Re: [PATCH v4 bluetooth] Fix lowpan_rcv

[Martin Townsend <martin.townsend@xsilon.com>]

Hi Jukka,

Very bizarre that the patch causes a locking issue in the Tx path.  Is 
it a side effect from not cloning the skb before calling process_data.  
I couldn't see why a clone was being made so maybe I'm missing something 
here.

- Martin.


On 02/10/14 14:55, Jukka Rissanen wrote:
> Hi Martin,
>
> On to, 2014-10-02 at 13:16 +0100, Martin Townsend wrote:
>> Hi Jukka,
>>
>> Thanks for investigating this.
>>
>> If you set the dev in your receive function does the oops go away,  BTW this is where the pkt_type is set to PACKET_HOST, for us this was causing problems, I don't know if this is the same for you guys.
>>
>> Here's the code:
>>
>>              local_skb->protocol = htons(ETH_P_IPV6);
>>>>>      local_skb->dev = dev;
>>              local_skb->pkt_type = PACKET_HOST;
>>
>>              if (give_skb_to_upper(local_skb, dev)
>>                              != NET_RX_SUCCESS) {
>>                  kfree_skb(local_skb);
>>
>> - Martin.
> So I applied this patch
>
> diff --git a/net/bluetooth/6lowpan.c b/net/bluetooth/6lowpan.c
> index e4f5ce5..aa64f91 100644
> --- a/net/bluetooth/6lowpan.c
> +++ b/net/bluetooth/6lowpan.c
> @@ -323,6 +323,7 @@ static int recv_pkt(struct sk_buff *skb, struct
> net_device *dev,
>   
>                  local_skb->protocol = htons(ETH_P_IPV6);
>                  local_skb->pkt_type = PACKET_HOST;
> +               local_skb->dev = dev;
>   
>                  if (give_skb_to_upper(local_skb, dev) != NET_RX_SUCCESS)
> {
>                          goto drop_local_skb;
> @@ -362,6 +363,7 @@ static int recv_pkt(struct sk_buff *skb, struct
> net_device *dev,
>   
>                          local_skb->protocol = htons(ETH_P_IPV6);
>                          local_skb->pkt_type = PACKET_HOST;
> +                       local_skb->dev = dev;
>   
>                          if (give_skb_to_upper(local_skb, dev)
>                                                          !=
> NET_RX_SUCCESS) {
>
>
> and I do not see the earlier null pointer oops any more.
>
>
> Unfortunately a new error is seen:
>
> [  340.676353] inconsistent {IN-SOFTIRQ-W} -> {SOFTIRQ-ON-W} usage.
> [  340.676353] kworker/u3:1/370 [HC0[0]:SC0[0]:HE1:SE1] takes:
> [  340.676353]  (&(&list->lock)->rlock#6){+.?...}, at: [<f8319d4c>]
> hci_send_acl+0xac/0x290 [bluetooth]
> [  340.676353] {IN-SOFTIRQ-W} state was registered at:
> [  340.676353]   [<c10915a3>] __lock_acquire+0x6d3/0x1d20
> [  340.676353]   [<c109325d>] lock_acquire+0x9d/0x140
> [  340.676353]   [<c1889c25>] _raw_spin_lock+0x45/0x80
> [  340.676353]   [<f8319d4c>] hci_send_acl+0xac/0x290 [bluetooth]
> [  340.676353]   [<f833abf0>] l2cap_do_send+0x60/0x100 [bluetooth]
> [  340.676353]   [<f833e7c0>] l2cap_chan_send+0x7f0/0x10e0 [bluetooth]
> [  340.676353]   [<f850691e>] send_pkt+0x4e/0xa0 [bluetooth_6lowpan]
> [  340.676353]   [<f8506d20>] bt_xmit+0x3b0/0x770 [bluetooth_6lowpan]
> [  340.676353]   [<c17742f4>] dev_hard_start_xmit+0x344/0x670
> [  340.676353]   [<c17749ad>] __dev_queue_xmit+0x38d/0x680
> [  340.676353]   [<c1774caf>] dev_queue_xmit+0xf/0x20
> [  340.676353]   [<c177b8b0>] neigh_connected_output+0x130/0x1a0
> [  340.676353]   [<c1812a63>] ip6_finish_output2+0x173/0x8c0
> [  340.676353]   [<c18182db>] ip6_finish_output+0x7b/0x1b0
> [  340.676353]   [<c18184a7>] ip6_output+0x97/0x2a0
> ...
>
> I have to investigate this more what your new code is doing as I did not
> see this error earlier before your patch.
>
> So I have very latest bluetooth-next (commit b57d4471fd18) + your v4
> patch + above patch that sets the dev pointer, and this combination
> triggers the above inconsistent softirq usage error.
>
>
> Cheers,
> Jukka
>
>