Experimental Physics and
Industrial Control System

Mark Rivers <[email protected]> · Fri, 24 May 2013 02:19:39 +0000

Hi Zen,

The modbus driver definitely does not lock the port and never unlock it.

However, it is possible that there is a deadlock, which can occur when there is more than one mutex, and they are not taken/released in the correct order.

The modification I was proposing previously was not to fix a deadlock, but rather to lock the port for even longer in the polling thread.

In order to figure out if this is a deadlock, please type the following iocsh command when your simple test driver is blocking:

epicsMutexShowAll 1

That will list all mutexes that are currently held, and what line of code those mutexes were created in.

Thanks,
Mark

________________________________________
From: Zenon Szalata [[email protected]]
Sent: Monday, May 20, 2013 10:19 PM
To: Mark Rivers
Cc: [email protected]
Subject: Re: More on Modbus

Hi Mark,
I have written a simple test device driver derived directly from
asynPortDriver.  The call to pasynInt32SyncIO->read(pasynUser,&val,tmo)
is blocking forever.  So, it seems that the modbus driver locks the port
and never unlocks it.  You did mentioned the issue with locks in an
earlier email on this topic.  Could you make the modification to the
modbus support module?
Thank you much,
Zen

On 05/18/13 16:38, Mark Rivers wrote:
> Hi Zen,
>
> Actually I don't think I need to expose any data structures in the Modbus driver.  Your motor driver should be able to communicate with the Modbus driver using only the following information:
>
> - The name of the Modbus port
> - The asyn "address" field, which is the offset of the register relative to the first register address for that port
> - The asyn interface (e.g. asynInt32, etc.)
> - The asyn drvInfo field, which you may not need if the default MODBUS_DATA is sufficient
>
> With the above information your motor driver will probably do something like the following:
>
> pasynInt32SyncIO->connect(port, addr, &pasynUser, drvInfo);  // In your driver's constructor
>
> pasynInt32SyncIO->write(pasynUser, value, timeout);   // In your driver's I/O routines
> pasynInt32SyncIO->read(pasynUser, &value, timeout);
>
> With the above scheme you will need a number of pasynUser structures (one for each port/addr pair).  You could also use the pasynInt32SyncIO->writeOnce and readOnce methods, which are less efficient but don't require storing multiple pasynUser structures.
>
> Mark
>
> ________________________________________
> From: Zenon Szalata [[email protected]]
> Sent: Saturday, May 18, 2013 1:55 PM
> To: Mark Rivers
> Cc: [email protected]
> Subject: Re: More on Modbus
>
> Hi Mark,
> This is just a short follow up on my last email.
> After looking at my Pico driver implementation and digging a bit more
> into drvModbusAsyn.c file, it seems to me that you might need to expose
> the data structures that are created for each port and then I should be
> able to call pasynOctetSyncIO routines as needed. Is that approximately
> what you had in mind?
> Thanks for helping me with this,
> Zen
>
> P.S. At this point I have a reasonably well working implementation, all
> in a collection of records in a database, about 1000 of them. The only
> problem is that one has to pay attention to reads and writes to the
> hidden registers as these sometimes go wrong.
> With that said, some people here are getting somewhat excited, because
> Beckhoff offers a very inexpensive hardware solution compared to XPS.
> And the XPS controllers are not problem free.
>
>
> On 05/18/13 08:45, Mark Rivers wrote:
>> Hi Zen,
>>
>> Hi Zen,
>>
>> I think I see a solution that should allow you to stop the poller thread when you want to access the same Modbus addresses in another thread.  The solution is to have the poller thread take the asyn port driver lock (pasynManager->lockPort()) for its operation loop.  Your other thread can then also lock the port, which will block the poller thread while you do a sequence of write and read operations.  This will work fine if your motor code is written as a C/C++ driver.  But if it is just a collection of records in a database I'm not sure how to do it.
>>
>> Since the problematic Beckhoff devices are motor controllers, you could write a real driver in C++ using the asynMotorController and asynMotorAxis base classes, which are themselves derived from asynPortDriver.  There are quite a few examples of relatively simple drivers using that model in the synApps motor module now.  Such drivers are normally used with the synApps motor record, but this is NOT required.  Since the driver uses standard asyn interfaces it is possible to just use standard records to move the motor, control the velocity, read the status, etc.  You lose the "state machine" aspect of the motor record (backlash correction, retries, etc.).  But for a simple controller the standard records may be fine.  In your C++ driver you will then call lockPort() for the underlying Modbus driver when needed.
>>
>> I can make the appropriate minor change to the Modbus driver for you to test if this seems like a reasonable approach to you.
>>
>> Mark
>>
>> ________________________________________
>> From: Zenon Szalata [[email protected]]
>> Sent: Friday, May 17, 2013 4:39 PM
>> To: Mark Rivers; [email protected]
>> Subject: More on Modbus
>>
>> Hi Mark,
>> Some of what I wrote in my previous email is no longer relevant.  I have
>> since played with controlling the polling timeout to reduce the
>> interference between the reads done in the polling thread and the
>> write-reads that I am attempting from my rather complicated set of EPICS
>> records.  I have reached a conclusion, hopefully correct, that what I
>> really need is a Modbus device driver that would give me a full control
>> when and in which order register read and write operations are performed.
>> Is this possible with the existing Modbus support module?
>> I am thinking that it would be very nice to have a C++ base class like
>> asynPortDriver, or perhaps a subclass of asynPortDriver, which would
>> implement all the details of Modbus protocol and basic IO.
>> Have you thought of writing something like that?
>> Do you think such a class would be useful?
>> Beckhoff are the only modbus devices that I have written software for
>> and the stepper motor controllers are the only devices for which I find
>> your Modbus support module too limited. For that reason I am a bit
>> hesitant to start a new project, that is to write the C++ class.  Could
>> you offer your insight on this?
>> Thanks in advance,
>> Zen

Experimental Physics and Industrial Control System

Experimental Physics and
Industrial Control System