Re: Losing CAN messages with socket-CAN

[Wolfgang Grandegger <wg@grandegger.com>]

Hi Jeroen,

On 04/05/2013 03:31 PM, j.p.lammertink wrote:
> Hi Wolfgang,
> 
> Thanks for your response. Still working just before midnight ...
> Unfortunatelly it doesn't work yet.
> I've collected test results and pasted them below.
> 
> Best Regards,
> 
> Jeroen Lammertink
> 
> --------------------------------------------------------------------------
...
> <>Can you tell me which sources I should look for? (Probably I've downloaded them
> <>already, but I wish to replace only the files that make the difference)
> 
> < I think that patch 1/7 of the series may already fix your TX problems:
> 
> <   http://marc.info/?l=linux-can&m=135526076021515&w=2
> < ´
> < If you apply the remaining patches, you can select the "c_can_pci"
> < driver. If you get trouble with these patches I need to adapt them to a
> < more recent kernel version.
> 
> My colleage has applied the patch you mentioned above. The patching was 
> straightforward without any problems. Also building went fine (->pch_can.ko).
> I've installed the module with:
>     sudo rmmod pch_can
>     sudo insmod pch_can.ko 

What kernel version did you finally use? Could you show your pch_can.c?

> Unfortunatelly, running the test application resulted in an assert. To obtain
> additional information I addapted the function CCanPort::Write() (See below).
> Running gave the following result:
> 
> write returned -1
> errno = 64
> CCanPort.cpp[116]: ASSERT(ret == sizeof(*pFrame)) FAILED 
> 
> The errno 64 is not mentioned in man write
> #define ENONET 64 /* Machine is not on the network */

Hm, I'm also puzzled where this errno comes from? If you send a lot of
messages without delay I would expect an errno ENOBUFS. Could you please
handle the error code properly. In case of an ENOBUF errno just sleep
for a shot time (e.g. usleep(10000)) and then retry.

> At the end of this E-mail I've also included messages of dmesg.
> 
> Do you recognize these errors?

See above.

> Should we continue with the pch_can driver and also apply remaining patches,
> or should we switch to c_can?

Probably yes. I do not really want to care about the pch_can driver.

> <><>--------------------------------------------------------------------------
> 
> <><>size_t CCanPort::Write(can_frame *pFrame)
> <><>{
> <><>    ssize_t ret = write(m_Sock, pFrame, sizeof(*pFrame));
> <><>    ASSERT(ret == sizeof(*pFrame));
> <><>    return ret;
> <><>}
> 
> <><>CCanPort::CCanPort()
> <><>{
> <><>    int rcvbuf_size = BUFFER_SIZE;
> 
> <><>    m_Sock = socket( PF_CAN, SOCK_RAW, CAN_RAW );
> 
> <><>    struct ifreq ifr;
> <><>    strcpy(ifr.ifr_name, "can0");
> <><>    ioctl(m_Sock, SIOCGIFINDEX, &ifr);
> 
> <><>    struct sockaddr_can addr;
> <><>    addr.can_family = AF_CAN;
> <><>    addr.can_ifindex = ifr.ifr_ifindex;
> 
> <><>    // Set receive buffer size
> <><>    setsockopt(m_Sock, SOL_SOCKET, SO_RCVBUF, &rcvbuf_size, sizeof(rcvbuf_size));
> 
> <><>    // Bind
> <><>    bind( m_Sock, (struct sockaddr*)&addr, sizeof(addr) );
> <><>}
> 
> --------------------------------------------------------------------------
> 
> size_t CCanPort::Write(can_frame *pFrame)
> {
>     ssize_t ret = write(m_Sock, pFrame, sizeof(*pFrame));
> 
>     switch(ret)
>     {
>     case 0:
>         std::cout << "write return 0" << std::endl;
>         break;
>     case -1:
>         std::cout << "write returned -1" << std::endl;
>         switch(errno)
>         {
>         case EAGAIN:
>             std::cout << "errno = EAGAIN (=EWOULDBLOCK)" << std::endl;
>             break;
>         case EBADF:
>             std::cout << "errno = EBADF" << std::endl;
>             break;
>         case EDESTADDRREQ:
>             std::cout << "errno = EDESTADDRREQ" << std::endl;
>             break;
>         case EFAULT:
>             std::cout << "errno = EFAULT" << std::endl;
>             break;
>         case EFBIG:
>             std::cout << "errno = EFBIG" << std::endl;
>             break;
>         case EINTR:
>             std::cout << "errno = EINTR" << std::endl;
>             break;
>         case EINVAL:
>             std::cout << "errno = EINVAL" << std::endl;
>             break;
>         case EIO:
>             std::cout << "errno = EIO" << std::endl;
>             break;
>         case ENOSPC:
>             std::cout << "errno = ENOSPC" << std::endl;
>             break;
>         case EPIPE:
>             std::cout << "errno = EPIPE" << std::endl;
>             break;
>         default:
>             std::cout << "errno = " << errno << std::endl;
>             break;
>         }
>         break;
>     default:
>         std::cout << "write returned " << ret << std::endl;
>     };
> 
>     ASSERT(ret == sizeof(*pFrame));
>     return ret;
> }
> 
> --------------------------------------------------------------------------
> 
> bmterra@q7buntu:~/user/jla$ dmesg | grep can
> [    0.181560] [Firmware Bug]: ACPI: No _BQC method, cannot determine initial brightness
> [    1.316862] isapnp: Scanning for PnP cards...
> [    2.885786] rtc_cmos 00:08: RTC can wake from S4
> [   11.119391] pch_can 0000:02:0c.3: irq 47 for MSI/MSI-X
> [   11.119435] pch_can 0000:02:0c.3: (unregistered net_device): PCH CAN opened with MSI
> [   11.119537] pch_can 0000:02:0c.3: setting latency timer to 64
> [  635.622115] can: controller area network core (rev 20090105 abi 8)
> [  635.628256] can: raw protocol (rev 20090105)
> [  660.069395] pch_can 0000:02:0c.3: can0: can_put_echo_skb: BUG! echo_skb is occupied!
> [  660.069561] pch_can 0000:02:0c.3: can0: can_put_echo_skb: BUG! echo_skb is occupied!
> [  660.069583] pch_can 0000:02:0c.3: can0: can_put_echo_skb: BUG! echo_skb is occupied!
> [  660.069602] pch_can 0000:02:0c.3: can0: can_put_echo_skb: BUG! echo_skb is occupied!
> [  660.069622] pch_can 0000:02:0c.3: can0: can_put_echo_skb: BUG! echo_skb is occupied!
> [  660.069641] pch_can 0000:02:0c.3: can0: can_put_echo_skb: BUG! echo_skb is occupied!

These are real problems. Did you see them without the patch as well?

Wolfgang.