WO1988008571A1

WO1988008571A1 - Method and apparatus for initiating transactions on a multiprocessor computer system employing multiple lock indications

Info

Publication number: WO1988008571A1
Application number: PCT/US1988/001359
Authority: WO
Inventors: Richard B. Gillett, Jr.; Douglas D. Wiiliams
Original assignee: Digital Equipment Corporation
Priority date: 1987-05-01
Filing date: 1988-04-22
Publication date: 1988-11-03
Also published as: CA1303233C; DE3882988D1; KR930000992B1; AU601285B2; DE3882988T2; EP0358690B1; JPH01502628A; AU1725088A; KR890702118A; US4941083A; EP0358690A1

Abstract

A processor node providing exclusive read-modify-write operations in a computer system having multiple processors interconnected by a pended bus (25) and employing multiple lock bits. The processor generates an interlock read command which is transmitted as a transfer over the pended bus (25) to a memory, or I/O node. Acknowledge confirmations are transmitted by the memory back to the processor two bus cycles after each bus cycle of the command transfer. The command transfer, including an interlock read command, is stored in an input queue (306) in memory and is processed in turn by the memory. A first interlock read command to a specified memory location causes a lock bit to be set in (310) for that location and a first type of response message including the contents of the specified location to be generated by the memory and stored in an output queue (318). The memory obtains access to the pended bus through an arbitration process and transmits a response message including the contents of the memory location specified in the interlock read command at an unspecified time after initiation of the interlock read command. A subsequent interlock read command from the processor to the same memory location will result in a denial of access (314) to the specified location and in the generation of a second type of response message by the memory which indicates that the specified location is locked.

Description

METHOD AND APPARATUS FOR INITIATING TRANSACTIONS ON A

MULTIPROCESSOR COMPUTER SYSTEM EMPLOYING

MULTIPLE LOCK INDICATIONS

II. BACKGROUND OF THE INVENTION

The invention relates to computer systems and, more particularly, to computer systems having multiple processors interconnected by a pended bus•

Modern computer systems may have multiple processors, memory resources, and input/output (I/O) devices interconnected by a common bus to achieve high total computational power. Such construction can provide very powerful systems capable of executing many millions of instructions per second. However, the interconnection of multiple processors can create difficul¬ ties when there is a need to perform an instruction sequence known as a read- odify-write (RM ) operation. In an RMW opera¬ tion, one processor retrieves data from a memory location, per¬ forms an operation on the data, and writes the modified data back to the o-riginal memory location. Unpredictable results affecting data integrity can occur if one processor has started an RMW operation for one memory location, and a second proces¬ sor attempts an RMW operation for the same memory location in the time period between the "read" operation of the first pro¬ cessor's RMW operation and the "write" portion of that RMW operation.

One way to prevent multiple processors 'from per¬ forming RMW operations on the same memory locations is to pro¬ vide an "interlock read" capability. This involves the use of- a "lock" indicator, such as a lock bit, which is set when a "read" portion of an RMW operation is performed and which is reset after the "write" portion of the RMW operation is completed. A second processor attempting to initiate an RMW operation on a location in memory when the lock bit is set will cause the memory to return lock: status information by means of a "busy" or "retry" confirmation a predetermined number of bus cycles after the second processor generated its interlock read command. The busy confirmation indicates to the processor that the second interlock read command was not accepted by the memo¬ ry.

The interlock read operation alleviates problems caused by multiple processors each attempting to perform an RMW operation. Processors are granted equitable access to the bus for such interlock read operations by arbitration processes using, for example, a round-robin algorithm. However, perfor¬ mance bottlenecks can still occur. For example, under certain bus traffic conditions, a specific processor may repeatedly encounter locked memory locations and will be unable to obtain needed access to memory resources in a timely manner. Such problems could be reduced by providing multiple lock bits for a memory node with each lock bit associated with a portion of the memory node rather than with the whole memory node. Such mul¬ tiple lock bits would provide finer "granularity" of interlocked read operations on a memory node, tying up a smaller portion of memory after an interlock read operation. This solution would also permit a higher success rate of RMW operations, thus improving system throughput. However, imple¬ menting multiple lock bits on prior art pended bus multiprocessor systems would result in unacceptably complex circuitry for- detecting and transmitting lock status informa¬ tion.

Although the preceding discussion has emphasized the operation of a computer system employing processor nodes, memo¬ ry nodes, and I/O nodes, a more general discussion of such a system is in the terms of commander nodes, that is, nodes which initiate a transaction on a bus, and responder nodes, that is, nodes which respond to a transaction initiated by a commander node. At various times, a single device can function as either a commander node or a responder node. It is desirable to provide a computer system in which devices are interconnected over several busses, each having different characteristics. However, this was extremely diffi¬ cult to accomplish in prior ar pended bus systems using interlock read operations, in which lock status information was transmitted with a fixed time relationship to the initial interlocked read command.. . III. SUMMARY OF THE INVENTION

It is therefore an object of the present invention to provide a commander node in a multiprocessor system having mul¬ tiple lock bits and simplified circuitry for transmi ting lock status information.

It is a further object of the present invention to provide a commander node in a multiprocessor pended bus com¬ puter system having interlock read operations in which lock status information is not transferred with a fixed time rela¬ tionship to the initial interlock read command.

Additional objects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. The objects and advantages of the invention may be realized and attained by means of the in¬ strumentalities and combinations particularly pointed out in the appended claims.

The present invention overcomes the problems and dis¬ advantages of the prior art by providing a commander node re¬ sponsive ^"to an acknowledge confirmation received a predeter¬ mined time after an interlock read command and to a lock status message received at an unspecified time after the interlock read message.

In accordance with the principles of the present invention, there is provided apparatus for initiating exclusive read-modify-write operations on a pended bus, the operations having a set of distinct transactions on the pended bus including an interlock read command to retrieve information stored at a specified location and restrict access to the stored information by subsequent interlock read commands, and an unlock write command to store information in the specified -al¬

location and restore access to the stored information. The apparatus comprises means for initiating, over the pended bus to a responder node containing the specified location, an interlock read command to retrieve information from the specified location and an unlock write command to write modi¬ fied information in the storage location. The apparatus fur¬ ther comprises means for receiving from the responder node, at an unspecified time subsequent to initiation of the interlock read command, a lock status messages indicating whether the interlock read command was executed by the responder node.

In another aspect, the invention provides a methoα for initiating exclusive read-modify-write operations bn a pended bus, the operations having a set of distinct transac¬ tions on the pended bus including an interlock read command to retrieve information ^"stored at a specified location and re¬ strict access to the stored information by subsequent interlock read commands, and an unlock write command to store information in the specified location and restore access to the stored in¬ formation. The method comprises the steps of initiating, over the pended bus to a responder node containing the specified lo¬ cation, an interlock read command to retrieve information from the specified location; and receiving from the responder node, at an unspecified time subsequent to initiation of the interlock read command, a lock status messages indicating whether the interlock read command was executed by the re¬ sponder node.

The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate one embodiment of the invention, and, together with the descrip¬ tion, serve to explain the principles of the present invention. IV. BRIEF DESCRIPTION OF THE DRAWINGS

Fig. 1 is a block diagram of a data processing system using the present invention;

Fig. 2 is a block diagram of a node in the data pro¬ cessing system of Fig. 1;

Fig. 3 is a timing diagram showing timing signals used in the data processing system of Fig. 1; Fig . 4 is a block diagram of the data interface 61 in the node of Fig. 2;

Fig. 5 is a block diagram of the arbiter in the data processing system of Fig. 1;

_ Fig . 6 is a timing diagrams showing signals appearing on the system bus of Fig. 1 during an interlock read transac¬ tion; -** ^•

Fig. 7 is a block diagram of a processor node in the data processing system of Fig . 1 ;

Fig. 8 is a block diagram of a memory node in the data processing system of Fig. 1; and

Fig .9 is a block diagram of a lock controller in the memory node of Fig. 8.

V. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENT A. System Overview

Fig. 1 shows an example of a data processing system 20 which embodies the present invention. The heart of system 20 is a system bus 25 which is a synchronous bus that allows communication between several, processors, memory ^'subsystems, and I/O systems. Communications over system bus 25 occur syn¬ chronously using periodic bus cycles. A typical bus cycle time for system bus 25 is 64 nsec.

In Fig. 1, system bus 25 is coupled to two processors 31 and 35, a memory 39, one I/O interface 41 and one I/O unit 51. I/O unit 53, is coupled to system bus 25 by ^"way of I/O bus 45 and I/O unit interface 41.

A central arbiter 28 is also connected to system bus 25 in the preferred embodiment of data processing system 20. Arbiter 28 provides certain timing and bus arbitration signals directly to the other devices on system bus 25 and shares some signals with those devices.

The implementation shown in Fig. 1 is one which is presently preferred and^' should not necessarily be interpreted as limiting the present invention. For example, I/O unit 53 could be coupled^■ directly to system bus 25, and arbiter 28 need not operate in the manner described for the present invention.

In the nomenclature used to describe the present invention, processors 31 and 33, memory 39, and I/O interface 41, and I/O device 51 are all called nodes. A "node" is defined as a hardware device which connects to system bus 25.

According to the nomenclature used to describe the present invention, the terms "signals" or "lines" are used in¬ terchangeably to refer to the names of the physical wires. The terms "data" or "levels" are used to refer to the values which the signals or lines can assume.

Nodes perform transfers with other nodes over system bus 25. A "transfer" is one or more contiguous cycles that share a common transmitter and common arbitration. For exam¬ ple, a read operation initiated by one node to obtain informa¬ tion from another node on system bus 25 requires a command transfer from the first to the second node followed by one or more return data transfers from the second node to the first node at some later time.

A "transaction" is defined as the complete logical task being performed on system bus 25 and can include more than one transfer. For example, a read operation consisting of a command transfer followed later by one or more return data transfers is one transaction. In the preferred embodiment of system bus 25, the permissible transactions support the trans¬ fer of different data lengths and include read, write (masked) , interlock read, unlock write, and interrupt operations. The difference between an interlock read and a regular or noninterlock read is that an interlock read to a specific loca¬ tion retrieves information stored at that location and re¬ stricts access to the stored information by subsequent interlock read commands. Access restriction is performed by setting a lock mechanism. A subsequent unlock write command stores information in the specified location and restores access to the stored information by resetting the lock mecha¬ nism at that location. Thus, the interlock read/ lock write operations are a form of read-modify-write^' operation.

Since ^"system bus 25 is a "pended" bus, it fosters efficient use of bus resources by allowing other nodes to use bus cycles which otherwise would have been wasted waiting for responses. In a pended bus, after one node initiates a trans¬ action, other nodes can have access to the bus before that transaction is complete. Thus, the node initiating that trans¬ action does not tie up the bus for the entire transaction time. This contrasts with a non-pended bus in which the bus is tied up for an entire transaction. For example in system bus 25, after a node initiates a read transaction and makes a command transfer, the node to which that command transfer is directed may not be able to return the requested data immediately. .» Cycles on bus 25 would then be available between the command transfer and the return data transfer of the read transaction. System bus 25 allows other nodes to use those cycles.

In using system bus 25, each of the nodes can assume different roles in order to effect the transfer of information. One of those roles is a "commander" which is defined as a node which has initiated a transaction currently in progress. For example, in a write or read operation, the commander is the node that requested the write or read operation; it is not nec¬ essarily the node that sends or receives the data. In the pre¬ ferred protocol for system bus 25, a node remains as the com¬ mander throughout an entire transaction even though another node may take ownership of the system bus 25 during certain cycles of the transaction. For example, although one node has control of system bus 25 during the transfer of data in re¬ sponse to the command transfer of a read transaction, that one node does not become the commander of the bus. Instead, this node is called a "responder."

A responder responds to the commander. For example, if a commander initiated a write operation to write data from node A to node B, node B would be the responder. In addition, in data processing system 20 a. node can simultaneously be a commander and a responder.

Transmitters and receivers are roles which the nodes assume in an individual transfer. A "transmitter" is defined as a node which is the source of information placed on system bus 25 during a transfer. A "receiver" is the complement of the transmitter and is defined as the node which receives the information placed on system bus 25 during a transfer. During a read transaction, for example, a commander can first be a transmitter during the command transfer and then a receiver during the return data transfer. When a node connected to system bus 25 desires to become a transmitter on system bus 25, that node asserts one of two request lines, CMD REQ (commander request) and RES REQ (re¬ sponder request) , which are connected between central arbiter 28 and that particular node. In general, a node uses its CMD REQ line to request to become commander and initiate transac¬ tions on system bus 25, and a node uses its RES REQ line to become a responder to return data or message to a commander. Generally, central arbiter 28 detects which nodes desire access to the bus (i.e., which request lines-are asserted). The arbi¬ ter then responds to one of the asserted request lines to grant the corresponding node access to bus 25 according to a priority algorithm. In the preferred embodiment, arbiter 28 maintains two independent, circular queues: one for the commander re¬ quests and one for the responder requests. Preferably, the re¬ sponder requests have a higher priority than the commander re¬ quests and are handled before the commander requests.

The commander request lines and 'responder request lines are considered to be arbitration signals. As illustrated in Fig. 1, arbitration signals also include point-to-point con¬ ditional grant signals from central arbiter 28 to each node, system bus extend signals to implement multi-bus cycle trans- • fers, and system bus suppression signals to control the initiation of new bus transactions when, for example, a node such as memory is momentarily unable to keep up with traffic on the system bus.

Other types of signals which can constitute system bus 25 include information transfer signals, respond signals^', control signals, console/front panel signals, and a few miscel¬ laneous signals. Information transfer signals include data signals, function signals which represent the function being performed on the system bus during a current cycle, identifier signals identifying the commander, and parity signals. The ^"re¬ spond signals generally include acknowledge or confirmation signals from a receiver to notify the transmitter of the status of the data transfer.

Control signals include clock signals, warning sig¬ nals, such as those identifying low line voltages or low DC voltages, reset signals used during initialization, node fail¬ ure signals, default signals used during idle bus cycles, and error default signals. The console/front panel signals include signals to transmit and receive serial data to a system con¬ sole, boot signals to control the behavior of _a boot processor during power-up, signals to enable modification of the erasable PROM of processors on system bus 25, a signal to control a RUN LIGHT on the front panel, and signals providing battery power to clock logic on certain nodes. The miscellaneous signals, in addition to spare signals, include identification signals which allow each node to define its identification code.

Fig. 2 • shows an example of a node 60 connected to system bus 25. Node 60 could be a processor, a memory, an I/O unit or an I/O interface. In the example shown in Fig. 2, node 60 includes node specific logic 65, a node bus 67, and a system bus interface 64 containing a data interface 61 and a clock decoder 63. Preferably,, data interface 61, clock decoder 63, and node bus 67 are standard elements for nodes connected to system bus 25. The node specific logic 65, which uses differ¬ ent integrated circuits from system bus interface 64, prefer¬ ably includes, in addition to the circuitry designed by a user to carry out the specific function of a node, standard cir¬ cuitry to interface with the node bus 67. In general, data interface 61 is the primary logical and electrical interface between node 60 and system bus 25, clock decoder 63 provides timing signals to node 60 based on centrally generated clock signals, and node bus 67 provides a high speed interface between data interface 61 and node specific logic 65.

In the preferred embodiment of node 60 and system .bus interface 64 shown in Fig. 2, clock decoder 63 contains control circuitry for forming signals to be placed on system bus 25 and processes clock signals received from central arbiter 28 to obtain timing signals for node specific logic- 65 and data interface 61. Since the timing signals obtained by clock • decoder 63 use the centrally generated clock signals, node 60 will operate synchronously with system bus 25.

Fig. 3 is a timing diagram showing one bus cycle, the clock signals received by clock decoder 63, and certain of the timing signals generated by clock decoder 63. The clock sig¬ nals received by clock decoder 63 include a Time H signal, a Time L signal, and a Phase signal as shown in Fig. 3. Time H and Time L are inverses of the fundamental clock signals and the Phase signal is obtained by dividing the fundamental clock signal by three. The timing signals generated by clock decoder 63 include C12, C23, C34, C45, C56 and C61, all of which are shown in Fig. 3. Those timing signals required by data inter¬ face 61, which occur once per bus cycle, are provided to data interface 61, and a complete set of timing signals, including ^'equivalent ones of the timing signals provided to data inter¬ face 61, is buffered and provided to the node specific logic 65. The purpose of buffering is to insure that node specific logic 65 cannot adversely effect the operation of the system bus interface 64 by improperly loading the timing signals. Clock 63 uses the clock signals to create six subcycles for each bus cycle and then uses the subcycles to create the six timing signals CXY, where X and Y represent two adjacent subcycles which are combined to form one timing signal.

Each node in the sytem bus has its own corresponding set of timing signals generated by its clock decoder 63. While nominally the corresponding signals occur at exactly the same time in every node throughout the system, variations between clock decoder 63 and other circuitry in multiple nodes intro¬ duce t-iming variations between corresponding signals. These timing variations are commonly known as "clock skew. "

Fig. 4 shows a preferred embodiment of data interface 61. Data interface 61 contains both temporary storage cir¬ cuitry and bus driver circuitry to provide a bidirectional and high speed interface between each of the lines of node bus 67 and each of the lines of system bus 25. As shown in Fig. 4, data interface 61 preferably includes storage elements 70 and 72 and system bus driver 74 to provide a communication path from node bus 67 to system bus 25. Data interface 61 also in¬ cludes storage element 80 and node bus driver 82 to provide communication path from system bus 25 to node bus 67. As used in the description of data interface 61, the term "storage ele¬ ment" refers generally to bistable storage devices such as transparent latch or a master-slave storage element, and not to a specific implementation. Persons of ordinary skill will rec¬ ognize which types of storage elements are appropriate.

As shown in Fig. 4, storage element 70 has an input connected to receive data from node bus 67 and an output con- ^~ nected to the input of storage element 72. The output of stor¬ age element 72 is connected to an input of system bus driver 74 whose output is connected to system bus 25. Storage elements 70 and 72 are controlled by node bus control signals 76 and 78, respectively, which are derived from the timing signals gener¬ ated by clock decoder 63. Storage elements 70 and 72 provide a two-stage temporary storage for pipelining data from node bus 67 to system bus 25. Different numbers of storage stages can also be used.

System bus driver 74 is controlled by system bus driver enable 79. According to the state of the system bus driver enable 79, the input of system bus driver 74 either is coupled to its output, thereby transferring the data at the output of storage element 72 to system bus 25, or decoupled from that output. When system bus drive enable 79 decouples the input and output of the system bus driver 74, system bus driver 74 presents a high impedance to system bus 25. The sys¬ tem bus drive enable 79 is also generated by clock decoder 63 in accordance with clock signals received from system bus^' 25 and control signals received from the node specific 'logic 65.

Storage element 80 has an input terminal connected to system bus 25 and an output terminal connected to an input of node bus driver 82. The output of node bus driver 82 is con¬ nected back to node bus 67. Storage element 80, preferably a transparent latch, is controlled by a system bus control signal 85 which is derived. from the timing signals generated by clock decoder 63. A node bus drive signal 87 controls node bus driv¬ er 82 similar to the manner in which system bus drive signal 79 controls system bus driver 74. Thus, in response to node bus driver signal 87, node bus driver 82 either couples its input to its output or decouples its input from its output and pro¬ vides a high impedance to node bus 67. In order to explain how data is transferred over sys¬ tem bus 25, it is important to understand the relationship between system bus drive enable 79 and control signal 85. In the present embodiment, this relationship is shown in Fig. 3. System bus drive enable 79 is nominally driven from the begin¬ ning to the end of a bus cycle. The new data become available for receipt from system bus 25 at some time later iα the bus cycle after driver propagation and bus settling time has oc¬ curred. In the present embodiment, storage element 80 is a transparent latch. Control signal 85 is logically equivalent to clock C45. The bus timing assures that system bus 25 data is available for receipt sometime prior to the deassartion of control signal 85. Storage element 80 stores bus data that is stable at least a set-up time prior to the deassertion of con¬ trol signal 85 and remains stable a hold time after the deassertion of control signal 85.

Node bus 67 is preferably a very high speed data bus which allows bidirectional data transfer between the node spe¬ cific logic 65 and system bus 25 by way of data interface 61. In the preferred embodiment of node 60 shown in Fig. 2, node bus 67 is a an interconnect system consisting of point-to-point connections between the system bus interface 64 and the node specific logic 65. In accordance with the present invention, however, there is no requirement for such point-to-point inter¬ connection.

Fig . 5 shows a preferred embodiment of the central arbiter 28 which is also connected to system bus 25. Central arbiter 28 provides the clobk signals for system bus 25 and grants ownership of the bus to the nodes on system bus 25. Central arbiter 28 preferably includes an arbitration circuit 90, a clock circuit 95, and a oscillator 97. Oscillator 97 generates the fundamental clock signals. Clock 95 provides timing signals for arbitration circuit 71 and the basic Time H, Time L, and Phase clock signals for timing on system bus 25. Arbitration circuit 71 receives the commander and responder re¬ quest signals, arbitrates conflicts between nodes desiring access tδ system bus 25, and maintains the queues referred to above for the commander and responder requests. Arbitration circuit 71 also rovides certain control si nals to clock 95. B. Discussion of Interlock Operations As briefly discussed above, a number of different types of transactions are allowed on bus 25. In each case, the transaction is composed of one or more separate transfers from one node to another. When the responder node successfully re¬ ceives a command transfer during one or more bus cycles, it generates an acknowledge confirmation at the beginning of the second bus cycle after each cycle of the transfer. Such acknowledgement signals do not indicate successful execution of the command contained in the original transfer, but merely in¬ dicate that the transfer was successfully placed in an input queue in the desired responder node. The transactions relevant to the present invention will be briefly described below.

A read transaction is used to move data in four-byte, eight-byte, sixteen-byte, or thirty two-byte blocks from a spe¬ cific location in a responder node managing a region of address space to a commander node. In the preferred embodiment, memory and I/O operations are referenced to a common address space. A responder node can be either a memory node, a processor node, or an I/O node.

Interlock read transactions are similar to read transactions. However, the exact effect of an interlock read transaction depends on the state of lock tags in the responder node in a manner to be described later in greater detail. Lock tags prevent access to locations or groups of loca-tions in address space. The effect of lock tags can be understood by vis^'ualizing address space of system 20 as appearing across a metallic "blackboard. " Lock tags operate like magnetic tags removably placed on top of locations or groups of locations on the address space "blackboard. " If the location in address space specified in an interlock read transaction is already covered with a lock tag, that is, if the specified address space is "locked, " the responder node responds to the interlock read request with a "locked" response message and no data is returned. This signifies to the commander that the location in address space specified in the interlock read command is not accessible. This locked response message is transmitted to the commander after the responder node services the interlock read command and after the responder node can gain access to bus 25. Thus, the commander receives the locked response message at an unspecified time after the command transfer of the interlock read transaction.

If the specified location is not locked, that is, not associated with a lock tag, information stored in the address specified in the interlock read command is returned in a re¬ sponse message to the commander node which generated tne interlock read command. The responder node also attaches a lock tag to the location in address space specified in the interlock read command, thus denying access to the specified location in address space to subsequent interlock read commands.

The unlock write transaction is the complement to the interlock read" transaction. When a commander node successfully completes- the read and modify location in a read-modify-write operation, it must unlock the location in address space which it tem orarily locked by an interlock read command. The com¬ mander accomplishes this operation by performing an unlock write transaction to the specified location in address space to write the appropriately modified data into the specified loca¬ tion. The responder node processes the unlock write command by unlocking the address space and writing the data as requested. The lock tag is then cleared in a manner to be described in greater detail.

Messages transmitted over bus 25 during an interlock read command transfer includes data on sixty four data lines. That data contains a four-bit command field, a. two-bit length field which specifies the number of words to be transferred from e.g., memory 39 to processor node 31, and a thirty-bit address field which specifies the address location in memory 39 from which data is desired to be read. Other lines of system bus 25 carrying information during an interlock read command include four function lines carrying a four-bit function code indicating a command transfer, six ID lines carrying a six-bit code identifying the commander node which initiated the interlock read command, and three parity lines. As briefly discussed above, system bus 25 includes respond signals which are used by a receiver to indicate suc¬ cessful reception of information placed on the bus by a trans¬ mitter. In the preferred embodiment, the respond signals in¬ clude three identical wire-ORed confirmation (CNF) lines. Three lines are provided since it is extremely important to the integrity of bus transactions that a commander know exactly what a responder has done in response to each command, particu¬ larly in the case of an interlock command or a write to an I/O register. Therefore, a receiver will send either an acknowl¬ edge (ACK) indication by asserting all three CNF lines or a no acknowledge (NACK) signal by not asserting all three CNF lines. Error correction logic is provided in the receiver to determine the true CNF status if all three CNF lines are not received by the receiver at the same logic level.

An ACK confirmation indicates that a responder has accepted information from one cycle of command transfer or that a commander has accepted information from one cycle of a re¬ sponse message. A read command transfer cycle resulting in an ACK confirmation indication indicates that the responder will return a read response message at some later time.

A NACK confirmation returned on the CNF lines indi¬ cates that no receiver has accepted the information from that bus cycle of the command transfer. This could be for three reasons: (1) a parity error has occurred on the system bus 25, (2) the receiver was temporarily unable to accept the command, for example, when the .receiver' s input queue is full, or (3) there is no responder node corresponding to the specified address.

The confirmation indications corresponding to a bus cycle are placed on the CNF lines by the receiver node at the beginning of the second cycle after cycle of the bus cycle.

An example of interlock read transactions will be described in connection with Figure 6. The horizontal axis at the top of Figure 6 indicates successive bus cycles on bus 25. The labels appearing vertically along the left side of Figure 6 indicate groups of lines contained in bus 25, that is, function lines, data lines, ID lines, confirmation lines, and arbitration lines. The entries in the matrix formed by the horizontal and vertical axes of Figure 6 describe the type of data appearing on the specified bus lines during the specified bus cycles.

At bus cycle 0, a first commander node, for example, node 31 of Figure 1, asserts its CMD REQ arbitration request line (one of the point to point lines connected to arbiter 28 and shown in Fig. 1) to arbiter 28. Figure 6 thus indicates a "cmdr #1" request present on arbitration lines of system bus 25 at cycle 1. Assuming that no other node of higher priority is simultaneously requesting access to the bus, processor- 31 obtains bus access on cycle 1 and transmits a message into sys¬ tem bus 25.

During cycle 1 information placed on the function lines of bus 25 indicates.that the information on the bus is command (cmd) information.- The data placed on data lines of bus 25 consist of command and address (c/a) data identifying the current transaction as an interlock read transaction and specifying the address in memory 39 from which data is to be returned to processor 31. The ID lines during bus cycle 1 con¬ tain the identification code of processor (commander/cmdr) node 31 currently transmitting on bus 25.

During bus cycle 2, no information is placed on bus 25 in connection with the present interlock read transac¬ tion.

At the beginning of bus cycle 3, which is two cycles^" (i.e., a predetermined time) after initiation of the interlock read transaction, memory node 39 transmits an ACK confirmation on the confirmation lines of bus 25 if memory 39 successfully received the command transfer transmitted during bus cycle 1. Memory 39 then places the command message in the input queue of memory 39.

The end of bus cycle 3 constitutes the end of the first transfer in the interlock read transaction. Due to the pended nature of transactions on bus 25, the time when the re¬ quested information will be returned from memory 39 to processor 31 is not precisely defined. The response time depends on the length of time required by memory 39 to process the request and the amount of time necessary for system bus 25 to handle additional traffic on bus 25 generated by other nodes. The unspecified nature of the time between the two transfers of an interlock read transaction is indicated by the dotted line in Figure 6 between bus cycles 3 and 4. Thus, although subsequent information is indicated by Figure 6 to occur over bus cycles 4 through 7, it is to be understood that this is only a specific example of the timing involved in an interlock read transaction and that the second transfer of such transaction could occur in any subsequent cycle of bus 25.

Memory 39 processes the interlock read command by re¬ moving the interlock read transfer message from its input queue in turn and examining the address information contained in the transfer. The information is compared to address values stored in lock tags to be more completely described. If there is a match between the stored address values and the address infor¬ mation of the interlock read transfer, this is an indication that the desired address location has been locked by a previous interlock read command. Memory 39 then generates' a locked re¬ sponse message including a "locked" function code, along with other information required for a response message, in an output queue of memory node 39.

If the comparison of address values stored in lock tags with the interlock read transfer address information does not yield a "hit, " that is, if the transferred address does not correspond to any stored address, memory node 39 constructs a response message consisting of a valid read response ^"node such as "good read data" (grdO) code for functions lines, the con¬ tents of the specified address location for the data lines, and the commander identification code of the commander node which initiated the interlock read command for the ID lines. This response message is loaded into the output queue of memory node 39.

When memory 39 has processed the interlock read transaction and generated a response message within its output queue in a manner to be more completely described, memory 39 asserts its RES REQ request line (another point-to-point line shown in Fig. 1) to arbiter 28. The arbitration lines thus carry a responder request (resp) indication as shown in Figure 6 at bus cycle 4. Assuming that no other nodes have higher priority at this time, arbiter 28 grants memory 39 access to bus 25 during bus cycle 5. Memory 39 transmits the response message including "good read data" (.grdO) signal onto function lines of system bus 25, eight bytes (i.e. 64 bits) of data over the data lines of system bus 25 from the memory locations specified by the address field of the initial transfer from processor 31 to memory 39, and the ID of processor Jl onto the ID lines of bus 25 to associate the returning data with the commander (i.e. processor 31) which initially issued the interlock read request.

During bus cycle 6, no traffic appears on system bus 25 related to this interlock read transaction. Finally, the interlock read^"transaction concludes in bus cycle 7 when processor 31 transmits an ACK confirmation onto the confirma¬ tion lines of bus 25.

A second interlock read transaction to the same specified location in memory will result in data appearing on bus 25 as shown in cycles 8-15 of Fig. 6. At cycle 8, a second commander (cmdr #2 ) initiates a commander request to arbitor 28. Bus cycles 9-12 result in traffic on bus 25 identical to cycles 1-4. However, memory 39, upon processing of the re¬ ceived interlock read command, found a match between the address values stored in lock tags and the address transmitted with the interlock read command. Accordingly, a LOC response is presented on function lines of bus 25 at, for example, cycle 13. Bus cycles 14 and 15 are identical to cycles 6 and 7.

C. Description of Processor

Referring now to Figure 7, there is shown a more detailed block diagram of certain elements of the node specific logic 65 in processor 31. Processor node 31 includes, as do all the nodes, bus interface circuit 64. Process node 31 also includes a processor logic 202. As shown in Fig. 7, processor logic 202 includes central processing unit (CPU) circuitry re¬ quired to execute software in a manner well known to those skilled in the art. ' Processor logic 202 also generates command and address information as required by system 20 to execute the necessary application functions as well as to control transfers over system bus 25.

Processor node 31 also includes a parity error check circuit 204 which monitors information on the function, data, ID, and parity lines of system bus 25 received from bus inter¬ face circuit 64 to perform a parity check on those signals in a manner well known in the art. A detected parity error will result in the generation of a parity error indication on signal line 206.

The information on the ID lines is monitored by a comparator circuit 207 which is also supplied with the identi¬ fication code of processor 31 from a hard-wired connection 210 on the backplane which is determined by the position of proces¬ sor 31 in a mounting cabinet.

In a preferred embodiment of the invention, processor 31 comprises means for transmitting, over the pended bus to the responder node at a predetermined time subsequent to the lock status message, an acknowledge confirmation corresponding to the lock status message indicating succesful receipt of the lock response message. As embodied herein, the confirmation means comprises an acknowledge generator 208. Preferably, the acknowledge generator 208 comprises means for transmitting, over the pended bus to the responder node at a predetermined time subsequent to the lock status message, a no-acknowledge confirmation indicating failure to successfully receive the lock status message. The comparison result from comparator 207 is supplied, along with information on the parity error signal line 206, to acknowledge confirmation generator 208. If no parity error was detected and if the ID code received over bus 25 for a response message matches the ID code of processor 31, ACK confirmations are transmitted over the CNF lines of bus 25 by an acknowledge generator 208 by asserting all three CNF lines at the beginning of the second bus cycle after each cycle of a response transfer directed to processor 31. Otherwise, the CNF lines are not asserted, thus constituting a no-acknowledge confirmation.

In accordance with the present invention, the proces¬ sor includes means for receiving from the responder node, at an unspecified time subsequent to initiation of the interlock read command, a lock status message indicating whether the interlock read command was executed by the responder node. As embodied herein, such means comprise bus interface 64 and a response decoder 212. Information on function and data lines of bus 25 is supplied through bus interface 64 to response decoder 212. Decoder 212 is enabled by comparator 207 when a message over bus 25 is intended for processor 31. This is determined by a positive comparison result from comparator 207. If decoder 212 is enabled by comparator 207, decoder 212 extracts function codes from function lines of system bus 25, and for certain function codes, supplies command and data information from data lines of bus 25 to processor logic 202 for appropriate action.

In accordance with the present invention, the proces¬ sor includes means for initiating, over the pended bus to a re¬ sponder node containing the specified location, an interlock read command to retrieve information from the specified loca¬ tion and an unlock write command to write modified information in the storage location. As- embodied herein, the transmitting- means comprises a command generator 214. When processor 31 desires to initiate a transaction on bus 25, command, address, and data information is supplied to command generator 214, along with the ID of this node supplied from connection 210. Command generator 214 prepares a command transfer message and asserts the node CMD REQ arbitration line 216 to indicate to arbiter 28 (not shown in Figure 7) that processor 31 desires access to bus 25 to transmit a commander message. Using an arbitration system, arbiter 28 grants bus access to processor 31 at an unspecified time after the original interlock read transfer.

Upon being granted access, command generator 214 causes bus interface 64 to transmit the command message from command generator 214 to system bus 25.

The responder node to which the interlock read command is directed will generate an acknowledge confirmation two cycles after the interlock read command transfer.

In accordance with the present invention, the com¬ mander node includes means for receiving, over the pended bus from the memory at a predetermined time subsequent to initiation of the interlock read command, an acknowledge con¬ firmation corresponding to the interlock read command. As embodied herein such means comprise bus interface 64, command generator 214, and CNF lines connecting bus interface 64 and command generator 214. As specifically shown in Fig. 7, command generator 214 monitors CNF lines from bus interface 64 to detect the presence of an ACK confirmation on the CNF bus lines exactly two bus cycles after each cycle of a command transfer transmitted by a processor 31 over system bus 25. Failure to detect the presence of an ACK confirmation will result in appropriate corrective action which, in the preferred embodiment, consists of a retransmission by command generator 214 of the previous command. When the transfer is complete, the responder node will process the interlock read command and return a response message on system bus 25. Because of uncer¬ tainties due to traffic on system bus 25 and queue lengths, the responder node will generate a response message at an unspecified time after the command transfer. D. Description of Memory 39

Figure 8 shows a block diagram of memory 39 which may function as a responder node. As can be seen in Figure.8, mem¬ ory 39 includes a command decode and address and parity check circuit 300. Circuit 300 is connected to the bus function, address, and ID lines and performs a parity check in a well- known manner. Circuit 300 also compares the information on bus address lines to the limits of address space served by memory 39, as supplied from a register 302, and supplies the results of this comparison on an address match line 301.. If the address information received over bus 25 is within the range of address space served by memory 39 and if no parity error has occurred, an acknowledge generator 304, connected to circuit 300, will generate^' an ACK confirmation by asserting all three CNF lines at the beginning of the second cycle after the trans¬ mission cycle of a transfer destined for memory 39.

Memory 39 includes an input queue 306 for receiving interlock read commands from the processor nodes. Input queue 306 permits messages, received at high speed over bus 25, to be stored until the relatively slower logic of memory 39 allows such messages to be acted upon. Input queue 306 is enabled to store a message from bus 25 when address information appearing in a message on bus 25 is within the limits of address space for memory 39, as determined by address match signal 301.

The output of input queue 306 is supplied to a decoder 308.which extracts address and command information from messages stored in input queue 306. Although decoder Jϋ8 sup¬ plies multiple indications to decode the various commands and provides address information on a set of parallel signal lines, the address and command outputs of decoder 306 are respectively shown in Figure 8 as bundled lines 309 and 311 for purposes of clarity.

As is well known in the art, information is stored in a plurality of^* discrete locations in memory array 312 identi¬ fied by addresses which may be specified by read and write commands supplied to the array 312.

The address and command information is supplied to lock controller 310 which implements a locking scheme to be described later in greater detail. Address and command infor¬ mation from decoder 308 is also supplied to a memory array 312. Memory array 312 responds to read and write commands to read data from and write data to locations in array 312 specified by address information received from decoder 308.

Lock status signal 314 from controller 310 and memory data from memory array 312 are supplied to response generator 316 which generates an output response message to be described later in greater detail. Response messages from generator 312 are supplied to an output queue 318 for storage until memory 39 obtains access to- the bus through the arbitration process described previously.

Response generator 316 prepares a response message based on data received from memory 312, lock status signal 314 received from controller 310, and command and ID information received from decoder 308. The response message prepared by- generator 316 is either of two types, depending on whether mem¬ ory 39 is permitted to supply the requested data. If the command being responded to is a non-interlock read command, or if the command is an interlock read command and lock status signal 314 is unasserted, response generator 316 prepares a first type of message including the requested contents of the specified location in memory 312. However, if the command is an interlock read command and the lock status line 314 is asserted, response generator 316 prepares the second type of message with a "locked" code on the function lines indicating that the specified address of the interlock read command was in a locked condition and that the requested data is not provided in the response message transmitted by memory 39 in response to the received interlock read command.

When generator 316 has compiled the response message, it is supplied to output queue 318. Output queue 318 signals bus interface 64 that memory 39 desires access to bus 25. The response message is stored in output queue 318 for an unspecified time until such access is obtained.

When memory 39 is granted access to bus 25, the re¬ sponse message contained in output queue 318 is placed on sys¬ tem bus 25 for transmission to the commander node which origi¬ nally generated the command. Since it is not known at what time memory 39 will complete the execution of the command orig¬ inally transmitted by the commander node and since it is fur¬ ther uncertain at what time memory 39 will obtain access to bus 25 to provide either the requested data or the lock status in¬ formation, the ^' lock- status information corresponding to the interlock read command will appear on the function lines of bus 25 at the- commander node at an unspecified time subsequent to initiation of the original interlock read command.

If a commander node fails receive the response mes¬ sage generated by the responder node after a successful interlock read command, the commander node will not generate an acknowledge indication. When the responder node fails to re¬ ceive or acknowledge confirmation from its response message, it will clear the lock bit set by that interlock read command.

E. Description of Lock Controller 310

Referring now to Figure 9, there is shown a more detailed diagram of lock controller 310. In accordance with the invention the lock means includes lock tag means for receiving a selected address corresponding to an address in memory array 39 to which an interlock read command is to be prevented. As embodied herein, the lock tag means include four lock tags 352a, 352b, 352c, and 352d which, together with a logic controller 35o, constitute lock controller 312. It is to be understood that more or fewer lock tags could be provided depending on the specific application. Lock tags 352a-d are identical in construction and operation. For purposes of clar¬ ity, detailed circuitry is shown only for lock tag 352a.

Each lock tag 352a-d includes a storage register 354 for storing a value corresponding to a location in address space of system 20. Register 354 includes an output terminal 356 at which appears the value stored in register 354. Regis¬ ter 354 includes an enable terminal 358 and an input terminal 360 connected to address line 309. Activation of enable termi¬ nal 358 causes register 354 to load the signals present on address lines 309.

Register output terminal 356 is connected to -one input terminal 366 of a comparator 368. Comparator 366 has^' another input terminal 370 connected to address lines 309. An output terminal 372 of comparator 366 constitues a "match" sig¬ nal which -is supplied to one input terminal of a two-input AND gate 374. -The other input terminal of AND gate 374 is con¬ nected to unlock write line 380 of command lines 311. The out¬ put terminal of AND gate 374 is connected to the reset terminal of a latch 382. The output terminal of latch 382 constitutes a LOCK signal which is supplied to one input terminal 387 of a two-input AND gate 386. The other input terminal of AND gate 386 is connected to the match signal output of comparator 368. The output of AND gate 386 constitutes a "hit" signal indicating that an address appearing on address lines is "locked" by lock tag 352a.

The final component of lock tag 352a is a four-input AND gate 388. One input of AND gate 388 is connected to a line 390 of command lines 311 indicating the command currently being processed by memory node 39 is an interlock read command. A second input of AND gate 388 is connected to a clock signal 389 to properly gate the operation of lock tag 352a and prevent race conditions. A third input of AND gate 388 is connected to an "allocate" terminal of logic controller 350 to be described below. The fourth input terminal of AND gate 388 is connected- to an inverted lock status signal 314. The output terminal of AND gate 388 is connected to the enable input 358 of register 354 and the set terminal of latch 382.

Logic controller 350 includes a lock tag allocation circuit 392 which functions as a selection encoder to select an idle lock tag. Allocation circuit 392 determines, by the sta¬ tus of lock bits from lock tags 352a-d, which lock tags are free and assigns one of the available lock tags to provide the locking function by raising an "allocate" signal for the se¬ lected lock tag. If all lock tags are currently assigned, an "all busy" output signal is supplied to one input of a five- input OR gate 394. The other inputs of OR gate 394 are supplied by respective "hit" . signals of lock tags 352a-d.

The operation of lock controller 310 to process an interlock read command' will now be described. Address values on address lines 309 are constantly compared to stored address -values in registers 354. If none of the address values stored in registers 354 is equal to the address values appearing on address lines 309, no match signal is asserted, and no "hit" signal is asserted. Assuming that the "all busy" signal of circuit 392 is also not asserted, no input of OR gate 394 is active and lock status line 314 is not set. Memory array 312 (Figure 8) then supplies the contents of the specified location to response generator 316. The nonassertion of lock status line 314 causes response generator 316 to generate the first type of response message in which a "good read data" code is set in the bits of the response message which will eventually be transmitted over function lines of bus 25 to the commander node which requested them.

The inverted value of lock status line 314 is now supplied to AND gate 388. Circuit 392 is supplying one of the allocate signals of lock tags 352a-d. Since an interlock read is being processed, interlock read line 390 is set by decoder 308 (Figure 8). Thus, when clock signal 389 is activated, AND gate 388 of lock tag 352a is activated to enable register 354. The address values appearing on address lines 309 are stored in register 354 of lock tag 352a. Activation of AND gate 388 also sets latch 382, thus asserting the lock bit of 384 of lock tag 352a. Access to the specified location contained in register 354 of lock tag 352 is now denied to subsequent interlock read commands.

A subsequent interlock read command to_the locked lo¬ cation will result in the following operation. Address values appearing on address lines 309 are equal to the value stored in register 354 of lock tag 352a. The match signal at terminal 372 of lock tag 352a is thus set. Since the lock bit at termi¬ nal 384 for lock tag 352a is set from the previous interlock read operation, both inputs of AND gate 386 are now active, causing the hit signal of lock tag 352a to be asserted. This in turn activa'tes OR gate 394 to activate lock status line 314. Activation of lock status line 314 causes response generator 316 (Figure 8) to generate a response message of a second type in which a LOCKED response code is set on function bits of the message.

The operation of an unlock write command to clear a lock bit will now be described. An unlock write command to a location previously locked will result in the values present on address lines 309 being equal^" to a value stored in a register 354 of a lock tag. For example, assume that an unlock write command has been transmitted to unlock a location locked by lock tag 352a. When the address value appears on address lines 309, the output of comparator 368 will cause a match signal to be set. Since the unlock write line 391 will also be high at this time, AND gate 374 will be activated, causing latch 382 to reset the lock bit signal at output terminal 384. AND gate 386 will be deactivated removing the active hit signal for lock tag 352a from the input terminal of OR gate 394. The data trans¬ mitted with the unlock write command is then written into the specified location in memory.

By providing a lock status message supplied to a pro¬ cessor as a data transfer over the system" bus at an unspecified time after an interlock read command, the present invention allows the functions of transfer acknowledge and lock status transmission to be separated, thereby providing for the use of multiple lock bits without the expense and complexity which would be required if lock status information were required to- be transmitted at a predetermined time or over dedicated lock status lines. This also permits lock status information to be obtained from nodes connected to the system through adapters and buses separate from the system bus.

Throughout the previous discussion, locked memory, or address space, was said to be specified by a "location." It is to be understood that each address stored register may consti¬ tute a range of* addresses, such that a single interlock read command or unlock write command can respectively lock and unlock a range of locations in address, and not merely a single location.

It will be apparent to those skilled in the art that various modifications and variations can be made in the bus interface circuitry and interface of this invention without departing from the spirit or scope of the present invention. The present invention covers such modifications and variations which are within the' scope of the appended claims and their equivalents.

Claims

What is Claimed is:

1. Apparatus for initiating exclusive read-modify-write operations on a pended bus, said operations having a set of distinct transactions on said pended bus including an interlock read command to retrieve ^'information stored at a specified location and restrict access to said stored information by subsequent interlock read commands, and an unlock write command to store information in said specified location and restore access to said stored information, said apparatus comprising: means for initiating, over said pended bus to a re¬ sponder node containing said specified location, an interlock read command to retrieve informa¬ tion from said specified location and an unlock write command to write modified information in said storage location; means fo.r receiving, over said pended bus from said responder node at a predetermined time subse¬ quent" to initiation of said interlock read command, an acknowledge confirmation correspond¬ ing to said interlock read command; and means for receiving from said responder node, at an unspecified time subsequent to initiation of said interlock read command, a lock status mes¬ sages indicating whether said interlock read command was executed by said responder node.

2. An apparatus as recited in claim 1 comprising confirmation means for transmitting, over said pended bus to said responder node at a predetermined time subsequent to said lock status message, an acknowledge confirmation corresponding to said lock status message indicating succesful receipt of said lock response message.

3. An apparatus as recited in claim 2 wherein said confirmation means comprises means for transmitting, over said pended bus to said responder node at a predetermined time sub¬ sequent to said lock status message, a no-acknowledge confirma¬ tion indicating failure to successfully receive said lock sta¬ tus message.

4. An apparatus as recited in claim 1 further com¬ prising means for repeating said interlock read command in re¬ sponse to receipt of a lock status message indicating that said interlock read command was not executed by said responder node.

5. Apparatus for initiating exclusive read-modify-write operations on a pended bus, said operations having a set of distinct transactions on said pended bus including an interlock read command to retrieve information stored at a specified location and restrict access to said stored information by subsequent interlock read commands, and an unlock write command to store information in said specified location and restore access to said stored information, said apparatus comprising: means for initiating, over said pended bus to a re¬ sponder node containing said specified location, an interlock read command to retrieve informa¬ tion from said specified location and an unlock write command to write modified information in said storage location; and means for receiving from said responder node, at an unspecified time subsequent to initiation of said interlock read command, a lock status mes¬ sage indicating whether said interlock read command was executed by .said responder node.