Reconfigurable Bus

This invention relates to data processing in general and more particularly, to a reconfigurable bus for use in a data processing system, which bus can be configured to either communicate with external memory or external I/O such as an additional processor.

The need for large amounts of memory in modern day processors is well recognized. It is also recognized that it is desirable to provide memory in variable amounts. To accomplish this, additional memory can be provided and accessed by means of a memory bus. In accordance with a system of the type disclosed in the present application, such a memory bus includes a configuration bus which provides information to the computer system concerning the configuration of the memory. Another requirement in computer systems which is becoming increasingly important is the ability to interface with other computer systems. Typically this is done over an I/O bus. In conventional systems, separate hardware is provided to implement separate memory buses and I/O buses. This becomes a particular problem when trying to place all of the necessary hardware on a single board. However, in some instances, both the I/O bus access and memory access are not needed at the same time. Thus, it would be beneficial if a single bus could be used for both purposes, despite the fact that the two types of devices, i.e. memory and I/O device, e.g. another processor have somewhat different requirements.

The present invention provides a solution to this problem by providing a reconfigurable bus. In accordance with the present invention, there is provided a memory bus which includes, in addition to address and data lines, a configuration bus, having 8 bits in the illustrated embodiment. When the external memory bus is utilized as a memory bus, the configuration bus is a unidirectional bus providing information concerning memory configuration back to the computer system through a memory controller. However, in accordance with the present invention, when utilized as an I/O bus, the configuration bus is reconfigured to provide both input and output, i.e. to become a bus with inputs and outputs to provide the necessary additional functionality needed for I/O operations, particularly operations in involving another processor and memory.

In accordance with the present invention, this is accomplished by first of all assigning to one bit of the configuration bus the function of indicating the presence of an external device other than a memory, i.e. an external processor. A small amount of additional hardware is provided to which, in the illustrated embodiment the first four bits of the configuration bus are connected. This hardware comprises a bi-directional buffer. The bi-directional buffer is turned on by the external processor bit from the configuration bus and converts the first four lines, in the illustrated embodiment, from configuration lines to two input lines and two output lines. The function of the two output lines is to provide a lock bit protocol. The lock bit protocol is a means of providing signals to the external device to indicate that the processor wishes to lock out its memory. For example, if the first processor is accessing the memory of the external processor and needs to continue to have access to that memory, it will wish to lock out the remote processor from obtaining access to that memory. The lock bit protocol provides this means.

On the other hand, there are instances where the external processor itself will have locked out its own memory to prevent access to it from the main processor. In this case, it is desirable that there be the opportunity for the processor to try again to access the memory. Thus, in accordance with the present invention, when the remote memory is locked out, logic sends back to the first computer system on the configuration bus a RETRY signal which provides an input to a RETRY logic in the computer system to initiate a RETRY cycle as soon as the RETRY signal is deasserted. In effect, if the first processor tries to access the remote memory and the lock bit is present, the RETRY signal causes the memory cycle to be aborted and to then restart when the RETRY signal is no longer asserted.

The other signal provided by the reconfigured bus is a signal called STALL. The bus when used as a memory bus is an asynchronous bus. A memory cycle from the processor operates the bus asynchronously. This means the processor expects a response from the memory a certain time after starting a cycle. In the illustrated embodiment this can be for example either a ready signal or an error signal. However, when operating and accessing a remote device over the I/O bus, the remote device is not synchronized to the cycle of the processor. Thus, it is not possible in all cases to provide the response at the required time. The processor is responsive to the STALL signal to extend its cycle, and so long as the STALL signal is asserted, wait for the response. Thus, providing the STALL signal on the reconfigured bus permits converting what was a synchronous bus to an asynchronous bus and permits communications with the remote device which would not otherwise be possible.

Thus, the reconfiguration of the bus in accordance with the present invention is done to provide the input and output signals, not found on a memory bus, which are necessary for proper I/O operations in any given case. In the illustrated embodiment these signals relate to the lock bit and RETRY and STALL signals. In other systems, other signals may be necessary.

FIG. 1 is a block diagram of a computer system employing the present invention.

FIG. 2 is an illustration of the local bus structure of the system of FIG. 1.

FIG. 3 depicts the system bus node space assignments of the system of FIG. 1.

FIG. 4A and FIG. 4B represent a block diagram of a memory controller contained in the memory system of the system of FIG. 1.

FIG. 5 is an illustration of the memory array bus structure of the memory system of FIG. 4.

FIG. 6 is a block diagram illustrating the configuration of the bus of FIG. 5.

FIGS. 7, 7A and 7B are timing diagrams showing the RETRY sequence.

Referring now to the drawings and initially to FIG. 1, illustrated in block diagram form, is a system configuration according to the present invention. The system includes a single chip microprocessor 10, a floating point coprocessor 11, a memory controller 12, a local RAM 13, a local ROM 14, an off board memory 15, a time of year (TOY) clock 25, an Octal ART 26, a PAL 20, a system bus interface 16 and 24, latches 28 and transceivers 27, 32 and 34. The system components 10, 11, 12, 13, 14, 15, 16 are interfaced with one another by means of a bus 17. The system components 10, 11, 12, 13, 14, 15, 16, 17 are interfaced through the bus interface 16 to bus 18.

A preferred system configuration for advantageous implementation of the teaching of the invention comprises a single board computer module 9 which contains system components 10, 11, 12, 13, 14, 16 24, 25, 26, 20, 24, 28, 27, 32, 34, 30, 31 AND BUS 17. The single board computer may be a module which is 32-bit, single-board processor that provides VAX functionality. The module is compatible with system bus 18 which is manufactured by Digital Equipment Corporation and hereinafter referred to as the system bus. The term "VAX" refers to a product line manufactured by Digital Equipment Corporation.

The module 9 may function either as a primary or a secondary CPU in a larger computer configuration. The module 9 thus serves as a computational engine for an operating system and for user application programs.

The module 9 implements console functions. A console 100 provides the user interface to the module 9 and to the system configuration in which it is used. Through console 100, a user can bootstrap the system image, invoke diagnostic routines, examine or write registers and memory throughout the system bus 18 configuration, and interactively communicate with the operating system and application programs.

In addition to a console interface, the module 9 provides an asynchronous serial interface 35 (RS-232C compatible) for up to eight terminals 103. The terminals operate independently of the console subsystem.

The module 9 has extensive on-board diagnostics. On power-up or node reset, it tests itself and makes its status (pass or fail) available over the system bus 18. A set of LEDs on the printed circuit board of module 9 also indicates the outcome of self-test. In addition, a user can invoke other on-board diagnostics to test the logic and functions of module 9 more extensively.

The microprocessor chip 10 in the illustrated embodiment is a 32-bit, virtual-memory processor chip. Chip 10, for example, can directly address a 1 Gbyte physical address space.

The floating-point unit 11 (78132) is a single-chip, floating-point processor 11 that functions as a slave to the processor 10. The floating-point unit (FPU) performs 32-bit, floating-point computation and accelerates the execution of certain integer and character-string instructions.

The local bus 17 is a 32-bit, asynchronous, bidirectional bus that connects the processor 10 with on-board memory 13 and 14 and 14 and on-board devices. The local bus 17 is time-multiplexed so that address information and data are transmitted over the same lines during different cycles of transaction.

The local bus interface logic 19 connects the local bus 17 with the processor 10, the chip 16, and components on the memory bus 22 (MBUS). The processor 10 and chip 16 transceivers generate byte parity on writes and check byte parity on reads. The bus transceivers are controlled by the MBUS PAL 20.

The memory bus 22 (MBUS) connects local memory, as well as other devices, to the local bus 17. The MBUS 22 is a 32-bit data bus that interfaces with the local bus 17 through transceivers 27. With the exception of the real-time clock chip 25, the MBUS devices are addressed over a physical address (PA) bus 23. During the address part of a local bus cycle, local bus address bits [29:0] are latched by the PA latches 28.

The module 9 has 1 Mbyte of on-board RAM 13. Byte parity is generated for RAM writes, and it is checked on RAM reads. The RAM can be backed up by batteries.

The module 9 has 128 Kbytes of ROM 14 that store firmware to perform initialization routines, the primary bootstrap routine, console program code, diagnostic routines and in-circuit tester pattern-generating code. The module 9 can accommodate up to 256 Kbytes of ROM.

The memory controller 12 controls access to the physical address space. The memory controller 12 decodes the address on the PA bus 23, enables the appropriate device, and returns status information to the local bus master. The memory controller 12 also drives on-board RAM 13 and provides refresh timing for all local RAM, which includes off-board RAM on memory array modules 15.

Pursuant to the teachings of the present invention, on power-up or node reset, the primary processor in the system, which could be the module 9 itself, assigns the module 9 a starting address offset (SAO). The SAO represents the base address for local RAM within system physical memory space. The memory controller 12 uses the SAO to decode memory references. If the module 9 is not the primary processor, the console software assigns the starting address from the system Starting Address Register.

The memory array bus 21 is an off-board extension of the local bus 17. The memory array bus 21 connects the module 9 with daisy chained memory array boards 15 in accordance with the present invention. The memory array bus 21 thus allows the module 9 to access an extended main memory.

Each memory array board 15 contains 2 Mbytes of dynamic RAM implemented on 256 Kbyte memory chips. Up to five memory array boards 15 can be daisy chained to module 9 memory array bus 21. Main memory can thus be expanded to a total of 11 Mbytes (1 Mbyte on-board plus 10 Mbytes on memory array boards 15).

Pursuant to the present invention, the local bus master recognizes there is no distinction between the on-board RAM 13 and memory array board 15. The access time and other operating characteristics are identical. Memory array board 15 can also be backed up by batteries.

The local bus 17 and the system bus 18 interface through interface chips 16 and 24. Chip 24 is a standard user interface to the system bus 18. The chip 16, which connects to the chip 24 through a bus 33, is a memory-mapped peripheral device of the processor 10.

The chip 16 is the window of module 9 onto the system bus 18. When the processor 10 references a nonlocal physical address, the chip 16 translates the local bus 17 transaction into a transaction on the system bus 18. When another system node addresses the module 9, the chip 16 translates the incoming system transaction into a local bus 17 transaction.

The chip 24 implements the system bus 18 commands and performs system error checking and arbitration.

The console 100 is the primary user interface to module 9 and to the system computer system as a whole. The console consists of an RS-232C interface 30, a control panel 29, and console firmware that implements console commands. A DLART (DIGITAL Link asynchronous receiver/transmitter) chip provides parallel-to-serial and serial-to-parallel conversion between the processor chip 10 and the RS-232C transceivers 30.

The console has two operating modes: program mode and console mode. In program mode, characters typed at the console terminal are passed to the operating system; this is the normal operating mode (run mode). In console mode, characters typed at the console terminal are interpreted as console commands; the operating system and all application programs are suspended.

The module 9 provides an asynchronous serial interface 35 (RS-232C compatible) for up to eight terminals 103. This interface is implemented in an octal ART 26 (octal asynchronous receiver/transmitter) chip. The octal ART 26 supplies eight independently programmed serial lines. One line (line 0) has modem-control signals that enable communication with a modem.

The module 9 provides a time-of-year (TOY) clock and a real-time clock. Both clock functions are implemented in the Motorola RTC chip 25 (MC146818). This chip 25 calculates and stores the time and date during system operation. The TOY clock 25 has 50 bytes of battery-backed up RAM used by the console firmware to store the state of processor 10 when console mode is entered. The real-time function permits programmable interrupts to the processor chip 10.

The local (integrated circuit interconnect) bus 17 is a 32-bit, asynchronous, bidirectional bus. The local bus 17 interfaces the processor chip 10 with local memory and on-board logic and supports direct memory access (DMA) transfers to and from local memory and the system bus.

The local bus 17 is time-multiplexed. That is, address information and data are transmitted over the same lines during different parts of the bus cycle. Toward the beginning of the cycle, the address is transmitted. Toward the end of the cycle, data is transmitted.

Communication between any two devices on the local bus 17 takes place within a master/slave relationship. The bus master initiates and controls the transaction. The slave responds to the bus master's commands and returns appropriate status information. The local bus 17 is an interlocked bus in that only one transaction can be on the bus at any given time.

Data transfers on the local bus 17 occur on naturally-aligned longword (32-bit) boundaries. For example, memory fetches occur on address boundaries ending in 0, 4, 8 and C (hex). The bus master asserts the appropriate Byte Mask lines to indicate which bytes within a transferred longword contain valid data.

The local bus's 17 data and control signals are divided into classes. FIG. 2 depicts the local bus 17 structure and also delineates the classes of signals. The first class is information transfer signals and that encompasses 32 data and address lines and 4 lines for byte mask information. The second class is bus control and it entails control signals such as write, address strobe, data buffer enable, data strobe, ready and error. The third class of signals is system control which is comprised of 3 lines of cycle status and reset. The fourth class of signals is DMA (direct memory access) control which contains DMA request and DMA grant. The last class of signals is interrupt control which is comprised of 4 lines for interrupt request, powerfail, halt and interval timer interrupt.

The data and address lines are 32 bidirectional lines that are time-multiplexed. Toward the beginning of a read or write cycle, DAL [31:0] provide address and control information via DAL [31:30] to indicate the length of data to be transferred as follows:

The processor chip 10 executes quadword transactions on the local bus 17 as two consecutive longword transactions and DAL [29:2] convey a physical longword address. Data is always transferred on naturally-aligned longword boundaries. Bit [29] distinguishes memory space from I/O space such that when DAL[29]=0, it is a memory reference and when DAL[29]=1, it's an I/O reference.

DAL [1:0] are the byte address but may differ from the address implied by BM [3:0] during an instruction prefetch (always an aligned longword), during a character string data prefetch (always an aligned longword), during a PTE (page table entry) read (always an aligned longword) or during the second cycle of an unaligned operation.

Toward the end of a write cycle, DAL [31:0] transmit data from the bus master to the slave device. Toward the end of a read cycle, DAL [31:0] transmit data from the slave to the bus master.

DAL [31:0] are also used for interrupt acknowledge cycles, external processor cycles and external processor register cycles. For an Interrupt Acknowledge Cycle, toward the beginning of an interrupt acknowledge bus cycle, DAL [4:0] output the interrupt priority level (IPL) of the interrupt. Toward the end of the cycle, DAL [9:2] input the interrupt vector.

For an external processor register cycle, the processor chip 10 does not implement all the internal processor registers (IPRs) specified in the VAX architecture. These omitted IPRs are implemented externally and accessed through MTPR (Move to Processor Register) and MFPR (Move from Processor Register) commands. In the module 9 the console registers in the DLART chip are accessed as external processor registers. These bus cycles are similar to external processor cycles.

The Byte Mask signals specify which bytes on DAL [31:0] contain valid information during the data transfer part of a read or write cycle. When BM [3] is asserted, DAL [31:24] contain valid data. When BM [2] is asserted, DAL [23:16] contain valid data. When BM [1] is asserted, DAL [15:8] contain valid data and when BM [0] is asserted, DAL [7:0] contain valid data.

During a read cycle, the Byte Mask signals indicate which bytes the slave device must place on DAL [31:0]. The bus master ignores bytes that do not contain valid data. During a write cycle, the Byte Mask signals indicate which bytes on DAL [31:0] contain valid data. The slave device ignores the masked bytes. During unmasked reads and writes, and during interrupt acknowledge cycles, all four Byte Mask signals are asserted. The Byte Mask signals are valid only when the Address Strobe line (AS) is asserted.

The Write line specifies the direction of a data transfer on DAL [31:0]. If the bus master asserts WR, then it will drive data onto DAL [31:0]. If the bus master does not assert WR, then the slave device will drive data onto DAL [31:0]. WR is valid only when AS or EPS is asserted.

The Address Strobe line provides timing information for address transfers. The bus master asserts AS when the address on DAL [29:2] is valid during a read, write, or interrupt acknowledge cycle. The bus master deasserts AS at the end of the bus cycle.

The Data Buffer Enable line, in conjunction with WR, controls the DAL transceivers. The bus master asserts DBE to enable the transceivers and deasserts DBE to disable them.

The Data Strobe line provides timing information for data transfers. During a read cycle or an interrupt acknowledge cycle, the bus master asserts DS to indicate that it is ready to receive incoming data. The bus master then deasserts DS to indicate that it has received and latched the incoming data. During a write cycle, the bus master asserts DS to indicate that DAL [31:0] contain valid write data. The bus master then deasserts DS to indicate that it is about to remove the write data from DAL [31:0].

External logic asserts the Ready line to indicate the normal termination of a read, write, or interrupt acknowledge cycle. During a read cycle, this indicates that the slave device will place the required data on DAL [31:0]. During a write cycle, this indicates that the slave device has received data on DAL [31:0]. During an interrupt acknowledge cycle, this indicates that the interrupt vector has been placed on DAL [31:0]. When the bus master recognizes an asserted RDY, it terminates the current bus cycle. External logic then deasserts RDY.

External logic asserts the Error line to indicate the abnormal termination of a read, write, or interrupt acknowledge cycle. After receiving an asserted ERR, the bus master terminates the current bus cycle. External logic then deasserts ERR. On receiving an asserted ERR, the processor chip executes a machine check.

A response PAL 41 in the memory controller 12 asserts ERR whenever a parity error occurs on a read from local RAM 13, provided that parity error reporting is enabled.

The Cycle Status signals, in conjunction with WR, provide status about the current bus cycle. When AS is asserted during a read, write, or interrupt acknowledge cycle, CS [2:0] are encoded as shown in Table 1.

When system DC LO L is asserted, the RESET line is asserted, forcing logic on the local bus 17 to an initialized state. A reset aborts any local bus 17 transactions and/or chip internal operations in progress, and causes the module 9 to enter its initialized state.

External logic requests control of the local bus 17 by asserting the DMA Request (DMR) line. DMR must be asserted as long as external logic controls the bus. At the end of the DMA bus cycle, external logic deasserts DMR. In the module 9 the chip 16 is the only device that uses the local bus 17 for DMA transfers.

The processor chip 10 asserts the DMA Grant (DMG) line to give external logic control of the local bus 17. The processor 10 asserts DMG as long as external logic has control of the bus and deasserts DMG after external logic has deasserted DMR. In the module 9, the chip 16 is the only device that requests and is granted control of the local bus 17. DMG is also asserted during retry cycles.

The Interrupt Request signals are the device interrupts to the processor chip 10. External logic asserts IRQ[3:0] to post interrupt requests.

Each IRQ line has a unique interrupt priority level (IPL). The processor 10 accepts an interrupt request only if the request's IPL is higher than the IPL at which the processor 10 is currently operating. Each interrupt line also corresponds to a unique IPL on the system bus. An incoming system interrupt is translated by the chip into a corresponding local bus device interrupt.

The IPL assignments for IRQ [3:0] are: IRQ [3] is IPL 17 (hex); IRQ [2] is IPL 16 (hex); IRQ [1] is IPL 15 (hex); and IRQ [0] is IPL 14 (hex). IRQ [3:0] are level-sensitive signals that the processor chip 10 samples every cycle (200 ns). The processor 10 synchronizes an IRQ signal internally. When the processor 10 accepts an interrupt request, it completes the current macroinstruction and then executes an interrupt acknowledge bus cycle. During this cycle, the processor 10 acknowledges the IPL at which the IRQ is posted and then inputs the vector for the interrupt. In the module 9, the chip 16 inputs the acknowledged IPL and outputs the appropriate interrupt vector on DAL [9:2].

On the assertion of system AC LO L, indicating that the line voltage is below specification, the powerfail interrupt (BCI AC LO) is asserted to the processor 10. This interrupt vectors program control to a powerfail routine. When the module 9 asserts system DC LO L, indicating DC system voltages are about to fall below specification, the RESET line is asserted. RESET initializes the processor 10 and other logic on the local bus 17.

When the module 9 is in program mode and the key switch is enabled, pressing the BREAK key on the console terminal keyboard asserts HALT to the processor 10, forcing the module 9 to enter console mode. After halting, the processor 10 executes an external processor write cycle. During this cycle, Cycle Status [1:0]=10 and DAL [5:0]=111111. The module 9 ignores this bus cycle.

HALT is a nonmaskable interrupt. HALT always interrupts the processor chip 10 after the current macroinstruction completes, regardless of the processor's 10 current IPL. HALT is an edge-triggered signal that is sampled every cycle (200 ns).

The interval counter overflows every 10 milliseconds, thereby generating an interval timer interrupt (INTTIM) to the processor 10 at IPL 16.

The interval counter is controlled by the Interval Clock Control Register (ICCR), an IPR in the processor 10. The counter is enabled when bit [6] in the ICCR is set and disabled when bit [6] is clear.

The module 9 supports different local bus 17 cycle types. The types include idle, read, write, interrupt acknowledge, DMA and retry.

During an idle cycle, DAL [31:0] are asserted but are undefined, and the bus control signals are deasserted. An idle cycle lasts four clock phases (nominally 200 ns).

In a CPU read cycle, the bus master inputs data from memory or an on-board I/O device. A CPU read cycle lasts at least eight clock phases (nominally 400 ns) and may last longer in increments of four clock phases (nominally 200 ns).

For a CPU read cycle, the bus master drives a physical longword address onto DAL [29:2]. The bus master asserts BM [3:0] as required and deasserts WR. The bus master asserts AS, indicating that the address is valid, and then asserts DS, indicating that DAL [31:0] are free to receive incoming data. If no error occurs, the slave device drives the required data onto DAL [31:0], and external logic asserts RDY. If an error occurs, external logic asserts ERR, which aborts the bus cycle. This causes the processor 10 to execute a machine check. Finally, the bus master deasserts AS and DS to end the cycle.

In a CPU write cycle, the bus master outputs data to memory or an on-board I/O device. A CPU write cycle lasts at least eight clock phases (nominally 400 ns) and may last longer in increments of four clock phases (nominally 200 ns).

For a CPU write cycle, the bus master drives a physical longword address onto DAL [29:2]. The bus master asserts BM [3:0] as required and asserts WR. The bus master asserts AS, indicating that the address is valid. The bus master then drives data onto DAL [31:0] and asserts DS, indicating that the data is valid on DAL [31:0]. If no error occurs, the slave device reads the data, and the external logic asserts RDY. If an error occurs, external logic asserts ERR, which aborts the bus cycle. This causes the processor 10 to execute a machine check. Finally, the bus master deasserts AS and DS to end the cycle.

In an interrupt acknowledge cycle, the processor chip 10 acknowledges a posted interrupt and inputs an interrupt vector. An interrupt acknowledge cycle lasts at least eight clock phases (nominally 400 ns) and may last longer in increments of four clock phases (nominally 200 ns). The detailed timing of an interrupt acknowledge cycle is identical to that of a CPU read cycle.

At the beginning of the cycle, the processor 10 outputs on DAL [4:0] the IPL of the interrupt being acknowledged. The processor 10 then asserts AS to indicate that DAL [4:0] contain a valid IPL. The chip 16 reads the IPL from the bus. The processor 10 then asserts DS to indicate that DAL [9:2] are free to receive the interrupt vector. The chip outputs the appropriate interrupt vector on DAL [9:2]. If an error occurs, external logic asserts ERR, which causes the processor 10 to ignore the data and abort the transaction. Finally, the processor chip 10 deasserts AS and DS to end the interrupt acknowledge bus cycle.

The processor 10 initiates the cycle by driving the cycle status onto CS [1:0], precharging and sustaining CS [2] high, and asserting EPS. The external processor responds by placing the required information on DAL [31:0]. The processor 10 reads the information on DAL [31:0] and deasserts EPS. The external processor then removes the information from DAL [31:0] to end the cycle.

In an external processor write/command cycle, the processor chip 10 outputs either a command or data to an external processor. The cycle lasts four clock phases (nominally 200 ns).

In a DMA cycle, the processor chip 10 relinquishes control of the local bus 17 to an on-board DMA device. The cycle begins when a DMA device (the chip) asserts DMR to request control of the bus. At the conclusion of the current bus cycle, the processor 10 tri-states DAL [31:0], AS, DS, WR, and DBE; drives high and tristates BM [3:0] and CS [2:0], and asserts DMG. The DMA device may now use the bus to transfer data. On completion of the transfer, the DMA device stops driving AS, DBE, and DS and deasserts DMR. The processor 10 then deasserts DMG and begins the next bus cycle.

If a transaction request is not granted, the processor chip 10 retries the transaction. The local bus 17 address is stored in the chip.

The local bus 17 supplies eight hardware interrupts to the processor 10. Four of these interrupts are external and four are internal. External interrupts allow on-board devices to interrupt the processor 10. Internal interrupts alert the processor 10 to impending power failures, console halt requests, and interval timer overflows.

The Interrupt Request lines, IRQ[3:0], are the four external interrupts. IRQ [3:0] are assigned IPLs 17 through 14 (hex), respectively. The higher the IPL, the higher the interrupt priority. The processor 10 responds to an interrupt request when the request's IPL is higher than the IPL at which the processor 10 is currently operating.

The processor 10 responds to an external interrupt after completing the current macroinstruction by executing an interrupt acknowledge bus cycle. During this cycle, the processor 10 outputs the IPL of the interrupt on DAL [4:0], and then inputs the interrupt vector from the requesting device on DAL [9:2]. In the module 9, the chip inputs the IPL and outputs the appropriate vector for all interrupt acknowledge cycles.

The three external events which generate internal interrupts to the processor 10 are (1) pressing the BREAK key on the console terminal keyboard (if the halt-on-BREAK function is enabled), (2) a power failure and (3) an interval counter overflow. The following sections describe these events. Internal interrupts do not cause the processor 10 to execute an interrupt acknowledge bus cycle. Instead, program control is vectored to a routine appropriate to the type of internal interrupt.

When the module 9 is in program mode, pressing the BREAK key on the console terminal keyboard asserts HALT to the module 9, forcing the module 9 to enter console mode after completing the current macroinstruction. The halt-on-BREAK function can be disabled by setting the front panel key switch to the Secure position.

When the input line voltage goes below specification, the module 9 system control unit in the system asserts system AC LO L, which generates a powerfail interrupt to the processor 10 at IPL 1E. This interrupt vectors the processor 10 to a powerfail routine.

The interval counter overflows every 10 milliseconds, thereby generating an interval timer interrupt (INTTIM) to the processor 10 at IPL 16.

Interval timer interrupts are controlled by the Interval Clock Control Register (ICCR), an IPR in the processor 10. Setting bit [6] in the ICCR enables the interrupts. Clearing this bit disables the interrupts.

On-board peripheral devices request and are granted control of the local bus 17 through DMA Request (DMR) and DMA Grant (DMG), respectively. The chip uses DMAs to transfer data between the local memory 13 and the system bus 18.

The Error line on the local bus 17 alerts the bus master to an error condition during the current bus cycle. The memory controller 12 asserts ERR whenever a parity error occurs on a read from local RAM 13, provided that parity is enabled. This causes the processor 10 to perform a machine check.

The local bus 17 carries a 28-bit physical address on DAL[29:2] that can reference 1 Gbyte physical address space. The local bus address space is equally divided into I/O space and memory space. For I/O references, address bit [29] is cleared. All on-board peripheral devices and ROM 14 reside in local bus 17 I/O space. Local RAM 13 resides in local bus memory space.

The module 9 communicates with other system bus 18 nodes by interfacing the local bus 17 to the system bus 18. This interface lets the module 9 access I/O controllers, bus adapters, system memory modules, and other system processors.

Most of the logic that provides this interface to the system bus 18 is implemented by the chip 24 (bus interconnect interface chip), and the chip 16 (interface chip). Additional interface logic is provided by the node ID buffers and the Console Display/Control Register (CDCR).

On one side, the chip 24 connects directly to the system bus 18. On the other side of the chip 24 is the BCI bus, which is the standard user interface to the system bus 18. In fact, from the perspective of the module 9, the chip 24 appears to be the system bus. The BII chip also implements the system protocol and error checking.

The chip 16 chip interfaces the local bus 17 to the BCI bus 33. The BCI bus 33 is the user side of the chip 24 chip. The 16 is thus the window on the system bus 18. The processor 10 sees the chip 16 as a memory-mapped peripheral device.

Whenever the module 9 processor issues a read or write to non-local memory space, a response PAL 41 (see FIG. 4) asserts system REFERENCE to the chip 16. The chip 16, aided by the chip 24, then translates the local bus 17 transaction into a system transaction.

Whenever the module 9 processor issues a read or write to non-local I/O space, the chip 16 automatically translates the transaction onto the system bus 18. The chip 16 reports the status of windowed transactions back to the processor 10.

In addition, the chip 16, aided by the chip 24, translates incoming system transactions into local bus 17 transactions. When the local bus 17 transaction completes, the chip 16 reports the transaction's completion status and, if appropriate, returns data to the originating system node.

If contention arises between an incoming system transaction and an outgoing local bus 17 transaction, the chip 16 asserts RETRY to a retry PAL 40 (see FIG. 4) in the memory controller 12. The retry PAL 40 then suspends the outgoing local bus 17 transaction until the incoming system transaction completes. The chip 16 stores the address for all local bus 17 transactions in its Retry Address Latch. When a suspended local bus 17 transaction is retried, the memory controller 12 fetches the stored local bus 17 address from the chip 16.

The chip 16 provides an interrupt controller for the processor chip 10. When the processor chip 10 acknowledges an interrupt from any source, the chip 16 inputs the acknowledged IPL (interrupt priority level) and supplies the processor 10 with the appropriate interrupt vector. In addition, the chip 16 translates system interrupts targeted at the module 9 processor into corresponding local bus 17 interrupts to the processor 10.

The chip 16 provides 14 general purpose registers in the node space. GPR 0, the Receive Console Data (RXCD) Register, is the console interface to the system bus 18. GPR 5, the Permanent Power-Up Diagnostic Register, stores a copy of the system self-test results. GPRs 7 through 11 are used by the console entry sequence. GPR 11, the Console Status Register, contains status on initialization. The rest of the chip 16 GPRs are not used.

The chip 16 implements two functions that the processor 10 does not use during normal operation but that may be useful for diagnostic purposes and special operations. The processor 10 can construct commands for the system bus 18 by writing certain chip 16 registers. The chip 16 also contains a special data mover that can move data over the system bus in octal word packets.

The Console Display/Control Register (CDCR) provides direct reads of several control lines on the system bus 18. The CDCR can also be written by the processor chip 10 to assert system RESET L and system BAD L on the system bus 18.

The system node identification number of module 9 is determined by four signal pins on the system backplane. This node ID number is independent of the slot location of module 9 in the system card cage. On power-up or reset, the node ID is loaded into the system Control and Status Register of the chip 24. The node ID determines which 8 Kbyte block in system nodespace is assigned to the module 9. This block contains the addresses of the system accessible registers. Blocks of nodespace are allocated according to the mapping scheme shown in FIG. 3.

The system bus 18 registers, which have addresses in system nodespace, are accessible to other system nodes. System access to the local RAM is determined by the Starting Address and Ending Address Registers in the chip 24. These registers are loaded on power-up or reset by the primary processor in the system (which could be the module 9 itself). During normal operation, all of the local RAM 13 is accessible over the system bus 18.

Table 2 describes the implementation of the system commands by module 9. Since the processor 10 does not have cache memory, it does not issue read or write commands with the intention of caching data.

When the reference is a nonlocal memory address, the memory controller 12 asserts system REFERENCE to the chip 16. The chip 16 then translates (windows) the local bus 17 transaction into a system transaction. Normally, the local bus 17 is unavailable for other transactions until the windowed system transaction completes. However, if the chip 16 receives an incoming system bus 18 transaction, it takes control of the local bus 17 to service the transaction and then restarts the interrupted local bus 17 transaction. If the system transaction completes successfully, the chip 16 asserts the Ready line. Otherwise, the chip 16 interrupts the processor 10 or, for serious errors, asserts the Error line.

Table 3 lists the translated local bus 17 transactions and their system counterparts. Table 4 indicates the conditions on the system bus 18 that cause the chip 16 to assert the Error line to the processor 10.

When the module 9 (processor 10) is addressed over the system bus 18, the chip 16 asserts the DMA Request line (DMR) to the processor 10 to request mastership of the local bus 17. When granted bus mastership, the chip 16 translates the incoming system transaction into a local bus 17 transaction. The system bus 18 is interlocked during translated local bus 17 reads but not during translated local bus 17 writes. When the local bus 17 transaction completes, the chip 16 reports the completion status and, if appropriate, transfers data to the originating system node;

Table 5 lists the system transactions to which the module 9 responds and their local bus 17 translations. Table 7 indicates how the chip 16 translates local bus 17 status onto the system bus 18. Quadword and octaword transactions on the system bus 18 are translated into a series of longword transaction on the local bus 17.

The module 9 module accepts both Interrupt (INTR) commands and Interprocessor Interrupt (IPINTR) commands from the system bus 18.

On receiving an INTR command, the chip 16 asserts the corresponding local bus 17 Interrupt Request (IRQ) line. System IPLs BR 7-4 correspond to local bus 17 interrupts IRQ [3:0], respectively. The processor chip 10 responds to the interrupt request by executing an interrupt acknowledge transaction, which the chip 16 windows onto the system bus 18 as an IDENT transaction. The chip 16 inputs the vector from the interrupting node and returns this vector to the processor chip 10.

On receiving an IPINTR, the chip 16 asserts IRQ [14] to the processor chip 10. When the processor chip 10 acknowledges the interrupt request, the chip 16 returns 80 (hex), the vector for all IPINTRs.

Pursuant to an important feature of the present invention, the module 9 supports both local memory 13 (on-board dynamic RAM), expansion memory (off-board RAM) on memory array boards 15, which are daisy chained, and 128 Kbytes or 256 Kbytes of on-board ROM 14. The module 9 does not have cache memory.

In one embodiment, the module 9 has one Mbyte of on-board RAM 13, which is composed of 256 Kbyte dynamic RAM chips. The local bus 17 master (processor 10 or chip 16) can access this memory only on naturally-aligned longword boundaries. However, the bus master can use the Byte Mask signals to specify which bytes in a transferred longword contain valid data.

On-board RAM 13 is parity protected. Byte parity is written to RAM on all writes and checked on all reads. Parity generation and checking is performed by the processor transceivers (for processor-initiated transactions) and by the chip 16 transceivers (for chip 16-initiated transactions). Parity is generated and checked only for bytes specified to be valid by the Byte Mask signals.

The memory access time depends on the local bus 17 master. A processor 10 initiated read cycle takes 400 ns, while a chip 16 initiated read cycle takes 50 to 100 ns longer, due to the synchronization requirements of chip 16. Write cycles are the same for both bus masters.

The system of the present invention can support a plurality of memory array boards 15 daisy chained to the memory array bus 21, as will appear. In the illustrated embodiments, because of physical constraints there is room for only five boards. Also in this embodiment each memory array board 15 contains two Mbytes of RAM. Each board is driven by its own on-board logic and by bus control signals.

An important feature of the present invention is that to the local bus 17 master, there is no distinction between on-board RAM and off-board RAM. Off-board RAM supports byte parity, masked reads and writes, and has the same access times as does on-board RAM.

Because of the daisy chaining, memory array boards 15 must be placed in slots adjacent to the board 9. There can be no intervening slots or boards between the memory array boards 15 and the board 9.

The main memory and the memory controller 12 support longword read, read interlock, longword write, masked longword write and unlock write mask memory cycle types. All memory bus cycles are on naturally-aligned longword boundaries.

A longword read is a simple read from memory. The read interlock bus cycle locks all of main memory so that another read interlock cannot access memory until the lock is cleared. However, noninterlocked types of read operations can still access memory. A read interlock sets the Memory Locked bit in the memory controller 12.

A longword write is a simple write to memory. In the masked longword write, certain bytes within the referenced longword may be masked out. The Byte Mask signals indicate which bytes are valid. The unlock write mask clears the lock on main memory set by a previous interlock read. An unlock write mask clears the Memory Locked bit in the memory controller 12.

According to the teachings of the present invention, on power-up or node reset, the initialization code sizes the local RAM 13 and makes the size available in chip 24 GPR 1. The module 9 then assigns a starting address offset (SAO) to the local RAM 13. The SAO is the base address of the local RAM 13 in system memory space. If the module 9 is a secondary processor, the SAO value is based upon the contents found in the chip 24 Starting Address Register (SADR).

The module 9 contains independent power connections for battery-backup voltages. When provided, battery backup maintains the local RAM 13 and refresh circuitry. Battery-backup voltage is monitored by the system control unit.

The standard 128 Kbyte ROM provides 32K longword locations at addresses 2004 0000 hex through 2004 5FFF hex. The optional 256 Kbyte ROM provides 64K longword locations at addresses 2004 0000 hex, through 2004 7FFF hex. Both ROM configurations are accessed on naturally-aligned longword boundaries.

The 128 Kbyte ROM 14 configuration has a 15-bit address (PA [16:2]). The 256 Kbyte ROM configuration uses the same 15 address bits plus the signal 256 Kbyte ROM as the 16th (high-order) address bit. When 128 Kbyte ROM 14 is present, the signal 256 Kbyte ROM is allowed to float, that is, the bit is not connected to the ROM chips and therefore can be neither high nor low. When the 256 Kbyte ROM is present, however, the signal 256 Kbyte ROM must be connected to the ROM chips by adding a jumper to the module. The jumper connects ROM address 17 coming from the processor physical address latch with the 256 Kbyte ROM signal.

The standard 128 Kbyte ROM 14 provides 32K longword locations, but the total ROM space can accommodate 64K longwords. To completely fill the available boot ROM space, the 32K body of ROM code is replicated twice throughout this range of addresses.

Corresponding addresses from each set map to the same longword. This complete coverage of ROM space eliminates errors resulting from referencing nonexistent (unimplemented) locations. If a nonexistent location is referenced, the data returned is undefined.

The standard ROM configuration 14 contains the initialization routines, the primary bootstrap routine (VMB), diagnostic routines, and the console program, which are written in macrocode.

On power-up or node reset, program control jumps to the first location, 2004 0000 hex, in boot ROM space, and the module 9 executes its initialization routine. Note that the initialization code resides in the I/O space at addresses 2004 0000 hex, through 2008 0000 hex.

The primary bootstrap's main function is to locate the secondary bootstrap routine, bring it into memory, and then transfer control to it. The secondary bootstrap then loads the operating system.

The primary bootstrap also performs the following initialization in which it, tests all of local memory and builds a bit map that flags physical memory pages (512 Kbytes each) as good or bad, finds out what devices are present and operative in the system, and sets up the restart parameter block (RPB), a data structure used by powerfail and recovery routines to save volatile processor state.

VMB contains a boot driver that loads the secondary bootstrap program into memory and also writes a dump file to disk if the operating system detects a fatal error.

The diagnostic code consists of a self-test, which runs automatically on power-up and reset, and additional diagnostics, which can be invoked from a console. The self-test provides a quick confidence check of the hardware at the node, including any attached memory array boards 15. The additional diagnostics test the module 9 more extensively. The diagnostic code also contains a parser that parses diagnostic commands.

The memory controller 12 controls access to the on-board memory, registers, and devices, as well as access to off-board memory that resides on attached memory array boards 15. Whenever the processor 10 or chip 16 references the physical address space, the memory controller 12 decodes the physical address and the local bus 17 cycle type, and provides completion status to the local bus 17 master. In addition the memory controller 12 controls the addressing of on-board RAM 13, refresh timing for both on-board and off-board RAM 15, parity error reporting and local bus retry timing.

FIG. 4 is a block diagram showing the major components of the memory controller 12. These include I/O address PAL 45, memory address comparator 44, status PAL 43, cycle PAL 42, response PAL 41, retry PAL 40, DRAM controller chip 46, parity error logic and the memory control and status registers (MCSRs) 47.

The I/O address PAL 45 decodes all addresses on the physical address (PA) bus 23. The address can be a memory address or an I/O address.

If the address is an implemented I/O address, the I/O address PAL 45 asserts KASPC (module 9 SPACE) to the response PAL 41, indicating that the address is implemented. The PAL 45 also selects or enables the referenced device which includes ROM 14, CDCR (Console Display/Control Register) PAL, Octal ART chip 26, MCSRs 47 (Memory Control and Status Registers) and TOY (time-of-year) clock 25. If the I/O reference is to an unimplemented I/O location (addresses 2008 5000 hex through 203F FFF hex), the address PAL 45 asserts NSSPC (Nonsupported Space) to the response PAL 41. The response PAL 41 then asserts RDY to the local bus 17 master, however, the data on DAL [31:0] is undefined.

Pursuant to the present invention, the memory address comparator 44 decodes all addresses on the physical address (PA) bus 23. The comparator 44 inputs PA [29:28] and compares address bits DAL [27:0] with the starting address offset (SAO [7:0]) for local RAM 13. The starting address offset specifies the local RAM's 13 base address in system memory space.

If PA [29] is set, the reference is to I/O space, otherwise, the reference is to memory space. If PA [28] is set, the reference is above the upper limit, 0FFF FFFF hex, of the 256 Mbyte address range supported by the memory controller 12, and therefore cannot be implemented in the local RAM 13. If both PA [29] and PA [28] are clear, the comparator compares DAL [27:0] with SAO [7:0]. If DAL [27:0] is less than SAO [7:0], the referenced address is below the starting address offset, and therefore cannot be implemented in local RAM 13. If DAL [27:0] matches SAO [7:0], the referenced address is to on-board RAM 13. If DAL [27:0] is greater than SAO [7:0], the reference is not to on-board RAM 13 but may be to off-board RAM 15, that is, RAM on memory array boards 15.

According to the present invention, the comparator outputs two signals, CPU NXM (CPU Nonexistent Memory) and LOW NXM (Low Nonexistent Memory) to the cycle 42 and response 41 PALs. When neither signal is asserted, the memory reference is to on-board RAM 13. In this case the cycle PAL 42 asserts RASIN (Row Address Strobe In) to the DRAM controller 46. The DRAM controller 46 then latches the PA address and outputs the address and control strobes for on-board RAM 13. When CPU NXM is asserted, the memory reference is not to on-board RAM 13, but may be to off-board RAM 15 (RAM on the memory array bus 21). In this case, the attached memory array boards 15 take over the address decoding. If the reference is to one of these memory array boards 15, the selected board will respond to the bus cycle. Otherwise, the BUS NXM (Bus Nonexistent Memory) is asserted to the response PAL 41, causing the PAL 41 to assert system REFERENCE to the chip 16.

If the bus cycle was initiated by the processor chip 10, the chip 16 translates the transaction into a system transaction. If the bus cycle is an incoming system transaction, the chip 16 returns NO ACK on the system bus 18. When LOW NXM is asserted, the memory reference is not to local RAM 13. This causes the response PAL 41 to assert system REFERENCE to the chip 16. If the bus cycle was initiated by the processor chip 10, the chip 16 translates the transaction into a system transaction. If the bus cycle is an incoming system transaction, the response PAL 41 asserts ERR on the local bus 17 and the chip 16 returns NO ACK on the system bus.

The memory controller 12 provides status information to the local bus 17 master (processor 10 or chip 16) during a memory cycle. The bus master uses this status to control the cycle. The response logic for memory cycles is implemented in the status PAL 43 and the response PAL 41.

The status PAL 43 has an internal flip-flop called the Memory Locked bit. This flip-flop sets when an interlock read bus cycle locks the local memory 13. If another interlock read is attempted while the Memory Locked bit is set, the status PAL 43 asserts RETRY. The unlock write mask bus cycle clears the Memory Locked bit. Normally, the interlock read and unlock write mask bus cycles are a paired transaction and are generated by the same bus master.

The status PAL 43 also reports parity errors by asserting ERROR PENDING to the response PAL 41. If parity checking is enabled, the response PAL 41 then asserts ERR to the bus master to indicate abnormal termination of the bus cycle. ERROR PENDING remains asserted until cleared by a subsequent D-stream (data stream) read or write cycle.

If a bus cycle encounters no parity errors or is not restricted by a previous read lock, it may complete successfully. In this case the response PAL 41 asserts RDY to the bus master. If the processor chip 10 is the bus master, the response PAL 41 asserts RDY during the first response window in the cycle. This allows the processor 10 to execute a 400-ns (nominal) cycle. If the chip 16 is the bus master, the response PAL 41 asserts RDY when read data is valid on the local bus 17, which results in a long bus cycle of 500 ns (nominal). The chip 16 receives RDY early in the cycle for writes, which allows a minimum length bus cycle.

Table 7 describes the sequence of responses of memory controller 12 to any given event in a memory cycle. Table 8 describes the response.

A DP8409 DRAM controller 46 drives the on-board RAM 13. The DRAM controller 46 generates the address, row and column address strobes, and refresh count for the 256 Kbyte DRAM chips.

The Column Address Strobe (CAS) from the DRAM controller 46 is divided into four CAS signals that are gated by BM [3:0]. When a Byte Mask line is asserted, the corresponding byte in the longword is to be read or written.

Memory access and refresh cycles are controlled externally to the DRAM controller 46. The control logic is implemented in the cycle PAL 42 and the status PAL 43.

The status PAL 43 uses the synchronized Address Strobe (AS) and Refresh signals to arbitrate between memory and refresh cycles. When there is contention (both signals assert on the same clock edge), the memory cycle wins. If Synchronized Refresh alone is asserted, the PAL asserts the Refresh line to initiate a refresh cycle.

The status PAL 43 allows memory refreshing while a windowed local bus 17 transaction is outstanding on the system bus 18. While waiting for the transaction to complete, the processor 10 asserts AS to interlock the local bus 17. Normally, this would prevent a refresh cycle from occurring. However, the status PAL 43 can detect this situation, and will allow refreshing even though AS is asserted.

The cycle PAL 42 controls the address timing for both on-board 13 and off-board RAM 15. When on-board RAM 13 is referenced, the PAL 42 asserts RASIN (Row Address Strobe In) and CASIN (Column Address Strobe In) to the DRAM controller. The DRAM controller 46 synchronizes these signals internally and outputs RAS (Row Address Strobe) and CAS (Column Address Strobe) to the RAM. When the memory cycle is a possible reference to off-board RAM 15 (CPU NXM asserted and LOW NXM deasserted), the cycle PAL 42 asserts ARR RASIN (Array Row Address Strobe In) on the array bus. ARR RASIN serves as the start pulse for off-board memory and refresh cycles.

Under certain conditions, the retry PAL 40 suspends and then restarts (retries) local bus 17 transactions initiated by the processor 10. For a detailed description of the retry sequence mechanism, reference should be made to the description below of co-pending application Ser. No. 496,097 entitled "Method and Apparatus for Suspending and Restarting A Bus Cycle" filed on even date herewith in which this is claimed.

Byte parity on the local bus 17 is generated for processor initiated transactions, by the processor transceivers 36, and for chip 16 initiated transactions, by the chip 16 transceivers. Parity is checked by the chip 16 transceivers 34. When the parity error signal is generated, the memory controller 12 generates two parity error reporting signals called PERR and PARITY ERROR.

Whenever a parity error is detected, PERR is asserted. This sets the Parity Error bit in the Memory Status Register 47 (MCSR0) when the cycle PAL 42 asserts DONE to indicate that the bus cycle has finished. The Parity Error bit is only a status bit. It does not cause the operating system to respond to the parity error.

PARITY ERROR is a logical AND of PERR and of the Parity Enable signal from the Memory Control Register 47 (MCSR1). If the Parity Enable signal is asserted, PARITY ERROR is asserted whenever PERR is asserted to the status PAL 43. The status PAL 43 then asserts Error Pending to the response PAL 44. The response PAL 41 terminates the bus cycle by asserting ERR to the bus master.

The memory controller 12 has four control and status registers. The memory status register reports parity errors and enables latching of the address at which a parity error occurs. The memory control register controls parity reporting and indicates whether local memory was successfully sustained by battery backup when main system power was off. The memory configuration register indicates the highest Mbyte boundary of the last memory array board 15 that was accessed. The memory starting address offset/error address register on power-up or reset is written to by the primary processor's memory configuration routine. The routine writes the starting address offset for local memory into this register. Under certain conditions, the register indicates the address of a memory location that was read with bad parity.

The memory array boards and associated buses are described in detail in application Ser. No. 093,580, entitled "Memory Configuration System", filed on even date herewith and assigned to the same assignee as the present invention.

As described in connection with FIGS. 4 and 5, additional memory may be connected to the memory array bus. However, in some instances, additional memory is not required but instead there is a need of an I/O bus, e.g. when a connection to an external processor 400 must be made. I n accordance with the present invention, such I/O can be accommodated using the memory array bus, thereby eliminating the need for an additional I/O bus. In a single board processor this reduction in hardware is of great significance.

In accordance with the present invention, to accomplish this purpose, the configuration bus 407 is used for dual purposes. When used with memory, it provides unidirectional configuration information as described above. However, when used as an I/O bus, e.g., with an external processor 400, the bus becomes partially a bi-directional bus. The first four lines of the bus are taken off and provided into a bi-directional buffer 401. Inputs to this buffer are the signals CS0 and CS1 from the local bus 17. Outputs, which are obtained from the external processor 400 and provided into the buffer 401, are signals designated as STALL and RETRY. -The RETRY signal is the same signal described above and provides an additional input into the RETRY PAL 40. The seventh bit of the configuration bus is utilized to indicate whether or not memory or an external processor is present on the end of the bus. If a memory is present, this bit is not asserted. If a processor is present the bit is asserted and this signal is provided as an input to buffer 401 to turn on the buffer and automatically convert what was a configuration bus into a bi-directional bus which permits operation as an I/O bus with an external processor.

FIG. 6 is helpful in further understanding this type of operation. There are shown two computer systems, system 412 which includes the previously described processor 415, one megabyte of memory 421, interface 417 to interface with an external bus 419, status PAL 425 which controls the lock and retry logic and the retry PAL 423 itself. It of course includes all of the other hardware described above in connection with FIG. 4. Only this much of the apparatus is set out in block diagram form in order to give a better understanding of the reconfigured bus.

At the other side is a system 413 including one or more processors 431 with, for example, 16 megabytes of dynamic random access memory 433 and lock and retry logic 435. Shown with the figure is a definition of what the configuration bus bits mean when configured with an external processor present.

As shown, the seventh bit indicates that the external processor is present. Bits 4 to 6 are not used. Bits 2 and 3 are the internal bus signals CSI and CS0 which control the lock bit protocol. (See Table 1.) Finally, bit 1 is a RETRY signal and bit 0 is a STALL signal. Bits 2 and 3 are outputs from system 412 to system 413. The remaining bits 0, 1, and 7 are inputs from the system 413 to the system 412.

The computer system 413 can be a graphics system-of the type described in copending application Ser. No. 085,081, entitled High Performance Graphics Workstation, filed Aug. 13, 1987 and assigned to the same assignee as the present invention, the disclosure of which is hereby incorporated by reference. For a better understanding of the general use of STALL and RETRY, reference may be made to the aforementioned copending application Ser. No. 534,720.

As previously explained, RETRY can be asserted within the system 412 under a number of conditions. First of all RETRY may be asserted, the lock bit is set. The lock bit is set if the processor 415 is to be denied access to the memory 421. Such can be initiated by the memory itself or may be caused because the bus 419 is accessing the memory. RETRY can also be asserted by the interface device 16 if it has an incoming transaction at the same time there is a request for an outgoing transaction from processor 10. As previously explained, what occurs during RETRY is that the cycle then present and being carried out by the processor 415 is aborted and subsequently restarted. In addition, the arrangement of the present invention permits also asserting RETRY remotely from the system 413.

The processor 415 at times requires access to the memory 433 in the system 413. For example, when utilized in the aforementioned graphics system, the memory 433 will be storing data structures. Data structures are quite large and for proper operation it is necessary that a complete data structure be obtained all at one time from memory. Thus, if the processor 431 is accessing memory 433 to obtain a data structure it can interlock its memory 433 by setting a lock bit in the logic 435. If the processor 415 now attempts to access the memory 433, the presence of this lock bit causes a RETRY signal to be sent back over the configuration bus 407 to the system 412 where it will become an input to the RETRY PAL and cause aborting and restarting of the cycle of processor 415 in a manner more fully described below. RETRY may also be asserted as a result of the bus timeout specification of system 413, e.g. in order to refresh memory 433. Processor 415 can also lock out the remote memory through the use of the two CS0 and CS1 bits which are also provided on the configuration bus, e.g. when it is accessing the memory to obtain a data structure.

The STALL signal which is on bit 0 of the configuration bus in effect converts what is a synchronous bus when operated in the memory mode to an asynchronous bus when operated in the I/O mode. A normal memory cycle of the processor 413 operates in a synchronous manner. The processor expects to receive either a ready (RDY) or error (ERR) signal from response PAL 41 at a certain point in the cycle. Because of the fact that when operating with the system 413, it is operating with a system to which it is not synchronized, the bus cycle cannot always be completed at the proper time. What is done to avoid completion of the cycle is to assert the STALL signal on reconfigured bit 0. This may be done selectively or, as in the illustrated embodiment, each time the remote system is addressed. STALL is an input to the response PAL 41 of FIG. 4 and causes the response PAL 41 to delay providing the ERR or RDY output until STALL is no longer asserted. Thus, in response to the STALL signal the processor extends its cycle. When STALL is asserted the cycle is suspended. Once the STALL signal is deasserted, the cycle continues and the response PAL 41 can provide the RDY signal or ERR signal.

The need for some mechanism to resolve potential conflicts between the processor 10 and the requests on bus 18, as well cases where a memory, be it memory 13 or remote memory in a remote computer system 400, is clear in light of the foregoing overview of the system. The Retry method and apparatus which will now be described in detail provides that mechanism. It does so by operating various transceivers to electronically decouple processor 10 from the local bus 17, while at the same time storing the memory address which the processor 10 was using. During the time the processor 10 is decoupled, the potential conflict or deadlock is resolved by allowing the devices other than the processor 10 to perform their task. Upon resolution, Retry allows the processor 10 to resume operation by recoupling it to the local bus 17 and reasserting, from PAL 40, the proper addresses and signals. The method avoids the necessity of reinitializing processor 10 after such a break, which is required by some other systems, and also maximizes processor utilization.

In the illustrated embodiment three events can trigger the occurrence of a Retry.

The first of these events is when an outgoing local bus 17 transaction contends with a system bus 18 transaction. Since bus interface chip 16 cannot translate transactions in both directions at once, the outgoing local bus transaction is suspended and then retried after the incoming system bus 18 transaction has been completed.

The second of these events is when the time-of-year (TOY) clock 25 is read from or written to. TOY clock 25 has both a slow setup time and a slow access time. The Retry sequence is thus well-suited for accessing this device.

The final event which may trigger the Retry sequence is when the processor 10 attempts an interlock read operation on memory, be it local memory 13 or memory in a remote processor coupled as described in connection with FIG. 6, when that memory is locked.

Although these three events trigger Retry in the illustrated embodiment, this should not be understood as a limitation upon the invention. Retry could be implemented with a plurality of other input/output devices or with other computer architectures.

The Retry sequence commences when bus interface chip 16, the status PAL or the remote computer asserts RETRY or when the I/O address PAL asserts TOY.

Referring now to FIG. 7, when RETRY 500 is asserted, due to LOCK 402 being low, retry PAL 40 deasserts the following control signals: (1) AS EN (address Strobe Enable) 404, (2) DBE EN 406 (Data Buffer Enable) and (3) BMEN (Byte Mask Enable) 408. FIG. 7 is a timing diagram illustrating the sequence of these operations.

When the above signals are deasserted, the Byte mask signals become invalid, the address on DAL (29:2) lines is considered invalid, and the DAL transceivers 36 are disabled. In short, the transceivers and buffers that output the processor 10 bus control signals, DAL lines, and byte mask lines onto local bus 17 are all disabled. This removes all processor 10 signals from bus 17 and electronically isolates the processor 10 chip. At the same time that the previous signals are deasserted, the bus interface chip 16 stores the address output of processor 10 in its Retry address latch.

An internal flip-flop in retry PAL 40, called Retry Pending 419, is asserted (see FIG. 7A) to latch the address stored in bus interface chip 16 and to indicate to the retry PAL 40 that it must later restart the suspended local bus transaction. A retry pending input to Response PAL 41 prevents it from asserting RDY or ERR.

If the retry sequence was triggered in response to a signal from bus interface chip 16, in response to control signals asserted by processor 10, prior to asserting RETRY 500, bus interface chip 16 asserts DMR 420 (DMA Request), shown in FIG. 7A, to request that control of local bus 17 be given to the bus interface chip 16, as opposed to processor 10. In a normal operational sequence, processor 10 would assert UV DMG (processor DMA Grant) to the retry PAL 40 causing the retry PAL 40 to assert DMG (DMA Grant) to bus interface chip 16. However, in the Retry sequence, processor 10 is waiting for acknowledgment from the stalled local bus 17 transaction and cannot assert UV DMG. Retry PAL 40 senses that a retry is in progress, and takes over this function, and asserts DMG 422 to give the bus interface chip 16 bus control. The bus interface chip 16 deasserts RETRY and takes control of the local bus 17 issuing control signals 424 and completes the incoming system bus 18 transaction. Then, the bus interface chip 16 gives up control of the local bus 17 by deasserting DMR at 430, which causes the retry PAL 40 to deassert DMG at 432. Retry PAL 40 then restarts the stalled local bus 17 transaction with control signals at 434.

If the retry was in response to a TOY clock 25 reference, retry PAL 40 deasserts the DAL transaction control signals and asserts a TOY select. Retry PAL 40 then restarts the stalled local bus 17 transaction.

If the Retry was originally in response to a lock bit, the response PAL 41 also restarts the transaction when the lock bit is cleared and DMG is deasserted. This is shown in FIG. 7B. The interface 16, for example, asserts DMG at 450 and the processor asserts DMG at 451. At 453, the interface 16 asserts the control signals starting the bus (read) cycle, indicating that it is a read lock. The lock bit is set in status PAL 43 by CS (1:0) at 455. When processor 10 asserts control signals and tries to read from memory at 457, Retry 500 is asserted by status PAL 43 causing Retry AL 40 to deassert the control signals at 458 and set the retry pending bit at 459. Subsequently, when interface 16 does a write unlock, which completes at 461, the lock bit is cleared at 463, and Retry PAL 40 at 465 asserts the control signals that were deasserted at 458.

If the retry was initiated over the bus 407 from a remote device, the transaction is restarted when RETRY is no longer asserted.

Returning to FIG. 7, the local bus 17 transaction is restarted by the retry PAL 40 asserting Byte Mask Enable 408, which enables the processor 10 Byte Mask and Control Status lines. The retry PAL 40 then asserts RLOE (Retry latch Output Enable) to output the stored address from bus interface chip 16.

Next, RLOE Delay 412, a timing signal internal to retry PAL 40 is asserted, causing the Address Strobe Enable 404, and Data Buffer Enable 406 to assert at staggered intervals as shown in FIG. 7.

These enable signals gate processor 10 control signals to the local bus 17 at the proper intervals. The Retry sequence is completed when the response PAL 41 asserts either RDY or ERR to the processor 10 chip to indicate a successful or unsuccessful completion of the transaction.

By electronically isolating processor 10 from the local bus 17 during the previously described events, potential conflicts with other devices on the bus are avoided. Since the processor 10 is only suspended and the Retry PAL regenerates the beginning of the cycle, operation can resume without reinitializing the chip and with only a minimal loss of available processing capacity.

The logic equations for retry PAL 40 are as follows: