Multiprocessor system with write generate method for updating cache

• US 5,524,212 A
• Filed: 04/27/1992
• Issued: 06/04/1996
• Est. Priority Date: 04/27/1992
• Status: Expired due to Fees

• Current Assignee: Board of Regents of The University of Washington (University of Washington)
• Original Assignee: University of Washington
• Inventors: Johnson, Robert E., Cooper, Kenneth H., Wittenbrink, Craig M., Chen, Chung-Ho, Somani, Arun K., Haralick, Robert M.
• Primary Examiner(s): Black, Thomas G.
• Assistant Examiner(s): Choules, Jack M.
• Application Number:
  US07/876,775
• Time in Patent Office: 1,499 Days
• Field of Search: 395/425, 395/800, 395/468, 395/471, 395/448, 395/200.08, 364/228.7, 364/228.9, 364/241.9, 364/243.41, 364/243.5, 364/940.67, 364/964.31, 364/964.27
• US Class Current: 711/121
• CPC Class Codes: G06F 12/084 with a shared cache

Abstract

A plurality of program processors, shared memory, dual port memory, external cache memory and a control processor form a multiprocessor system. A shared memory bus links the program processors, shared memory, dual port memory and external cache memory. Program processor I/O occurs through a pair of serial I/O channels coupled to one port of the dual port memory. A write generate mode is implemented for updating cache by first allocating lines of shared memory as write before read areas. For such lines, cache tags are updated directly on cache misses without reading from memory. A hit is forced for such line, resulting in valid data at the updated part and invalid data at the remaining portion. Thus, part of the line is written to and the rest invalidated. The invalid portions are not read, unless preceded by a write operation. The mode reduces the number of bus cycles by making write misses more efficient.

Claims

1. A method for updating cache memory in a multiprocessor system having shared memory, comprising the steps of:

allocating a first area of shared memory as being for data that is written before being read, and indicating at one of line or page units of shared memory that a given unit is write before read;

implementing a write operation of data through cache memory for a shared memory address within the write before read area by (1) updating a tag for a first line of the cache memory corresponding to said shared memory address without reading from shared memory, (2) writing said data to a corresponding portion of said first line irrespective of whether said data is a byte, word, or double word and irrespective of whether said data is word-aligned in the first line, (3) setting a third status for said first line denoting valid data is present; and

(4) setting a second status for said first line denoting that the first line has been write generated; and

checking said second status during a later write operation to said first line to determine whether to treat the later write operation as a cache miss.

2. The method of claim 1, in which said first cache memory line comprises the second status, the third status, a tag and a data portion, the data portion for storing data, the tag for linking data stored in the data portion to shared memory, the third status for indicating whether valid data is stored in the data portion;

and

in which a write operation for first data having an address within the first area and not present in the cache memory comprises;

(a) loading a first tag for the first data into a first cache line without reading corresponding data from shared memory,

(b) writing the first data into the data portion of said first cache line irrespective of write alignment,

(c) setting the third status to valid, and

(d) setting the second status to dirty; and

in which a write operation for second data having an address encompassed by said first tag of said first cache line, a corresponding third status set to valid and a corresponding second status set to dirty is treated as a cache miss.

3. The method of claim 1, in which the second status comprises two bits and wherein two writes are allowed to a write generated line in cache.

4. The method of claim 1, in which page units of shared memory are defined as write before read areas;

and in which the second status is located in a page table of a translation look-aside buffer, the second status comprising a fourth status bit per line of a write generated page, each fourth status bit for denoting whether a corresponding line in the write generated page has been write generated.

5. A multiprocessor system having a single-board processor cluster, the cluster comprising:

a plurality of program processors;

shared memory accessible by each one of said plurality of program processors via a shared memory bus;

dual port memory coupled to the shared memory bus and having a first port and a second port;

a pair of serial I/O channels accessible by each one of said plurality of program processors via said dual port memory, a first I/O channel of said pair for receiving first data from outside the cluster and a second I/O channel of said pair for outputting second data received from within the cluster;

wherein the first port of dual port memory is coupled to the pair of I/O channels and the second port is coupled to the shared memory bus;

a control processor for (1) scheduling processing tasks among program processors, (2) managing time-shared I/O operations to and from respective program processors across the pair of serial I/O channels, and (3) managing communication between program processors; and

external cache memory coupled to the shared memory bus, in which a first area of shared memory is allocated as being for data that is written before being read, the first area allocated in one of line or page units of shared memory, a first status associated with each unit indicating whether unit is write before read;

the external cache memory having a cache write generate mode during which a write generate operation specifying an address in the first area is validated on a cache-write miss without reading data from shared memory, the write generate operation comprising;

(1) updating a tag for a first line of the cache memory corresponding to said first area address without reading from shared memory, (2) writing said data to a corresponding portion of said first line irrespective of whether said data is a byte, word, or double word and irrespective of whether said data is word-aligned in the cache memory line, (3) setting a third status for said first line denoting valid data is present; and

(4) setting a second status for said first line denoting that the first line includes dirty data; and

in which said second status is checked during a later write operation to said first line to determine whether to treat the later write operation as a cache miss.

6. The system of claim 5, in which said first cache memory line comprises the second status, the third status, a tag and a data portion, the data portion for storing data, the tag for linking data stored in the data portion to shared memory, the third status for indicating whether valid data is stored in the data portion;

and

in which a write operation for first data having an address within the first area and not present in the cache memory comprises;

(a) loading a first tag for the first data into a first cache line without reading corresponding data from shared memory,

(b) writing the first data into the data portion of said first cache line irrespective of write alignment,

(c) setting the third status to valid, and

(d) setting the second status to dirty; and

in which a write operation for second data having an address encompassed by said first tag of said first cache line, a corresponding third status set to valid and a corresponding second status set to dirty is treated as a cache miss.