A multiprocessor computer system which maintains cache coherency includes first and second microprocessors each having an associated cache memory storing lines of data. Each line of data has associated protocol bits that indicate a protocol state consistent with write-through, write-back, or write-once cache coherency policies that are selected via a protocol selection terminal for different system configurations. In one configuration, the output and external address terminals of the first microprocessor are coupled to the external and output address terminals, respectively, of the second microprocessor. This configuration enables each microprocessor to snoop memory cycles to main memory initiated by the other microprocessor so that it can be readily determined if a particular cache has the latest version of data.
This is a continuation of application Ser. No. 08/195,827, filed Feb. 10, 1994, now abandoned, which is a continuation of application Ser. No. 07/777,763, filed Oct. 11, 1991, now abandoned.
A processor is disclosed that can map a request from a central processing unit that uses memory-mapped input-output space to a second processing domain, such as a multithreaded processing domain. A request addressed to the input-output space of the central processing unit is converted to a corresponding command that simulates an operation between components in the multithreaded processing domain. The command is executed in the multithreaded processing domain. Information is accessed according to the request in response to executing the command.
A cache coherency directory for a shared memory multiprocessor computer system. A data structure is associated with each cacheable memory location, the data structure comprising locations for storing state values indicating an exclusive state, a shared state, an uncached state, a busy state, a busy uncached state, a locked state, and a pending state. The busy state and pending state cooperate to reserve a cache line for future use by a processor while the cache line is currently being used by one or more other processors.
A memory management system couples processors to each other and to a main memory. Each processor may have one or more associated caches local to that processor. A system port of the memory management system receives a request from a source processor of the processors to access a block of data from the main memory. A memory manager of the memory management system then converts the request into a probe command having a data movement part identifying a condition for movement of the block out of a cache of a target processor and a next coherence state part indicating a next state of the block in the cache of the target processor.
The present invention discloses a method and apparatus that uses extensions to the TLB entry to dynamically identify pages of memory that can be weakly ordered or must be strongly ordered and enforces the appropriate memory model on those pages of memory. Such identification and memory model enforcement allows for more efficient execution of memory instructions in a hierarchical memory design in cases where memory instructions can be executed out of order. From the page table, the memory manager constructs TLB entries that associate page frame numbers of memory operands with page-granular client usage data and a memory order tag. The memory order tag identifies the memory model that is currently being enforced for the associated page of memory. The memory manager updates the memory order tag of the TLB entry in accordance with changes in the client usage information. In the preferred embodiment, the TLB structure is a global TLB shared by all processors. In alternative embodiments, the TLB structure may comprise either multiple distributed TLBs with shared knowledge, each assigned to a different processor, or a combination of multiple local TLBs, each assigned to a different processor, that exchange information with a global TLB, which in turn provides data to the memory manager to access the hierarchical memory system.
A cache coherency protocol uses a "Tagged" coherency state to track responsibility for writing a modified value back to system memory, allowing intervention of the value without immediately writing it back to system memory, thus increasing memory bandwidth. The Tagged state can migrate across the caches (horizontally) when assigned to a cache line that has most recently loaded the modified value. Historical states relating to the Tagged state may further be used. The invention may also be applied to a multi-processor computer system having clustered processing units, such that the Tagged state can be applied to one of the cache lines in each group of caches that support separate processing unit clusters. Priorities are assigned to different cache states, including the Tagged state, for responding to a request to access a corresponding memory block. Any tagged intervention response can be forwarded only to selected caches that could be affected by the intervention response, using cross-bars. The Tagged protocol can be combined with existing and new cache coherency protocols. The invention further contemplates independent optimization of cache operations using the Tagged state.
