While similar to the UNIBUS both at a high level, in that it supported both memory and devices through read-write cycles, DMA, and interrupts, as well as in much of the low-level detail, such as being entirely asynchronous, and using the same drivers for its wired-OR transmission lines, it differed in a number of ways.
For example, one way was that it used multiplexed address and data lines (to reduce the pin count), as opposed to the separate address and data lines of the UNIBUS. Another was that although it also supported 4 levels of interrupt priority, that support was optional, and used a more complex signalling mechanism, with only a single shared grant line, to do so.
Like the UNIBUS, there are three basic kinds of cycles on the QBUS: data read/write cycles (in which a 'master' reads or writes data to/from a 'slave', which is usually, but not always, memory); DMA cycles (in which a device gains control of the bus so that it can do a read/write cycle); and interrupt cycles, in which a device causes the CPU to perform an interrupt.
Also like the UNIBUS, all QBUS transactions are asynchronous, and use interlocked request/response signals for control and timing; also, most QBUS signal lines are electrically bi-directional transmission lines (even though some of these are logically uni-directional); only the grant lines (BDMG and BIAK) are physically uni-directional, and are wired in a daisy-chain fashion.
Read/write cycles come in the same basic forms as on the UNIBUS: DATI for word reads, DATO for word writes, DATIO for word read-modify-write cycles, and DATOB and DATIOB for byte write/R-M-W cycles. On the UNIBUS, however, two control lines coded the cycle type; on the QBUS, discrete control lines exist for each type of cycle (BDIN, BDOUT, and BWTBT).
The QBUS later added block transfer modes, DATBI and DATBO; only later memory and devices support this mode. The QBUS signal BREF, used for external refresh of MOS memory, which had by that time fallen into desuetude, was re-purposed to allow a memory to signal that it supported block mode.
On the QBUS, multi-level priority interrupts share a single grant line; to do this, interrupt-requesting devices must monitor the higher-priority request lines, and refrain from intercepting a grant if there is a higher-priority request pending. Early QBUS devices did not implement this multi-level priority scheme.
QBUS backplanes come in two types, dual and quad. The QBUS itself is fully carried in a dual slot, so the latter are further sub-divided into two types, the so-called Q/Q and Q/CD.
In quad Q/Q backplanes, both sides of each quad slot are fully wired for QBUS, and so a single slot can hold two separate dual-width QBUS devices. The device locations are usually arranged for grant priority in so-called 'serpentine' order, i.e. one with the devices in the following kind of order (facing the backplane from the board side):
In a quad Q/CD backplane, the CD connectors form a private bus, sometimes called the CD interconnect, used to connect together board pairs. (The CD connectors run down the right-hand side, when facing the side of the backplane where the boards plug in, with the CPU at the top.)
NOTE WELL: For reasons which seem utterly incomprehensible, many boards designed for Q/CD slots (such as PMI cards) do not avoid the QBUS pins on the CD connectors which contain 'hazardous' (to TTL circuitry) voltages. Plugging such a card into a Q/Q backplane will generally destroy the card.
Variable address size
The QBUS was available in 16-, 18-, and 22-address-bit configurations (data width remained 16 bits in all three versions). The three versions are often referred to as Q16, Q18 and Q22.
CPUs, devices and backplanes all are one of the three alternatives; for instance, the earliest CPU, the LSI-11, is a Q16 device. Mixing cards and backplanes of differing address widths may, or may not, work; or may work, but with limitataions - and may sometimes initially appear to work, but, when examined carefully, not work.
Important note: The 16-bit and 18/22-bit backplanes are electrically incompatible and mixing the two may damage cards on the bus.
One example of the kind of limitation that may occur happens when using a Q18 DMA device in a Q22 system. The device will function correctly, but can only do transfers to the lower 256KB of memory; software that uses this device will have to work around that limitation.
An example of something that looks like it might work, but does not in fact work, is mixing Q18 and Q22 memory cards in a Q22 system, with more than 256Kbytes of memory in total. The problem is that the Q18 memory card will respond at multiple places in the 22-bit address space; e.g. if a Q18 card is configured at address 0, it will also respond at 1000000.
It is possible to upgrade 18-bit backplanes to 22-bit; see Upgrading QBUS backplanes.
QBUS pins are identified in the standard UNIBUS manner; there are two connectors, A and B; pins on the component side are 1, those on the solder side are 2. Pins are identified by the 'DEC alphabet', A-V, with G, I, O and Q dropped.
Signals marked with a "*" show cases where two signals use the same pin (not at the same time, obviously).
|SSpare2||AF1||alt SRUN/SMENBL on CF1||BRPLY||AF2|
|SSpare3||AH1||alt SRUN on CH1||BDIN||AH2|
References to pin "Cxy" refer to a quad-wide slot (e.g. as used by the original LSI-11 CPU board).