When adding 6lowpan devices very rapidly we sometimes see a crash:
[23122.306615] CPU: 2 PID: 0 Comm: swapper/2 Not tainted 4.9.0-43-arm64 #1 Debian 4.9.9.linaro.43-1
[23122.315400] Hardware name: HiKey Development Board (DT)
[23122.320623] task: ffff800075443080 task.stack: ffff800075484000
[23122.326551] PC is at expire_timers+0x70/0x150
[23122.330907] LR is at run_timer_softirq+0xa0/0x1a0
[23122.335616] pc : [<ffff000008142dd8>] lr : [<ffff000008142f58>] pstate: 600001c5
This was due to add_peer_chan() unconditionally initializing the
lowpan_btle_dev->notify_peers delayed work structure, even if the
lowpan_btle_dev passed into add_peer_chan() had previously been
initialized.
Normally, this would go unnoticed as the delayed work timer is set for
100 msec, however when calling add_peer_chan() faster than 100 msec it
clears out a previously queued delay work causing the crash above.
To fix this, let add_peer_chan() know when a new lowpan_btle_dev is passed
in so that it only performs the delay work initialization when needed.
Signed-off-by: Michael Scott <[email protected]>
---
net/bluetooth/6lowpan.c | 10 +++++++---
1 file changed, 7 insertions(+), 3 deletions(-)
diff --git a/net/bluetooth/6lowpan.c b/net/bluetooth/6lowpan.c
index e27be3ca0a0c..c282482edc2c 100644
--- a/net/bluetooth/6lowpan.c
+++ b/net/bluetooth/6lowpan.c
@@ -754,7 +754,8 @@ static void set_ip_addr_bits(u8 addr_type, u8 *addr)
}
static struct l2cap_chan *add_peer_chan(struct l2cap_chan *chan,
- struct lowpan_btle_dev *dev)
+ struct lowpan_btle_dev *dev,
+ bool new_netdev)
{
struct lowpan_peer *peer;
@@ -785,7 +786,8 @@ static struct l2cap_chan *add_peer_chan(struct l2cap_chan *chan,
spin_unlock(&devices_lock);
/* Notifying peers about us needs to be done without locks held */
- INIT_DELAYED_WORK(&dev->notify_peers, do_notify_peers);
+ if (new_netdev)
+ INIT_DELAYED_WORK(&dev->notify_peers, do_notify_peers);
schedule_delayed_work(&dev->notify_peers, msecs_to_jiffies(100));
return peer->chan;
@@ -842,6 +844,7 @@ static int setup_netdev(struct l2cap_chan *chan, struct lowpan_btle_dev **dev)
static inline void chan_ready_cb(struct l2cap_chan *chan)
{
struct lowpan_btle_dev *dev;
+ bool new_netdev = false;
dev = lookup_dev(chan->conn);
@@ -852,12 +855,13 @@ static inline void chan_ready_cb(struct l2cap_chan *chan)
l2cap_chan_del(chan, -ENOENT);
return;
}
+ new_netdev = true;
}
if (!try_module_get(THIS_MODULE))
return;
- add_peer_chan(chan, dev);
+ add_peer_chan(chan, dev, new_netdev);
ifup(dev->netdev);
}
--
2.11.0
Hi Michael,
On Tue, 2017-03-28 at 23:10 -0700, Michael Scott wrote:
> When adding 6lowpan devices very rapidly we sometimes see a crash:
> [23122.306615] CPU: 2 PID: 0 Comm: swapper/2 Not tainted 4.9.0-43-
> arm64 #1 Debian 4.9.9.linaro.43-1
> [23122.315400] Hardware name: HiKey Development Board (DT)
> [23122.320623] task: ffff800075443080 task.stack: ffff800075484000
> [23122.326551] PC is at expire_timers+0x70/0x150
> [23122.330907] LR is at run_timer_softirq+0xa0/0x1a0
> [23122.335616] pc : [<ffff000008142dd8>] lr : [<ffff000008142f58>]
> pstate: 600001c5
>
> This was due to add_peer_chan() unconditionally initializing the
> lowpan_btle_dev->notify_peers delayed work structure, even if the
> lowpan_btle_dev passed into add_peer_chan() had previously been
> initialized.
>
> Normally, this would go unnoticed as the delayed work timer is set
> for
> 100 msec, however when calling add_peer_chan() faster than 100 msec
> it
> clears out a previously queued delay work causing the crash above.
>
> To fix this, let add_peer_chan() know when a new lowpan_btle_dev is
> passed
> in so that it only performs the delay work initialization when
> needed.
>
> Signed-off-by: Michael Scott <[email protected]>
> ---
> net/bluetooth/6lowpan.c | 10 +++++++---
> 1 file changed, 7 insertions(+), 3 deletions(-)
>
> diff --git a/net/bluetooth/6lowpan.c b/net/bluetooth/6lowpan.c
> index e27be3ca0a0c..c282482edc2c 100644
> --- a/net/bluetooth/6lowpan.c
> +++ b/net/bluetooth/6lowpan.c
> @@ -754,7 +754,8 @@ static void set_ip_addr_bits(u8 addr_type, u8
> *addr)
> }
>
> static struct l2cap_chan *add_peer_chan(struct l2cap_chan *chan,
> - struct lowpan_btle_dev *dev)
> + struct lowpan_btle_dev *dev,
> + bool new_netdev)
> {
> struct lowpan_peer *peer;
>
> @@ -785,7 +786,8 @@ static struct l2cap_chan *add_peer_chan(struct
> l2cap_chan *chan,
> spin_unlock(&devices_lock);
>
> /* Notifying peers about us needs to be done without locks
> held */
> - INIT_DELAYED_WORK(&dev->notify_peers, do_notify_peers);
> + if (new_netdev)
> + INIT_DELAYED_WORK(&dev->notify_peers,
> do_notify_peers);
> schedule_delayed_work(&dev->notify_peers,
> msecs_to_jiffies(100));
>
> return peer->chan;
> @@ -842,6 +844,7 @@ static int setup_netdev(struct l2cap_chan *chan,
> struct lowpan_btle_dev **dev)
> static inline void chan_ready_cb(struct l2cap_chan *chan)
> {
> struct lowpan_btle_dev *dev;
> + bool new_netdev = false;
>
> dev = lookup_dev(chan->conn);
>
> @@ -852,12 +855,13 @@ static inline void chan_ready_cb(struct
> l2cap_chan *chan)
> l2cap_chan_del(chan, -ENOENT);
> return;
> }
> + new_netdev = true;
> }
>
> if (!try_module_get(THIS_MODULE))
> return;
>
> - add_peer_chan(chan, dev);
> + add_peer_chan(chan, dev, new_netdev);
> ifup(dev->netdev);
> }
>
Good catch!
Acked-by: Jukka Rissanen <[email protected]>
Cheers,
Jukka
Hi Michael,
> When adding 6lowpan devices very rapidly we sometimes see a crash:
> [23122.306615] CPU: 2 PID: 0 Comm: swapper/2 Not tainted 4.9.0-43-arm64 #1 Debian 4.9.9.linaro.43-1
> [23122.315400] Hardware name: HiKey Development Board (DT)
> [23122.320623] task: ffff800075443080 task.stack: ffff800075484000
> [23122.326551] PC is at expire_timers+0x70/0x150
> [23122.330907] LR is at run_timer_softirq+0xa0/0x1a0
> [23122.335616] pc : [<ffff000008142dd8>] lr : [<ffff000008142f58>] pstate: 600001c5
>
> This was due to add_peer_chan() unconditionally initializing the
> lowpan_btle_dev->notify_peers delayed work structure, even if the
> lowpan_btle_dev passed into add_peer_chan() had previously been
> initialized.
>
> Normally, this would go unnoticed as the delayed work timer is set for
> 100 msec, however when calling add_peer_chan() faster than 100 msec it
> clears out a previously queued delay work causing the crash above.
>
> To fix this, let add_peer_chan() know when a new lowpan_btle_dev is passed
> in so that it only performs the delay work initialization when needed.
>
> Signed-off-by: Michael Scott <[email protected]>
> ---
> net/bluetooth/6lowpan.c | 10 +++++++---
> 1 file changed, 7 insertions(+), 3 deletions(-)
patch has been applied to bluetooth-next tree.
Regards
Marcel