Hello,
I’m trying to get a minimal user space program to communicate bidirectionally with M4_0 on the Apalis.
I have added the following device tree overlay:
Note: For the following problems, I also tried changing the aliases for vdev2/3* to match vdev2/3, e.g. before vdev2vring0: vdev0vring0; after vdev2vring0: vdev2vring0. I also tried removing the @ in the alias for rpmsg0 to match the format of rpmsg1; before rpmsg0: rpmsg@0; after rpmsg0: rpmsg0.
I have built and loaded in uboot the M4 with rpmsg_lite_str_echo_rtos/cm4_core0 from NXP’s SDK 2.9.0 for MIMX8QM6xxxFF.
I modprobe imx_rpmsg_tty and run a minimal program in an attempt to read()/write() on /dev/ttyRPMSG30:
int rpMsgFd = -1;
struct termios tty;
rpMsgFd = open("/dev/ttyRPMSG30", 0, O_RDWR | O_NOCTTY);
if(rpMsgFd < 0) {
return -1;
}
if(tcgetattr(rpMsgFd, &tty) < 0) {
return -2;
}
cfmakeraw(&tty);
if(tcsetattr(rpMsgFd, TCSANOW, &tty) < 0) {
return -3;
}
rpmsgResult = write(rpMsgFd, &outgoingMsg, sizeof(outgoingMsg));
if(rpmsgResult != sizeof(outgoingMsg)) {
std::cerr << "Error sending RPMsg (" << rpmsgResult << ", " << errno << ", " << rpMsgFd << ").\n";
continue;
}
write() fails and returns -1, errno is 9 (bad file descriptor), with a file descriptor of 3. Since this isn’t a real file, I suspect errno may be invalid.
I can successfully use cat and echo, though:
# cat /dev/ttyRPMSG30 &
# echo test > /dev/ttyRPMSG30
test
This minimal example is what I’ve resorted to after trying to use a custom M4 program that uses the bare metal RPMsg library with the following preprocessor symbol modification in rpmsg_config.h:
#define RL_USE_STATIC_API (1)
Performing the same cat/echo above resulted in the M4 replying with a pre-programmed message, but the M4 subsequently received unexpected infinite messages, which it claimed were sourced by the Linux side, whose contents were the original M4->Linux message, but 1-2 bytes at a time, and eventually garbage. These erroneous receptions resulted in more M4->Linux transmissions. The addresses of the received message data seemed to be incrementing by 0x200 in the vdevbuffer memory, which appears to be correct.
I also attempted adding the following symbol to rpmsg_config.h, which was present in the RPMsg API documentation but not rpmsg_config.h itself. The documentation claimed that it was needed in a Linux scenario, but was not used in any of the SDK examples. No change in behavior was observed.
#define RL_ALLOW_CONSUMED_BUFFERS_NOTIFICATION (1)
Aside from using the static API and using my own Rx callback function to store the received data, I’m not sure what’s going wrong here. I will try to incrementally convert the example program into the static with custom callback program to see where things go wrong.
I can provide a minimal version of this M4 program if desired, but I think the write() problem is the first hurdle.
Apalis i.MX8QM 4GB IT v1.1 D
Apalis Evaluation Board Rev. 1.1
Custom Yocto build using toradex-manifest branch kirkstone-6.x.y
