On 10/09/2012 11:34 PM, Gilles Chanteperdrix wrote:
> On 10/09/2012 02:45 PM, Thierry Bultel wrote:
>> (snip)
>
>> (gdb) bt
>> #0 __ticket_spin_lock (lock=<optimized out>) at
>> /home/thierry/workspace/Buildroot/output/build/linux-3.2.21/arch/x86/include/asm/spinlock.h:63
>> #1 arch_spin_lock (lock=<optimized out>) at
>> /home/thierry/workspace/Buildroot/output/build/linux-3.2.21/arch/x86/include/asm/spinlock.h:129
>> #2 rtcan_raw_recvmsg (context=0xc7ff0000, user_info=0xd00b2290,
>> msg=0xd796df04, flags=<optimized out>) at
>> drivers/xenomai/can/rtcan_raw.c:655
>> (snip)
>>
>>
>> In my previous configuration, disabling CONFIG_SMP was enough to
>> workaround it, but in that new case
>> I have other bad side-effects (console and X server frozen at startup)
>> that prevent me to work correctly.
>
>
> That is essentially the same backtrace as in the first case. Could you
> try the following patch?
>
> diff --git a/ksrc/drivers/can/rtcan_raw.c b/ksrc/drivers/can/rtcan_raw.c
> index ded024e..b07f19c 100644
> --- a/ksrc/drivers/can/rtcan_raw.c
> +++ b/ksrc/drivers/can/rtcan_raw.c
> @@ -695,10 +695,7 @@ ssize_t rtcan_raw_recvmsg(struct rtdm_dev_context
> *context,
>
> /* Message completely read from the socket's ring buffer. Now check if
> * caller is just peeking. */
> - if (flags & MSG_PEEK)
> - /* Next one, please! */
> - rtdm_sem_up(&sock->recv_sem);
> - else
> + if ((flags & MSG_PEEK) == 0)
> /* Adjust begin of first message in the ring buffer. */
> sock->recv_head = recv_buf_index;
>
> @@ -707,6 +704,11 @@ ssize_t rtcan_raw_recvmsg(struct rtdm_dev_context
> *context,
> rtdm_lock_put_irqrestore(&rtcan_socket_lock, lock_ctx);
>
>
> + if (flags & MSG_PEEK)
> + /* Next one, please! */
> + rtdm_sem_up(&sock->recv_sem);
> +
> +
> /* Create CAN socket address to give back */
> if (msg->msg_namelen) {
> scan.can_family = AF_CAN;
>
>
Another possible fix would be to lock xenomai scheduler when taking an
rtdm_spin_lock, as it is almost always wrong to call xnpod_schedule in
such a situation. It works for other RT-CAN drivers, because they call
rtdm_sem_up from interrupt context where the scheduler is locked by the
XNINIRQ bit.
--
Gilles.
_______________________________________________
Xenomai mailing list
[email protected]
http://www.xenomai.org/mailman/listinfo/xenomai
