Re: CAN messages being lost on i.MX25 with flexcan

[Martin Kozusky <mkozusky@kkmicro.cz>]

Dne 19.4.2012 15:37, Wolfgang Grandegger napsal(a):
> On 04/19/2012 01:58 PM, Martin Kozusky wrote:
>> Dne 19.4.2012 13:44, Wolfgang Grandegger napsal(a):
>>> On 04/19/2012 01:21 PM, Martin Kozusky wrote:
>>>> Dne 19.4.2012 12:54, Wolfgang Grandegger napsal(a):
>>>>> On 04/19/2012 12:04 PM, Martin Kozusky wrote:
>>>>>> Hello,
>>>>>> I'm using Voipac i.MX25 module with flexcan, kernel 2.6.38.9.
>>>>>> I'm sending the data at 250kbps, around 1100 msgs/sec. When I
>>>>>> enable the
>>>>>> canbus interface (canconfig can0 start), CPU load is higher, that is
>>>>>> understandable, there are many interrupts. I'm not doing anything else
>>>>>> then using recvmsg (or recvmmsg which is little better), but some
>>>>>> messages are still lost (around 1500 messages lost from 467 000 being
>>>>>> send from another source). When I start doing something (like  "cat
>>>>>> /proc/interrupts",  or write to file), many more messages are lost.
>>>>>>
>>>>>> Do you have any idea how to fix this? I need to make some CAN messages
>>>>>> logger and I cannot lose any message (idealy :) So I made big
>>>>>> buffer in
>>>>>> my program so that I don't need to write the messages into the file
>>>>>> while "recording" is enabled, after "recording" is switched off, I
>>>>>> write
>>>>>> the buffer into the file, but that is still not good enough.
>>>>>> Is there any way how to write to some buffer directly in flexcan
>>>>>> driver
>>>>>> (the best would be in the IRQ routine) and then read messages from
>>>>>> this
>>>>>> buffer in my program?
>>>>>> Or are just interrupts lost when doing something else in the system
>>>>>> and
>>>>>> I cannot fix this? Or can I somehow specify that "can rx
>>>>>> interrupts" has
>>>>>> highest prioroty?
>>>>>
>>>>> Do you already know where the packets are dropped (get lost). Maybe
>>>>> your
>>>>> user space app is simply not faster enough to process them in time. You
>>>>> can check that with the candump option "-d".
>>>>
>>>> Hello Wolfgang,
>>>> I don't know where it gets lost. But I'm just making select() and
>>>> recvmmsg in my app now.
>>>> Now when I'm trying candump -d, it shows messages and between that:
>>>>
>>>>     can0  18FFC33D  [8] 4A E6 00 00 03 00 00 F6
>>>>     can0  1CEFB8F3  [8] 34 42 23 3D 8C 01 00 00
>>>> DROPCOUNT: dropped 245 CAN frames on 'can0' socket (total drops 30886)
>>>>     can0  18FF0AEF  [8] 70 76 46 7C A7 9E 2A 6A
>>>>     can0  18FF0AF0  [8] B0 7E FB 7D A7 DC 27 69
>>>>     can0  18FEBF0B  [8] F5 03 79 81 69 74 6E 79
>>>>
>>>> I've sent around 40 000 messages.
>>>> Does this help somehow to identify where the problem is?
>>>
>>> Yes, you are loosing messages because your app is not faster enough.
>>>
>>>> I found out that if I (in my program) don't printf every frame, then
>>>> dropcount is better, so I think when candump would only show dropps and
>>>> not every frame then it would also get better.
>>>
>>> printf takes time, of course. Therefore everything which makes your app
>>> reading the messages faster will help including increasing the
>>> scheduling priority of the process/thread. Also increasing the size of
>>> the receive buffer may help. Try candump with the "-r" option.
>>
>> I tried -r 10000000
>>
>> and the result was that dropping started some time later then it would
>> normaly. From 467 000 packets, total drops was 103064. That is better
>> ratio than before. But few times, my system was stucked for a while,
>> then it recovered and continued to work. So I will try to use this
>> buffer size in my program, without printf and I will see how many
>> packets I lose.
>
> If other activities delay your app too much, your are in trouble. Proper
> thread priorities may help.

So I tried my app with bigger receive buffer and saving data into memory 
and it looks like no messages are lost now :) (but when I write them 
into file, they are lost, but I can use the temp memory storage, so I 
can live with that :)

>> Is there a way how to get actual used size of the buffer?
>
> getsockopt?
>
>>>> This is my /proc/interrupts
>>>>
>>>> root@vmx25 /opt$ cat /proc/interrupts
>>>>              CPU0
>>>>     0:          0           -  spi_imx
>>>>     5:      32827           -  IMX-uart
>>>>     9:        129           -  sdhci
>>>>    13:          0           -  spi_imx
>>>>    25:          5           -  imxdi_rtc
>>>>    33:      12168           -  mxc_nand
>>>>    34:          1           -  mxc-sdma
>>>>    37:     109275           -  ehci_hcd:usb1
>>>>    43:    1624971           -  can0
>>>>    46:          0           -  mxcadc
>>>>    54:     195357           -  i.MX Timer Tick
>>>> 164:          0           -  ESDHCI card 0 detect
>>>> Err:          0
>>>
>>> In a first place, I don't think your are loosing messages because they
>>> are not read-out faster enough from the CAN controller.
>>>
>>>> I have another problem - when I send messages and nobody is acking them
>>>> (no other device on CANbus), system stops responding until I put another
>>>> device onto CANbus, but I think I should start new thread with this.
>>>
>>> What kernel version do you use? Do you use the flexcan driver from that
>>> kernel? The problem is bus errors coming at high rate.
>>
>> I'm using 2.6.35.9 Preempt, the flexcan.c driver is original.
>
> IIRC, that kernel version did not include the flexcan driver. It entered
> with 2.6.36. Anyway, maybe the following patch helps:
>
>    http://patchwork.ozlabs.org/patch/130761/

It seems that it helped :) when nobody is ACKing, it doesn't hang.

Thank you,
Martin

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