To use the SCI I/O model properly on series 5000 chips, perform the following tasks:
- Use the 
sci_in_request_ex()function instead of theio_in_request()function. The sci_in_request_ex() function has different parameters from the io_in_request()function. sci_in_request_ex() performs an interrupt-safe switch to the buffer & count passed in, and returns the count of bytes in the RX buffer during that interrupt-safe code section when the switch occurs.
 - Use a "double buffering" technique to switch between two different incoming data buffer so that you will not miss any data.
 
The following Neuron C code example demonstrates steps 2 and 3, above:
// pragmas
#pragma num_alias_table_entries 0
#pragma specify_io_clock "10 MHz" 
// SCI_4800 etc., are declared in this header file
#include <io_types.h>
// SCI declaration
IO_8 sci baud(SCI_115200) iosci; 
// Buffers and variable declarations
char szRxBufA[50];
char szRxBufB[50];
unsigned short rx_count;
int bufTog;
// Function declaration
void processRxData(char buffer[50], unsigned short rx_count);
when(io_in_ready(iosci))
{
 // Switch to the other buffer and keep the RX count
 rx_count = sci_in_request_ex((bufTog) ? szRxBufA : szRxBufB, sizeof(szRxBufA));
 bufTog ^= 1; // toggle this
 
 // Do something with RX data using the filled data buffer. 
 // It may be only a few bytes, or it may be close to 50 bytes. 
 processRxData((bufTog) ? szRxBufA : szRxBufB, rx_count);
}
when (reset)
{
 (void)io_in_request(iosci, szRxBufA, 50);
 bufTog = 0;  // indicates that A is in use.
}
void processRxData(char buffer[50], unsigned short rx_count)
{
 io_out_request(iosci, buffer, rx_count);
}