 @@ EXPORT_SYMBOL(ieee80211_alloc_hw);
 int ieee80211_register_hw(struct ieee80211_hw *hw)
 {
 	struct ieee80211_local *local = hw_to_local(hw);
-	const char *name;
 	int result;
 	enum ieee80211_band band;
 	struct net_device *mdev;
@@ -787,8 +786,8 @@ int ieee80211_register_hw(struct ieee80211_hw *hw)
 	mdev->header_ops = &ieee80211_header_ops;
 	mdev->set_multicast_list = ieee80211_master_set_multicast_list;
 
-	name = wiphy_dev(local->hw.wiphy)->driver->name;
-	local->hw.workqueue = create_freezeable_workqueue(name);
+	local->hw.workqueue =
+		create_freezeable_workqueue(wiphy_name(local->hw.wiphy));
 	if (!local->hw.workqueue) {
 		result = -ENOMEM;
 		goto fail_workqueue;
-- 
1.5.6.3

Re: [RFC] mac80211: don't assume driver has been attached on registration

[Johannes Berg]

[-- Attachment #1: Type: text/plain, Size: 1881 bytes --]

On Fri, 2008-11-14 at 13:12 -0800, Luis R. Rodriguez wrote:
> mac80211's ieee80211_register_hw() is often called within the
> probe path so it cannot assume the device's driver structure
> has been attached yet so to create a workqueue instead of
> using driver->name use the wiphy's phy%d name.
> 
> This should fix sporadic oopses found when we race to beat the
> driver pointer setting. Not even sure how this was working properly.
> 
> http://www.kerneloops.org/search.php?search=ieee80211_register_hw

Ok, umm, very odd. Why is it sometimes working? It should always fail...

Patch looks ok though, the name of the thing doesn't really matter
anyway...

> Signed-off-by: Luis R. Rodriguez <lrodriguez@atheros.com>

Acked-by: Johannes Berg <johannes@sipsolutions.net>

> ---
> 
> Tested with iwlagn. If you are running into this oops please test.
> 
>  net/mac80211/main.c |    5 ++---
>  1 files changed, 2 insertions(+), 3 deletions(-)
> 
> diff --git a/net/mac80211/main.c b/net/mac80211/main.c
> index d631dc9..cec9b6d 100644
> --- a/net/mac80211/main.c
> +++ b/net/mac80211/main.c
> @@ -722,7 +722,6 @@ EXPORT_SYMBOL(ieee80211_alloc_hw);
>  int ieee80211_register_hw(struct ieee80211_hw *hw)
>  {
>  	struct ieee80211_local *local = hw_to_local(hw);
> -	const char *name;
>  	int result;
>  	enum ieee80211_band band;
>  	struct net_device *mdev;
> @@ -787,8 +786,8 @@ int ieee80211_register_hw(struct ieee80211_hw *hw)
>  	mdev->header_ops = &ieee80211_header_ops;
>  	mdev->set_multicast_list = ieee80211_master_set_multicast_list;
>  
> -	name = wiphy_dev(local->hw.wiphy)->driver->name;
> -	local->hw.workqueue = create_freezeable_workqueue(name);
> +	local->hw.workqueue =
> +		create_freezeable_workqueue(wiphy_name(local->hw.wiphy));
>  	if (!local->hw.workqueue) {
>  		result = -ENOMEM;
>  		goto fail_workqueue;

[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 836 bytes --]