diff mbox

net/can/mscan: Enable interrupts when all TX buffers are occupied to get notified when they are available again

Message ID C3745E3C7FD07D429C54EB3D7606A5A7677E7C@ZRHEX021.ads.zuehlke.com
State Changes Requested, archived
Delegated to: David Miller
Headers show

Commit Message

Mosler, Martin Nov. 28, 2011, 1:25 p.m. UTC 
When all TX buffers of the mscan are occupied the network layer is told to stop sending further packets. However the TX interrupts are not enabled and therefore the driver has no chance to tell the network layer when it is ready to accept further packets.

Signed-off-by: Martin Mosler <mmo@zuehlke.com>

--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Comments

Wolfgang Grandegger Nov. 28, 2011, 1:37 p.m. UTC | #1 
On 11/28/2011 02:25 PM, Mosler, Martin wrote:
> When all TX buffers of the mscan are occupied the network layer is told to stop sending further packets. However the TX interrupts are not enabled and therefore the driver has no chance to tell the network layer when it is ready to accept further packets.
> 
> Signed-off-by: Martin Mosler <mmo@zuehlke.com>
> 
> --- linux-2.6/drivers/net/can/mscan/mscan.c.orig    2011-11-28 13:54:29.547850661 +0100
> +++ linux-2.6/drivers/net/can/mscan/mscan.c 2011-11-28 13:55:52.427849601 +0100
> @@ -214,6 +214,7 @@ static netdev_tx_t mscan_start_xmit(stru
>     case 0:
>         netif_stop_queue(dev);
>         dev_err(dev->dev.parent, "Tx Ring full when queue awake!\n");
> +       out_8(&regs->cantier, priv->tx_active);
>         return NETDEV_TX_BUSY;
>     case 1:
>         /*  

Hm, did you see the error message. Actually, it should never happen, IIRC.

Wolfgang.

--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Mosler, Martin Nov. 28, 2011, 1:59 p.m. UTC | #2 
>> When all TX buffers of the mscan are occupied the network layer is told to stop sending further packets. However the TX interrupts are not enabled and therefore the driver has no chance to tell the network layer when it is ready to accept further packets.
>>
>> Signed-off-by: Martin Mosler <mmo@zuehlke.com>
>>
>> --- linux-2.6/drivers/net/can/mscan/mscan.c.orig    2011-11-28 13:54:29.547850661 +0100
>> +++ linux-2.6/drivers/net/can/mscan/mscan.c 2011-11-28 13:55:52.427849601 +0100
>> @@ -214,6 +214,7 @@ static netdev_tx_t mscan_start_xmit(stru
>>     case 0:
>>         netif_stop_queue(dev);
>>         dev_err(dev->dev.parent, "Tx Ring full when queue awake!\n");
>> +       out_8(&regs->cantier, priv->tx_active);
>>         return NETDEV_TX_BUSY;
>>     case 1:
>>         /*
>
>Hm, did you see the error message. Actually, it should never happen, IIRC.
>
>Wolfgang.

It is in fact a very rare condition, but it was triggered during testing when pulling CAN-HI and CAN-LO lines
to GND and VCC in various combinations to verify how the system is recovering. I am working with a MPC5125 
if this is valuable information. (This was only one issue I ran into, I had to do other modifications for other 
issues as well, but I think they are chip specific, I'll share them in a separate thread if you like).

--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Wolfgang Grandegger Nov. 28, 2011, 2:08 p.m. UTC | #3 
On 11/28/2011 02:59 PM, Mosler, Martin wrote:
>>> When all TX buffers of the mscan are occupied the network layer is told to stop sending further packets. However the TX interrupts are not enabled and therefore the driver has no chance to tell the network layer when it is ready to accept further packets.
>>>
>>> Signed-off-by: Martin Mosler <mmo@zuehlke.com>
>>>
>>> --- linux-2.6/drivers/net/can/mscan/mscan.c.orig    2011-11-28 13:54:29.547850661 +0100
>>> +++ linux-2.6/drivers/net/can/mscan/mscan.c 2011-11-28 13:55:52.427849601 +0100
>>> @@ -214,6 +214,7 @@ static netdev_tx_t mscan_start_xmit(stru
>>>     case 0:
>>>         netif_stop_queue(dev);
>>>         dev_err(dev->dev.parent, "Tx Ring full when queue awake!\n");
>>> +       out_8(&regs->cantier, priv->tx_active);
>>>         return NETDEV_TX_BUSY;
>>>     case 1:
>>>         /*
>>
>> Hm, did you see the error message. Actually, it should never happen, IIRC.
>>
>> Wolfgang.
> 
> It is in fact a very rare condition, but it was triggered during testing when pulling CAN-HI and CAN-LO lines
> to GND and VCC in various combinations to verify how the system is recovering. I am working with a MPC5125 
> if this is valuable information. (This was only one issue I ran into, I had to do other modifications for other 
> issues as well, but I think they are chip specific, I'll share them in a separate thread if you like).

Sounds like a problem with bus-off recovery. The software has restarted
the queue but the TX objects are still active. How do you handle
bus-offs? Manually? restart-ms = ?

Would be great if you post patches for your other issues.

Wolfgang.




--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Mosler, Martin Nov. 28, 2011, 2:13 p.m. UTC | #4 
>>>> When all TX buffers of the mscan are occupied the network layer is told to stop sending further packets. However the TX interrupts are not enabled and therefore the driver has no chance to tell the network layer when it is ready to accept further packets.
>>>>
>>>> Signed-off-by: Martin Mosler <mmo@zuehlke.com>
>>>>
>>>> --- linux-2.6/drivers/net/can/mscan/mscan.c.orig    2011-11-28 13:54:29.547850661 +0100
>>>> +++ linux-2.6/drivers/net/can/mscan/mscan.c 2011-11-28 13:55:52.427849601 +0100
>>>> @@ -214,6 +214,7 @@ static netdev_tx_t mscan_start_xmit(stru
>>>>     case 0:
>>>>         netif_stop_queue(dev);
>>>>         dev_err(dev->dev.parent, "Tx Ring full when queue awake!\n");
>>>> +       out_8(&regs->cantier, priv->tx_active);
>>>>         return NETDEV_TX_BUSY;
>>>>     case 1:
>>>>         /*
>>>
>>> Hm, did you see the error message. Actually, it should never happen, IIRC.
>>>
>>> Wolfgang.
>>
>> It is in fact a very rare condition, but it was triggered during testing when pulling CAN-HI and CAN-LO lines
>> to GND and VCC in various combinations to verify how the system is recovering. I am working with a MPC5125
>> if this is valuable information. (This was only one issue I ran into, I had to do other modifications for other
>> issues as well, but I think they are chip specific, I'll share them in a separate thread if you like).
>
>Sounds like a problem with bus-off recovery. The software has restarted
>the queue but the TX objects are still active. How do you handle
>bus-offs? Manually? restart-ms = ?
>
>Would be great if you post patches for your other issues.
>
>Wolfgang.

We are using restart-ms = 500 for now.

I'll post the other patches after I know I did the submit process correctly for this one and when I reviewed 
them more closely, as it is my first submission to the kernel.





--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Wolfgang Grandegger Nov. 28, 2011, 2:22 p.m. UTC | #5 
On 11/28/2011 02:25 PM, Mosler, Martin wrote:
> When all TX buffers of the mscan are occupied the network layer is told to stop sending further packets. However the TX interrupts are not enabled and therefore the driver has no chance to tell the network layer when it is ready to accept further packets.

Please break lines after 72 chars, or so.

> Signed-off-by: Martin Mosler <mmo@zuehlke.com>
> 
> --- linux-2.6/drivers/net/can/mscan/mscan.c.orig    2011-11-28 13:54:29.547850661 +0100
> +++ linux-2.6/drivers/net/can/mscan/mscan.c 2011-11-28 13:55:52.427849601 +0100
> @@ -214,6 +214,7 @@ static netdev_tx_t mscan_start_xmit(stru
>     case 0:
>         netif_stop_queue(dev);
>         dev_err(dev->dev.parent, "Tx Ring full when queue awake!\n");
> +       out_8(&regs->cantier, priv->tx_active);
>         return NETDEV_TX_BUSY;
>     case 1:
>         /*  

I cannot apply this patch. It is white space mangled (no tabs but
spaces, etc.). Please fix your mail agent or, even better, use "git
send-email".

Wolfgang.
--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Wolfgang Grandegger Nov. 28, 2011, 2:31 p.m. UTC | #6 
On 11/28/2011 03:13 PM, Mosler, Martin wrote:
>>>>> When all TX buffers of the mscan are occupied the network layer is told to stop sending further packets. However the TX interrupts are not enabled and therefore the driver has no chance to tell the network layer when it is ready to accept further packets.
>>>>>
>>>>> Signed-off-by: Martin Mosler <mmo@zuehlke.com>
>>>>>
>>>>> --- linux-2.6/drivers/net/can/mscan/mscan.c.orig    2011-11-28 13:54:29.547850661 +0100
>>>>> +++ linux-2.6/drivers/net/can/mscan/mscan.c 2011-11-28 13:55:52.427849601 +0100
>>>>> @@ -214,6 +214,7 @@ static netdev_tx_t mscan_start_xmit(stru
>>>>>     case 0:
>>>>>         netif_stop_queue(dev);
>>>>>         dev_err(dev->dev.parent, "Tx Ring full when queue awake!\n");
>>>>> +       out_8(&regs->cantier, priv->tx_active);
>>>>>         return NETDEV_TX_BUSY;
>>>>>     case 1:
>>>>>         /*
>>>>
>>>> Hm, did you see the error message. Actually, it should never happen, IIRC.
>>>>
>>>> Wolfgang.
>>>
>>> It is in fact a very rare condition, but it was triggered during testing when pulling CAN-HI and CAN-LO lines
>>> to GND and VCC in various combinations to verify how the system is recovering. I am working with a MPC5125
>>> if this is valuable information. (This was only one issue I ran into, I had to do other modifications for other
>>> issues as well, but I think they are chip specific, I'll share them in a separate thread if you like).
>>
>> Sounds like a problem with bus-off recovery. The software has restarted
>> the queue but the TX objects are still active. How do you handle
>> bus-offs? Manually? restart-ms = ?
>>
>> Would be great if you post patches for your other issues.
>>
>> Wolfgang.
> 
> We are using restart-ms = 500 for now.

OK. How the hardware recovers on bus-off on the MPC51xx and especially
on the MPC51xx is still a mystery to me. I think the real problem is
that the software does not wait until the hardware has fully recovered
from bus-off by going to error active. Do you have a kernel log (dmesg
output) with CONFIG_CAN_DEBUG_DEVICES=y and CONFIG_PRINTK_TIME=y?

> I'll post the other patches after I know I did the submit process correctly for this one and when I reviewed 
> them more closely, as it is my first submission to the kernel.

I just commented on that.

Thanks.

Wolfgang.
--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
diff mbox

Patch

--- linux-2.6/drivers/net/can/mscan/mscan.c.orig    2011-11-28 13:54:29.547850661 +0100
+++ linux-2.6/drivers/net/can/mscan/mscan.c 2011-11-28 13:55:52.427849601 +0100
@@ -214,6 +214,7 @@  static netdev_tx_t mscan_start_xmit(stru
    case 0:
        netif_stop_queue(dev);
        dev_err(dev->dev.parent, "Tx Ring full when queue awake!\n");
+       out_8(&regs->cantier, priv->tx_active);
        return NETDEV_TX_BUSY;
    case 1:
        /*