Linux/PRU-ICSS-INDUSTRIAL-SW: RPMsg timing limitations

Hello Jesse,

I am not aware that TI has conducted any formal testing to see the fastest possible RPMsg communication without data loss. However, this thread indicates that 1ms has been verified by another user. Jason's suggestions on that thread might be helpful for you. One common issue is if the PRU writes RPMsgs to the kernel FIFO faster than the ARM reads from the FIFO, then data will be lost. One of several possible solutions is to only write data with the PRU when user space requests it.

What version of Processor SDK are you using? 

Regards, 

Nick

Also of note, I have been able to verify that the ARM can receive all the data from the PRU by having a read inside of a while loop and counting the number of bytes that I have read. By only sending a certain amount of bytes from the PRU i verify all bytes are received on the ARM, but the problem occurs when I try to write the data to a separate file descriptor and it starts missing chunks of data.

Here is the code I am using to receive:

while(1) {
if(FD_ISSET(fd3,&input_set)){
i = read(fd,&buf2[(bytecount/4)%500],4);
bytecount += i;
if(i >= 1){
if((bytecount%(max_buffer*4))==0) {
fwrite(buf2,max_buffer*4,1,stdout); //If I comment this line of code evrything works perfectly
gettimeofday(&now,NULL);
timersub(&detail_time,&now,&diff);
time2 = (diff.tv_sec+(diff.tv_usec/1000000.0))*1.0;
byte_write+=2000;
fprintf(stderr,"-------%d-------%f\n",byte_write,(diff.tv_sec+(diff.tv_usec/1000000.0))*1.0);
fprintf(stderr,"\nElapsed Time: %f\n\n",(time2-time)*-1.0);
time = time2;
}
}
if(bytecount==88000) break;
}
}


Is this more a linux issue than it is PRU firmware? I would think the ARM which is running at a much faster speed should be able to handle a read and write command in millisecond intervals without taking up too many resources.

Hello Jesse,

fwrite is a blocking function, as well as being expensive in terms of cycles to execute. It looks like this is a Linux related issue where the ARM code is taking so long to go through fwrite, fprintf, etc that it is missing reads. The application implementation is up to you, but I'd suggest

1) decoupling the higher priority read from the lower priority write (ie put fwrite and all those fprintf statements in a separate thread)
2) Look into reducing the number of times expensive functions are called

Are you using RT Linux or regular Linux?

Regards,
Nick