On Tuesday 17 June 2008 22:59:52 David Ellingsworth wrote:
> I ran into this error today from a kernel I built last night based on
> the latest wireless-testing branch. Correct me if I'm wrong, but it
> looks to be b43legacy related. I'm a bit new to kernel debugging but
> can try to provide additional information if instructions on how to do
> so are provided.
No, this is a mac80211 bug
> --------------------------------------------------------------------------------------
> dmesg output:
> --------------------------------------------------------------------------------------
> WARNING: at kernel/softirq.c:141 local_bh_enable+0x1f/0x59()
> Modules linked in: ipt_MASQUERADE iptable_nat nf_nat nf_conntrack_ipv4
> nf_conntrack ip_tables x_tables radeon drm binfmt_misc ppdev lp
> dm_snapshot dm_mirror dm_log dm_mod b43legacy mac80211 cfg80211
> led_class snd_ali5451 snd_ac97_codec ac97_bus snd_pcm_oss
> snd_mixer_oss ati_agp ibmcam usbvideo ssb pcmcia snd_pcm snd_timer
> evdev videodev v4l1_compat container battery ac button firmware_class
> agpgart i2c_ali1535 snd video output parport_pc parport yenta_socket
> rsrc_nonstatic pcmcia_core psmouse serio_raw i2c_ali15x3 i2c_core
> soundcore snd_page_alloc pcspkr reiserfs ide_cd_mod cdrom ide_disk
> alim15x3 natsemi ide_pci_generic ide_core ohci_hcd usbcore thermal
> processor fan
> Pid: 1371, comm: b43legacy Not tainted 2.6.26-rc6-wl #1
> [<c0113be7>] warn_on_slowpath+0x40/0x66
> [<c0111c60>] check_preempt_wakeup+0xa2/0xc0
> [<c0120d0d>] autoremove_wake_function+0xc/0x2b
> [<c010fc44>] __wake_up_common+0x2d/0x52
> [<c011126e>] __wake_up+0xf/0x15
> [<c0113f89>] wake_up_klogd+0x2b/0x2d
> [<c01981ed>] scnprintf+0x1f/0x2b
> [<c0117477>] local_bh_enable+0x1f/0x59
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
This function does not like being run with IRQs disabled.
WARN_ON_ONCE(irqs_disabled());
> [<dcf44045>] __sta_info_unlink+0xa9/0x134 [mac80211]
> [<dcf4453f>] sta_info_unlink+0x9/0xd [mac80211]
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
This function disables interrupts
> [<dcf48968>] ieee80211_associated+0x9f/0x148 [mac80211]
> [<dcf4b463>] ieee80211_sta_work+0x4cb/0x64f [mac80211]
> [<c0110392>] update_curr+0x3d/0x52
> [<c01104ee>] dequeue_entity+0xf/0x8d
> [<c0111979>] dequeue_task_fair+0x13/0x27
> [<c010fab2>] dequeue_task+0xa/0x14
> [<c0249b01>] schedule+0x22e/0x24a
> [<dcf4af98>] ieee80211_sta_work+0x0/0x64f [mac80211]
> [<c011e7a7>] run_workqueue+0x63/0xca
> [<c011ec96>] worker_thread+0x0/0xb7
> [<c011ed42>] worker_thread+0xac/0xb7
> [<c0120d01>] autoremove_wake_function+0x0/0x2b
> [<c011ec96>] worker_thread+0x0/0xb7
> [<c0120ba6>] kthread+0x36/0x5c
> [<c0120b70>] kthread+0x0/0x5c
> [<c010390b>] kernel_thread_helper+0x7/0x10
I'm not sure _why_ local_bh_enable() doesn't like getting called
with interrupts disabled.
The code is a bit weird:
134 void local_bh_enable(void)
135 {
136 #ifdef CONFIG_TRACE_IRQFLAGS
137 unsigned long flags;
138
139 WARN_ON_ONCE(in_irq());
140 #endif
141 WARN_ON_ONCE(irqs_disabled());
142
143 #ifdef CONFIG_TRACE_IRQFLAGS
144 local_irq_save(flags);
145 #endif
146 /*
147 * Are softirqs going to be turned on now:
148 */
149 if (softirq_count() == SOFTIRQ_OFFSET)
150 trace_softirqs_on((unsigned long)__builtin_return_address(0));
151 /*
152 * Keep preemption disabled until we are done with
153 * softirq processing:
154 */
155 sub_preempt_count(SOFTIRQ_OFFSET - 1);
156
157 if (unlikely(!in_interrupt() && local_softirq_pending()))
158 do_softirq();
159
160 dec_preempt_count();
161 #ifdef CONFIG_TRACE_IRQFLAGS
162 local_irq_restore(flags);
163 #endif
164 preempt_check_resched();
165 }
166 EXPORT_SYMBOL(local_bh_enable);
--
Greetings Michael.
On Wed, Jun 18, 2008 at 6:02 AM, Johannes Berg
<[email protected]> wrote:
> On Tue, 2008-06-17 at 23:20 +0200, Michael Buesch wrote:
>> On Tuesday 17 June 2008 22:59:52 David Ellingsworth wrote:
>> > I ran into this error today from a kernel I built last night based on
>> > the latest wireless-testing branch. Correct me if I'm wrong, but it
>> > looks to be b43legacy related. I'm a bit new to kernel debugging but
>> > can try to provide additional information if instructions on how to do
>> > so are provided.
>>
>> No, this is a mac80211 bug
>
> Can you try the patch below?
>
> johannes
>
> --- everything.orig/net/mac80211/sta_info.h 2008-06-18 10:07:24.000000000 +0200
> +++ everything/net/mac80211/sta_info.h 2008-06-18 11:58:32.000000000 +0200
> @@ -164,6 +164,7 @@ struct sta_ampdu_mlme {
> * @aid: STA's unique AID (1..2007, 0 = not assigned yet),
> * only used in AP (and IBSS?) mode
> * @flags: STA flags, see &enum ieee80211_sta_info_flags
> + * @flaglock: spinlock for flags accesses
> * @ps_tx_buf: buffer of frames to transmit to this station
> * when it leaves power saving state
> * @tx_filtered: buffer of frames we already tried to transmit
> @@ -186,6 +187,7 @@ struct sta_info {
> struct rate_control_ref *rate_ctrl;
> void *rate_ctrl_priv;
> spinlock_t lock;
> + spinlock_t flaglock;
> struct ieee80211_ht_info ht_info;
> u64 supp_rates[IEEE80211_NUM_BANDS];
> u8 addr[ETH_ALEN];
> @@ -198,7 +200,10 @@ struct sta_info {
> */
> u8 pin_status;
>
> - /* frequently updated information, locked with lock spinlock */
> + /*
> + * frequently updated, locked with own spinlock (flaglock),
> + * use the accessors defined below
> + */
> u32 flags;
>
> /*
> @@ -293,34 +298,41 @@ static inline enum plink_state sta_plink
>
> static inline void set_sta_flags(struct sta_info *sta, const u32 flags)
> {
> - spin_lock_bh(&sta->lock);
> + unsigned long irqfl;
> +
> + spin_lock_irqsave(&sta->flaglock, irqfl);
> sta->flags |= flags;
> - spin_unlock_bh(&sta->lock);
> + spin_unlock_irqrestore(&sta->flaglock, irqfl);
> }
>
> static inline void clear_sta_flags(struct sta_info *sta, const u32 flags)
> {
> - spin_lock_bh(&sta->lock);
> + unsigned long irqfl;
> +
> + spin_lock_irqsave(&sta->flaglock, irqfl);
> sta->flags &= ~flags;
> - spin_unlock_bh(&sta->lock);
> + spin_unlock_irqrestore(&sta->flaglock, irqfl);
> }
>
> static inline void set_and_clear_sta_flags(struct sta_info *sta,
> const u32 set, const u32 clear)
> {
> - spin_lock_bh(&sta->lock);
> + unsigned long irqfl;
> +
> + spin_lock_irqsave(&sta->flaglock, irqfl);
> sta->flags |= set;
> sta->flags &= ~clear;
> - spin_unlock_bh(&sta->lock);
> + spin_unlock_irqrestore(&sta->flaglock, irqfl);
> }
>
> static inline u32 test_sta_flags(struct sta_info *sta, const u32 flags)
> {
> u32 ret;
> + unsigned long irqfl;
>
> - spin_lock_bh(&sta->lock);
> + spin_lock_irqsave(&sta->flaglock, irqfl);
> ret = sta->flags & flags;
> - spin_unlock_bh(&sta->lock);
> + spin_unlock_irqrestore(&sta->flaglock, irqfl);
>
> return ret;
> }
> @@ -329,11 +341,12 @@ static inline u32 test_and_clear_sta_fla
> const u32 flags)
> {
> u32 ret;
> + unsigned long irqfl;
>
> - spin_lock_bh(&sta->lock);
> + spin_lock_irqsave(&sta->flaglock, irqfl);
> ret = sta->flags & flags;
> sta->flags &= ~flags;
> - spin_unlock_bh(&sta->lock);
> + spin_unlock_irqrestore(&sta->flaglock, irqfl);
>
> return ret;
> }
> @@ -341,10 +354,11 @@ static inline u32 test_and_clear_sta_fla
> static inline u32 get_sta_flags(struct sta_info *sta)
> {
> u32 ret;
> + unsigned long irqfl;
>
> - spin_lock_bh(&sta->lock);
> + spin_lock_irqsave(&sta->flaglock, irqfl);
> ret = sta->flags;
> - spin_unlock_bh(&sta->lock);
> + spin_unlock_irqrestore(&sta->flaglock, irqfl);
>
> return ret;
> }
> --- everything.orig/net/mac80211/sta_info.c 2008-06-18 11:56:44.000000000 +0200
> +++ everything/net/mac80211/sta_info.c 2008-06-18 11:56:58.000000000 +0200
> @@ -235,6 +235,7 @@ struct sta_info *sta_info_alloc(struct i
> return NULL;
>
> spin_lock_init(&sta->lock);
> + spin_lock_init(&sta->flaglock);
>
> memcpy(sta->addr, addr, ETH_ALEN);
> sta->local = local;
>
>
>
Yes, this patch seems to correct the issue. I no longer receive the
kernel warning.
Regards,
David Ellingsworth
On Tue, 2008-06-17 at 23:20 +0200, Michael Buesch wrote:
> On Tuesday 17 June 2008 22:59:52 David Ellingsworth wrote:
> > I ran into this error today from a kernel I built last night based on
> > the latest wireless-testing branch. Correct me if I'm wrong, but it
> > looks to be b43legacy related. I'm a bit new to kernel debugging but
> > can try to provide additional information if instructions on how to do
> > so are provided.
>
> No, this is a mac80211 bug
Can you try the patch below?
johannes
--- everything.orig/net/mac80211/sta_info.h 2008-06-18 10:07:24.000000000 +0200
+++ everything/net/mac80211/sta_info.h 2008-06-18 11:58:32.000000000 +0200
@@ -164,6 +164,7 @@ struct sta_ampdu_mlme {
* @aid: STA's unique AID (1..2007, 0 = not assigned yet),
* only used in AP (and IBSS?) mode
* @flags: STA flags, see &enum ieee80211_sta_info_flags
+ * @flaglock: spinlock for flags accesses
* @ps_tx_buf: buffer of frames to transmit to this station
* when it leaves power saving state
* @tx_filtered: buffer of frames we already tried to transmit
@@ -186,6 +187,7 @@ struct sta_info {
struct rate_control_ref *rate_ctrl;
void *rate_ctrl_priv;
spinlock_t lock;
+ spinlock_t flaglock;
struct ieee80211_ht_info ht_info;
u64 supp_rates[IEEE80211_NUM_BANDS];
u8 addr[ETH_ALEN];
@@ -198,7 +200,10 @@ struct sta_info {
*/
u8 pin_status;
- /* frequently updated information, locked with lock spinlock */
+ /*
+ * frequently updated, locked with own spinlock (flaglock),
+ * use the accessors defined below
+ */
u32 flags;
/*
@@ -293,34 +298,41 @@ static inline enum plink_state sta_plink
static inline void set_sta_flags(struct sta_info *sta, const u32 flags)
{
- spin_lock_bh(&sta->lock);
+ unsigned long irqfl;
+
+ spin_lock_irqsave(&sta->flaglock, irqfl);
sta->flags |= flags;
- spin_unlock_bh(&sta->lock);
+ spin_unlock_irqrestore(&sta->flaglock, irqfl);
}
static inline void clear_sta_flags(struct sta_info *sta, const u32 flags)
{
- spin_lock_bh(&sta->lock);
+ unsigned long irqfl;
+
+ spin_lock_irqsave(&sta->flaglock, irqfl);
sta->flags &= ~flags;
- spin_unlock_bh(&sta->lock);
+ spin_unlock_irqrestore(&sta->flaglock, irqfl);
}
static inline void set_and_clear_sta_flags(struct sta_info *sta,
const u32 set, const u32 clear)
{
- spin_lock_bh(&sta->lock);
+ unsigned long irqfl;
+
+ spin_lock_irqsave(&sta->flaglock, irqfl);
sta->flags |= set;
sta->flags &= ~clear;
- spin_unlock_bh(&sta->lock);
+ spin_unlock_irqrestore(&sta->flaglock, irqfl);
}
static inline u32 test_sta_flags(struct sta_info *sta, const u32 flags)
{
u32 ret;
+ unsigned long irqfl;
- spin_lock_bh(&sta->lock);
+ spin_lock_irqsave(&sta->flaglock, irqfl);
ret = sta->flags & flags;
- spin_unlock_bh(&sta->lock);
+ spin_unlock_irqrestore(&sta->flaglock, irqfl);
return ret;
}
@@ -329,11 +341,12 @@ static inline u32 test_and_clear_sta_fla
const u32 flags)
{
u32 ret;
+ unsigned long irqfl;
- spin_lock_bh(&sta->lock);
+ spin_lock_irqsave(&sta->flaglock, irqfl);
ret = sta->flags & flags;
sta->flags &= ~flags;
- spin_unlock_bh(&sta->lock);
+ spin_unlock_irqrestore(&sta->flaglock, irqfl);
return ret;
}
@@ -341,10 +354,11 @@ static inline u32 test_and_clear_sta_fla
static inline u32 get_sta_flags(struct sta_info *sta)
{
u32 ret;
+ unsigned long irqfl;
- spin_lock_bh(&sta->lock);
+ spin_lock_irqsave(&sta->flaglock, irqfl);
ret = sta->flags;
- spin_unlock_bh(&sta->lock);
+ spin_unlock_irqrestore(&sta->flaglock, irqfl);
return ret;
}
--- everything.orig/net/mac80211/sta_info.c 2008-06-18 11:56:44.000000000 +0200
+++ everything/net/mac80211/sta_info.c 2008-06-18 11:56:58.000000000 +0200
@@ -235,6 +235,7 @@ struct sta_info *sta_info_alloc(struct i
return NULL;
spin_lock_init(&sta->lock);
+ spin_lock_init(&sta->flaglock);
memcpy(sta->addr, addr, ETH_ALEN);
sta->local = local;