SOEM/doc/tutorial.txt

309 lines
10 KiB
Plaintext

/** \file
\section general General
The SOEM is a library that provides the user application with the means to send
and receive EtherCAT frames. It is up to the application to provide means for:
- Reading and writing process data to be sent/received by SOEM
- Keeping local IO data synchronised with the global IOmap
- Detecting errors reported by SOEM
- Managing errors reported by SOEM
The following sections show some basic examples on how to get the SOEM up
and running, as well as making use of the process data and checking
for errors. Since all code is local to the application or global
variables, it is possible to tweak and optimize when possible.
The following example shows how to add a main function that will be
called by startup code. In this example main's only purpose is to
spawn a new task that executes SOEM.
\code
int main (void)
{
rprintp("SOEM (Simple Open EtherCAT Master)\nSimple test\n");
task_spawn ("simpletest", simpletest, 9, 8192, NULL);
\endcode
\section configuration Configuration
Followed by start of the application we need to set up the NIC to be used as
EtherCAT Ethernet interface. In a simple setup we call ec_init(ifname) and
if SOEM comes with support for cable redundancy we call ec_init_redundant
that will open a second port as backup. You can send NULL as ifname if you
have a dedicated NIC selected in the nicdrv.c. It returns >0 if succeeded.
\code
/* initialise SOEM, bind socket to ifname */
if (ec_init(ifname) > 0)
\endcode
SOEM is a light weight ethercat master library used in embedded systems, It
supports only runtime configuration. It requests a BRD (Broad Cast Read) of
address 0, all fully functional slaves in the network will respond to this
request, and therefore we will get a working counter equal to the number of
slaves in the network. ec_config_init also sets up the mailboxes for slaves
that support it. When ec_config_init finishes it will have requested all slaves
to state PRE_OP. All data read and configured are stored in a global array
which acts as a placeholder for key values, consult ec_slave for detailed
information.
\code
/* find and auto-config slaves */
if ( ec_config_init(FALSE) > 0 )
{
rprintp("%d slaves found and configured.\n",ec_slavecount);
\endcode
SOEM has now discovered and configured the network it is connected to.
Now we can verify that all slaves are present as expected. These
definitions could be generated by an external tool in an offline .h file.
The definitions could be replaced by a struct keeping slave number.
\code
#define EK1100_1 1
#define EL4001_1 2
...
#define EL2622_3 8
#define EL2622_4 9
#define NUMBER_OF_SLAVES 9
snippet
...
uint32 network_configuration(void)
{
/* Do we got expected number of slaves from config */
if (ec_slavecount < NUMBER_OF_SLAVES)
return 0;
/* Verify slave by slave that it is correct*/
if (strcmp(ec_slave[EK1100_1].name,"EK1100"))
return 0;
else if (strcmp(ec_slave[EL4001_1].name,"EL4001"))
return 0;
...
else if (strcmp(ec_slave[EL2622_4].name,"EL2622"))
return 0;
return 1;
}
simpletest
...
if (network_configuration())
...
else
rprintp("Mismatch of network units!\n");
\endcode
We now have the network up and configured. Mailboxes are up for slaves that support
it. Next we will create an IOmap and configure the SyncManager's and
FMMU's to link the EtherCAT master and the slaves. The IO mapping is done
automatically, SOEM strives to keep the logical process image as compact as
possible. It is done by trying to fit Bit oriented slaves together in single
bytes. Below is an example of 8 slaves and how they are ordered. During
mapping SOEM also calculates an expected WKC for the IO mapped together.
That is the primary key to detect errors.
- Outputs are placed together in the beginning of the IOmap
- Inputs follow
When the mapping is done SOEM requests slaves to enter SAFE_OP.
\code
char IOmap[128];
int usedmem;
...
usedmem = ec_config_map(&IOmap);
if (usedmem <= sizeof(IOmap))
...
\endcode
\image html memory_layout.png "memory layout, mapping between physical and logical"
\image latex memory_layout.png "memory layout, mapping between physical and logical" width=15cm
To enter state OP we need to send valid data to outputs. The EtherCAT frame
handling is split into ec_send_processdata and ec_receive_processdata.
- ec_send_processdata sends the frame on the NIC and saves the frame on
the stack for receive to fetch.
- ec_receive_processdata(EC_TIMEOUTRET) tries to fetch the frames on the
stack. We send an argument for how long we will try to fetch the frame.
ec_receive_processdata returns the working counter.
\code
/* send one valid process data to make outputs in slaves happy*/
ec_send_processdata();
wkc = ec_receive_processdata(EC_TIMEOUTRET);
...
ec_writestate(0);
/* wait for all slaves to reach OP state */
ec_statecheck(0, EC_STATE_OPERATIONAL, EC_TIMEOUTSTATE);
\endcode
- Now we have a system up and running, all slaves are in state operational.
\section application Application
IO data is accessed through the IOmap, the ec_slave struct keep pointers
to the start byte in the IO map on slave level together with start bit within
the start byte. This way we can bit mask IO on bit level even though SOEM
has combined slave data to minimize the frame size to be sent. We'll use
slave 8 in the picture above as an example. From a printout from ec_slave we
have the following:
- Slave:8
- Name:EL2622
- Output size: 2bits
- Input size: 0bits
- Configured address: 1008
- Outputs address: 18cf6
- Inputs address: 0
- FMMU0 Ls:2 Ll: 1 Lsb:4 Leb:5 Ps:f00 Psb:0 Ty:2 Act:1
The Outputs address: 18cf6 is the pointer to slave 8's start byte. The FMMU's
Lsb:4 (LogicalStartBit) = ec_slave.Ostartbit telling us how to mask for the
individual bits in the combined byte. The same goes for byte addressed slaves,
but byte slaves only need the byte start address since they are byte aligned,
the start bit will be 0.
Some example on how to access different types of data
Set an output int 16 value when memory alignment needs to be considered,
arguments is:
- slave number in ethercat network
- module index as index internal to the slave in case more than one
channel
- value to write
\code
#define EL4001_1 2
...
void set_output_int16 (uint16 slave_no, uint8 module_index, int16 value)
{
uint8 *data_ptr;
data_ptr = ec_slave[slave_no].outputs;
/* Move pointer to correct module index*/
data_ptr += module_index * 2;
/* Read value byte by byte since all targets can't handle misaligned
* addresses
*/
*data_ptr++ = (value >> 0) & 0xFF;
*data_ptr++ = (value >> 8) & 0xFF;
}
...
set_output_int16(EL4001_1,0,slave_EL4001_1.out1);
\endcode
Target can handle non aligned pointers to the IOmap
\code
typedef struct PACKED
{
int16 outvalue1;
int16 outvalue2;
} out_EL4132t;
out_EL4132t *out_EL4132;
...
/* connect struct pointers to slave I/O pointers */
out_EL4132 = (out_EL4132t*) ec_slave[3].outputs;
out_EL4132->outvalue2 = 0x3FFF;
...
\endcode
Identify and manage errors. The key is the Working Counter, CRC errors
and errors local to the slave causing a state change can be detected by loss
of Working Counter since the syncmanagers won't get updated. When returning
Working Counter don't match Expected Working Counter something is wrong, then it
is up to an error handler to act, locate the erroneous slave and decide what action
to perform. The error may not be fatal. Some basic code from simple_test.
\code
wkc = ec_receive_processdata(EC_TIMEOUTRET);
expectedWKC = (ec_group[0].outputsWKC * 2) + ec_group[0].inputsWKC;
if( inOP && ((wkc < expectedWKC) || ec_group[currentgroup].docheckstate))
{
if (needlf)
{
needlf = FALSE;
printf("\n");
}
/* one ore more slaves are not responding */
ec_group[currentgroup].docheckstate = FALSE;
ec_readstate();
for (slave = 1; slave <= ec_slavecount; slave++)
{
if ((ec_slave[slave].group == currentgroup) && (ec_slave[slave].state != EC_STATE_OPERATIONAL))
{
ec_group[currentgroup].docheckstate = TRUE;
if (ec_slave[slave].state == (EC_STATE_SAFE_OP + EC_STATE_ERROR))
{
printf("ERROR : slave %d is in SAFE_OP + ERROR, attempting ack.\n", slave);
ec_slave[slave].state = (EC_STATE_SAFE_OP + EC_STATE_ACK);
ec_writestate(slave);
}
else if(ec_slave[slave].state == EC_STATE_SAFE_OP)
{
printf("WARNING : slave %d is in SAFE_OP, change to OPERATIONAL.\n", slave);
ec_slave[slave].state = EC_STATE_OPERATIONAL;
ec_writestate(slave);
}
else if(ec_slave[slave].state > EC_STATE_NONE)
{
if (ec_reconfig_slave(slave, EC_TIMEOUTMON))
{
ec_slave[slave].islost = FALSE;
printf("MESSAGE : slave %d reconfigured\n",slave);
}
}
else if(!ec_slave[slave].islost)
{
/* re-check state */
ec_statecheck(slave, EC_STATE_OPERATIONAL, EC_TIMEOUTRET);
if (ec_slave[slave].state == EC_STATE_NONE)
{
ec_slave[slave].islost = TRUE;
printf("ERROR : slave %d lost\n",slave);
}
}
}
if (ec_slave[slave].islost)
{
if(ec_slave[slave].state == EC_STATE_NONE)
{
if (ec_recover_slave(slave, EC_TIMEOUTMON))
{
ec_slave[slave].islost = FALSE;
printf("MESSAGE : slave %d recovered\n",slave);
}
}
else
{
ec_slave[slave].islost = FALSE;
printf("MESSAGE : slave %d found\n",slave);
}
}
}
if(!ec_group[currentgroup].docheckstate)
printf("OK : all slaves resumed OPERATIONAL.\n");
}
\endcode
This tutorial is just one way of doing it.
Enjoy and happy coding!
Andreas Karlsson, rt-labs AB, www.rt-labs.com
*/