Embedding Linux to Control Accelerators and Experiments

The experiments conducted at the ESRF beamlines require precise positioning of many motors to move goniometers, slits, translation stages, etc. Typically, a beamline has more than a hundred stepper motors to align the beamline and move the sample during the experiment (see Figure 5). Motors can be divided into two types: those which can be moved independently and those which have to be moved in sync with the data-taking process. For the independent motors, a TacoBox based on an embedded PC/104-based controller board is ideal—it is compact and can be installed close to the hardware. For the synchronized movements, VME is the preferred solution. It allows us to synchronize the motor movements with the data acquisition via the VME bus without requiring any extra cables.

Figure 5. An example of motor positioning on a beamline—a 6-legged hexapod used to position the crystal in the beam to select only a single wavelength of the beam.

Figure 5

In order to kill two birds with one stone, we chose a stepper motor controller which exists in PC/104 and VME format—the PC68 and VME58 from Oregon Micro Systems. These cards supports step rates of up to 1MHz, very useful for microstepping where the steps are divided by a factor of 1000 to ensure precise positioning. Both cards implement the same ASCII-based controller language. They differ in their register mappings on the bus (one is Intel I/O port-based and the other is m68k memory mapped). The differences are implemented (and hidden) at the level of the device driver. This ensures the TACO device server is identical for both cards. We were fortunate enough to get outside help from Richard Hirst (again) to write the device driver, which is based on an initial version we found on the OMS web site written by Tony Denault of the Institute of Astronomy, Hawaii. The driver implements ioctl calls to read position and status and to pass commands to the board. A multi-threaded device server which supports events was developed at the ESRF (based on an initial version by Lucile Roussier of Lure Laboratory in Paris). The server uses POSIX threads on Linux/m68k and Linux/x86. All the software (device driver and server) is available under the GPL from our FTP site at ftp://ftp.esrf.fr/pub/cs/ess/linux/drivers/oms/

Figure 6. Paolo and Andy in their bug-fighting gear with the stepper motor TacoBox (second shelf from the top on the right-hand side) in its rack on the ID27 beamline clean room. The other boxes are the stepper motor power drivers and the TacoBox power supply.

Figure 6

The PC/104 controller consists of a JumpTEC 486 CPU board and the OMS PC68 controller stacked together (see Figure 6). The TacoBox boots Linux from flashdisk, loads the device driver module, creates as many device descriptors as needed (using major number 26), then starts the device server. On VME, an extra step is needed to program the Vmechip2 according to the ESRF addressing standard.

Once the device server is running, the client requests the server to move the motors. A minor problem we had was ensuring that the PC68 card did not clash with any I/O address already in use, e.g., the address of the network card. This is easy to determine by listing (using cat) the /proc/ioports file and choosing one that is not attributed. A more serious problem was with the VME version of the card and the way it handles interrupts when hitting a limit switch. In the end, we abandoned using interrupts with limits; we simply read the status register to find out if a limit had been hit.

For the stepper motor-based TacoBox, the usual PC/104 cabling problems are reduced somewhat because there is only a single flat cable for the motor controller and an Ethernet cable to connect. The entire box cost us approximately 2000 euros for four stepper motor channels and 3000 euros for eight channels (using the PC68 extension board). In our application, we use the PC68 as a simple stepper motor controller; we haven't tried the other features yet (e.g., support for relative encoders, dc motors and servo loops).

Today, the main drawback of Linux for our applications is still the lack of device drivers for many kinds of I/O boards. PC boards are better supported than VME, but still lag far behind Windows drivers. Our aim is to install as many Linux-based controllers as possible because of their better stability, ease of programming, flexibility, lower cost, etc. In order to achieve this, we need drivers, drivers and yet more drivers.

We are only a small team working on device drivers and are therefore very interested in collaborating with other programmers on device drivers for all kinds of boards. All our drivers are developed under GPL and made available to the external world on our FTP site.

We would be interested in hearing from anyone who has written a device driver for an I/O board for PC/104 or VME. Please send e-mail to one of the authors to add your driver or name to our database, which we will make available on our web site.