Re: CM-ITC, pch_can/c_can_pci, sendto() returning ENOBUFS

[Jacob Kroon <jacob.kroon@gmail.com>]

Hi Oliver,

On 8/29/22 15:53, Oliver Hartkopp wrote:
> Hi Jacob,
> 
> On 29.08.22 15:20, Jacob Kroon wrote:
>> Hi Wolfgang and Marc,
>>
>> On 8/29/22 11:14, Jacob Kroon wrote:
>>> On 8/26/22 13:24, Jacob Kroon wrote:
>>>> On 8/25/22 15:25, Jacob Kroon wrote:
>>>>> Hi,
>>>>>
>>>>> I am using a CM-ITC board 
>>>>> (https://www.compulab.com/products/computer-on-modules/cm-itc/) 
>>>>> with an application that uses the CAN interface. After a while of 
>>>>> successfully sending packets, sendto() starts returning ENOBUFS. I 
>>>>> wait a whole second and try to send, several retries, but I get 
>>>>> ENOBUFS every time. I'm using kernel 5.15.59, and I've tried both 
>>>>> the pch_can and c_can_pci driver, but both show the same error.
>>>>>
>>>>> In the console I see several of:
>>>>> can0: can_put_echo_skb: BUG! echo_skb 0 is occupied
>>>>>
>>>>> I've also tried to increase the txqueuelen to 1000, as suggested here
>>>>>
>>>>> https://stackoverflow.com/questions/40424433/write-no-buffer-space-available-socket-can-linux-can
>>>>>
>>>>> but I think that if I increase the queuelen the threads just block 
>>>>> forever in sendto() (sockets are opened in blocking mode)
>>>>>
>>>>> If I bring down the interface with
>>>>>
>>>>> ifconfig can0 down
>>>>> ifconfig can0 up
>>>>>
>>>>> the transmitting does get unblocked.
>>>>>
>>>>> Is there anything I can do to debug this further ? Any other ideas ?
>>>>>
>>>>
>>>> First I get the print:
>>>>
>>>> can_put_echo_skb: BUG! echo_skb 0 is occupied!
>>>>
>>>> then netif_stop_queue() is called from here:
>>>>
>>>> https://github.com/torvalds/linux/blob/master/drivers/net/can/c_can/c_can_main.c#L469
>>>>
>>>> and then there is no call to netif_start_queue(), so the bus hangs.
>>>>
>>>
>>> Switching back to the pch_can driver. I'm guessing here but it would 
>>> seem that the driver is not receiving the TX interrupt for 
>>> 'PCH_TX_OBJ_END' that would wake up the netif queue, since with the 
>>> changes below I can't reproduce the hang, although I'm still seeing a 
>>> lot of "echo_skb <x> is occupied!":
>>>
>>>> diff --git a/drivers/net/can/pch_can.c b/drivers/net/can/pch_can.c
>>>> index 964c8a09226a..75ad2272d9b2 100644
>>>> --- a/drivers/net/can/pch_can.c
>>>> +++ b/drivers/net/can/pch_can.c
>>>> @@ -719,8 +720,7 @@ static void pch_can_tx_complete(struct 
>>>> net_device *ndev, u32 int_stat)
>>>>                           PCH_IF_MCONT_DLC);
>>>>         stats->tx_bytes += dlc;
>>>>         stats->tx_packets++;
>>>> -       if (int_stat == PCH_TX_OBJ_END)
>>>> -               netif_wake_queue(ndev);
>>>> +       netif_wake_queue(ndev);
>>>>  }
>>>
>>
>> Is there a recommended driver for the Intel EG20T PCH CAN controller ? 
>> pch_can or c_can_pci ?
> 
> I'm very happy that you show up with this question!
> 
> AFAIK the PCH_CAN driver was contributed by some OKI developers on 
> behalf of Intel to support their EG20T platform. This driver has not 
> been worked on for years now.
> 
> Later the C_CAN driver got its PCI support which should have made the 
> PCH_CAN driver obsolete as they are both using the C_CAN IP core.
> 
> So my suggestion would be to work with the well maintained c_can_pci and 
> mark the pch_can obsolete in the Linux kernel.
> 

Ok.

> IIRC Marc is very experienced with the C_CAN driver but he is still on 
> vacation this week.
> 
> Regarding your problem:
> "BUG! echo_skb 0 is occupied" should never show up but I wonder if you 
> probably have another CAN specific problem with your setup.
> 
> Did you set up a CAN network with at least two CAN nodes, identical 
> bitrate settings and 2x120 Ohms (== 60 Ohms) termination between CAN_L 
> and CAN_H?
> 

I have double checked that both endpoints of the network are terminated 
with 120Ohm resistors, and when I check CAN_H/CAN_L in an oscilloscope 
they look ok, and within the spec. voltage levels.

> Do you have another CAN node which can be attached to the EG20T setup 
> (e.g. some ECU or an USB CAN adapter)?

Yes I do have a CAN analyser from Microchip. I guess I can record all 
traffic with the analyzer, and compare it to what I see with "candump 
can0" on the host. Or do you have some other suggestion ?

Thanks for responding,
Jacob