Speaker
Description
Summary
The data acquisition system of the ATLAS experiment contains several large, VMEbus
based sub-systems. There will be about 120 9U VMEbus crates to house the Read-Out
Drivers. In addition there will be of the order of 30 6U crates to house the
modules of the Trigger and Timing Control system. Finally there is a large number
of VMEbus based test benches both at CERN and in the collaborating institutes.
It was felt necessary to prevent the groups responsible for the individual sub-
systems from selecting their favorite VMEbus Single Board Computer (SBC) by
standardizing on a single type of SBC for all installations.
In 2001 a small group of experts was formed to specify the requirements of a common
SBC both in terms of its technical features as well as the support which had to be
guaranteed by the manufacturer for at least nine years. At the same time a number
of VMEbus SBCs from different vendors was evaluated to identify potential problems
and to better understand the advantages and disadvantages of the different SBC
architectures. To avoid limiting the competition too much the final specification
did allow for a wide range of architectures (i.e. PowerPC and Pentium, all types of
VMEbus master interfaces). The SBC that was finally selected is based on a Pentium
CPU and the Tundra Universe PCI to VMEbus interface.
The evaluation of this and similar SBCs has identified a number of potential
problems with the Tundra Universe chip:
- It has no built in endian conversion. In case of a little endian CPU (e.g.
Pentium) this feature has to be provided by some vendor specific auxiliary logic.
- Moderate performance
o D32 R/W = 4 MB/s
o D32 posted W = 13 MB/s
o D32 BLT = 20 MB/s
o D64 MBLT = 40 MB/s
- Difficult bus error handling
- No true constant address DMA for reading out FIFOs
- Problems with BERR terminated BLTs
Depending on the PCI host bridge used on the SBC problems in the form of VMEbus
error can arise if the SBC has to deal with master and slave accesses at the same
time.
In parallel with the H/W the options for the VMEbus access S/W were analyzed.
Despite the effort made by the VISION standard a few years ago there is no widely
accepted standard API for access to the VMEbus. A number of free and commercial
drivers exist for the Tundra Universe chip but they are not compatible. Finally the
decision was made to program a dedicated Linux VMEbus driver and library for ATLAS.
This gives full control over the code and the API and allows optimizing the S/W for
the particular needs of ATLAS. The driver and library provide two ways of doing
single cycles, extensive support for chained block transfers, bus error handling
and the possibility to handle interrupts both synchronously and asynchronously.
Less frequently used features such a support for SYSFAIL interrupts or special AM
codes for geographical addressing (VME64x) have been added as well.
