Re: RTDM Kernel Driver exchanging with Linux Application via /dev/rtp0/ using RTIPC Protocol and XDDP socket

[Jan Kiszka <jan.kiszka@siemens.com>]

On 09.05.20 02:59, ayaida marwane wrote:
> 
> 
>> Le 5 mai 2020 à 21:22, Jan Kiszka <jan.kiszka@siemens.com 
>> <mailto:jan.kiszka@siemens.com>> a écrit :
>>
>> On 29.04.20 23:21, ayaida marwane via Xenomai wrote:
>>> Dear all,
>>> I am trying to develop a simple example of a RTDM driver running from 
>>> Xenomai Kernel, that exchanges a string ("Hello world!") with a Linux 
>>> Application using /dev/rtp0.
>>> I already succeeded doing this with Xenomai 2 using RT_PIPE. However, 
>>> as I understand, this using RT_PIPE from the Kernel depreciated in 
>>> Xenomai 3 and it is recommended to use RTIPC Protocol and a XDDP socket.
>>> In the documentation, there are some examples (xddp-echo.c, 
>>> xddp-label.c and xddp-stream.c), that worked well in my Xenomai 3.0.7 
>>> installed on a Raspberry Pi3.
>>> However, these examples are using a simple socket Protocol (socket(), 
>>> bind(), sendto() and recvfrom()) using the POSIX Skin. So, using this 
>>> in a RTDM module, will surely not work.
>>> Therefore, as I understand, I have to use the Real-time IPC defined 
>>> in the RTDM Skin  (socket__AF_RTIPC(), bind__AF_RTIPC(), 
>>> sendmsg__AF_RTIPC() and recvmsg__AF_RTIPC()).
>>
>> These calls do not exist (where did you find them?) as your build 
>> error also says. Check the inter-driver API for such a stacking 
>> scenario. It provides rtdm_socket, rtdm_sendmsg etc.
>>
>> Depending on what kind of Linux application shall use the API in the 
>> end, it can be simpler to compile that for Xenomai and use the RTDM 
>> userspace API directly to talk to the RTDM driver.
>>
>> Jan
>>
>> -- 
>> Siemens AG, Corporate Technology, CT RDA IOT SES-DE
>> Corporate Competence Center Embedded Linux
> 
> Dear Jan,
> 
> Thank you for your replay. I tried to use the inter-driver API to 
> communicate between the RTDM driver and the Linux side using /dev/rtp0. 
> The exemple is given below (inspired from xddp-echo.c):
> 
> ——————————————————————————————————————————————
> 
> #include <rtdm/driver.h>
> #include <rtdm/ipc.h>
> 
> #define XDDP_PORT 0     /* [0..CONFIG-XENO_OPT_PIPE_NRDEV - 1] and 
> /dev/rtpX  */
> 
> static const char *msg = "Hello world!";
> 
> static int __init my_module_init (void)
> {
>      struct sockaddr_ipc saddr;
>      int ret, s, len;
>      size_t poolsz;
>      char buf[128];
> 
>      rtdm_printk(KERN_INFO "%s.%s()\n", THIS_MODULE->name, __FUNCTION__);
> 
>      s = rtdm_socket(AF_RTIPC, SOCK_DGRAM, IPCPROTO_XDDP);
>      if (s < 0) {
>          rtdm_printk(KERN_INFO "socket error\n");
>      }
> 
>      poolsz = 16384; /* bytes */
>      ret = rtdm_setsockopt(s, SOL_XDDP, XDDP_POOLSZ,
>               &poolsz, sizeof(poolsz));
>      if (ret)
>          rtdm_printk(KERN_INFO "setsockopt error\n");
> 
>      memset(&saddr, 0, sizeof(saddr));
>      saddr.sipc_family = AF_RTIPC;
>      saddr.sipc_port = XDDP_PORT;
>      ret = rtdm_bind(s, (struct sockaddr *)&saddr, sizeof(saddr));
>      if (ret)
>          rtdm_printk(KERN_INFO "bind error\n");
> 
>      len = strlen(msg);
> 
>      rtdm_printk(KERN_INFO "%s: sent %d bytes, \"%.*s\"\n",  
> __FUNCTION__, len, len, msg);
> 
>      ret = rtdm_sendto(s, msg, len, 0, NULL, 0);
> 
>      if (ret != len)
>          rtdm_printk(KERN_INFO "sendto error = %d \n", ret);
> 
>      ret = rtdm_recvfrom(s, buf, sizeof(buf), 0, NULL, 0);
>      if (ret <= 0)
>          rtdm_printk(KERN_INFO "recvfrom error\n");
> 
>      rtdm_printk(KERN_INFO "%s: ==> Received from Linux %d bytes : 
> %.*s\n", __FUNCTION__, ret, ret, buf);
> 
> return 0;
> }
> 
> static void __exit my_module_exit (void)
> {
> rtdm_printk(KERN_INFO "%s.%s()\n", THIS_MODULE->name, __FUNCTION__);
> }
> module_init(my_module_init);
> module_exit(my_module_exit);
> MODULE_LICENSE("GPL");
> 
> ——————————————————————————————————————————————
> 
> Now, the RTDM driver compiles. However, the problem is that when I 
> execute it, it gives:
> 
> ——————————————————————————————————————————————
> 
> [ 5939.538234] rtdm_xddp.my_module_init()
> [ 5939.538437] my_module_init: sent 12 bytes, "Hello world!"
> [ 5939.538451] sendto error = -38
> [ 5939.538458] recvfrom error
> [ 5939.538468] my_module_init: ==> Received from Linux -38 bytes :
> [ 5944.279224] rtdm_xddp.my_module_exit()
> 
> ——————————————————————————————————————————————
> 
> The sendto fails. When I see the signification of this error is:
> 
> ——————————————————————————————————————————————
> 
> /*
>   * This error code is special: arch syscall entry code will return
>   * -ENOSYS if users try to call a syscall that doesn't exist.  To keep
>   * failures of syscalls that really do exist distinguishable from
>   * failures due to attempts to use a nonexistent syscall, syscall
>   * implementations should refrain from returning -ENOSYS.
>   */
> #define ENOSYS          38      /* Invalid system call number */
> 
> ——————————————————————————————————————————————
> 
> Why it does not work? I found that the parameters of sendto() and 
> rtdm_sendto() are the same.
> 

The init function of a module is running in the context of a non-RT 
Linux task (e.g. that of the modprobe process). You can't call any RTDM 
blocking services from there. Your driver needs to run its RT code paths 
inside a RTDM kernel task or (more commonly) inside the RT handler of 
its all RTDM services and, thus, in RT task context of a Xenomai 
application using it.

Jan

-- 
Siemens AG, Corporate Technology, CT RDA IOT SES-DE
Corporate Competence Center Embedded Linux