Mechanism for reducing latency of memory barrier operations on a multiprocessor system

• US 6,088,771 A
• Filed: 10/24/1997
• Issued: 07/11/2000
• Est. Priority Date: 10/24/1997
• Status: Expired due to Term

• Current Assignee: Hewlett Packard Enterprise Development LP (Hewlett-Packard Enterprise Company)
• Original Assignee: Digital Equipment Corporation (HP Inc.)
• Inventors: Steely, Simon C. Jr., Van Doren, Stephen R., Gharachorloo, Kourosh, Sharma, Madhumitra
• Primary Examiner(s): Chan, Eddie P.
• Assistant Examiner(s): Kim, Hong C.
• Application Number:
  US08/957,501
• Time in Patent Office: 991 Days
• Field of Search: 711/147, 711/150, 711/151, 711/152, 711/154, 711/163, 711/156, 711/158, 711/168, 709/213, 709/248, 709/201, 712/225, 712/16
• US Class Current: 711/154
• CPC Class Codes: G06F 8/458 Synchronisation, e.g. post-wait, barriers, locks synchronisation among tasks G06F9/52
• CPC Class Codes: G06F 9/3004 to perform operations on memory
• CPC Class Codes: G06F 9/30087 Synchronisation or serialisation instructions
• CPC Class Codes: G06F 9/3834 Maintaining memory consistency

Abstract

A technique reduces the latency of a memory barrier (MB) operation used to impose an inter-reference order between sets of memory reference operations issued by a processor to a multiprocessor system having a shared memory. The technique comprises issuing the MB operation immediately after issuing a first set of memory reference operations (i.e., the pre-MB operations) without waiting for responses to those pre-MB operations. Issuance of the MB operation to the system results in serialization of that operation and generation of a MB Acknowledgment (MB-Ack) command. The MB-Ack is loaded into a probe queue of the issuing processor and, according to the invention, functions to pull-in all previously ordered invalidate and probe commands in that queue. By ensuring that the probes and invalidates are ordered before the MB-Ack is received at the issuing processor, the inventive technique provides the appearance that all pre-MB references have completed.

Claims

1. A method for reducing the latency of a memory barrier (MB) operation used to impose an inter-reference order between sets of memory reference operations issued by a first processor to a multiprocessor system having a plurality of processors and a shared memory interconnected by a system control logic, the method comprising:

issuing a first set of memory reference operations from the first processor to the system control logic;

issuing the MB operation from the first processor to the system control logic immediately after issuing the first set of memory reference operations without waiting for responses to the first set of memory reference operations to arrive at the first processor;

ordering the first set of memory reference operations with respect to other memory reference operations issued by other processors of the system at an ordering point of a switch;

generating probe and invalidate packets for the ordered first set of memory reference operations at the ordering point;

loading the probe and invalidate packets into probe queues of the first and other processors for transmission to those processors;

ordering the MB operation at the ordering point after ordering of the first set of memory reference operations;

generating a MB acknowledgment (MB-Ack) in response to the ordered MB operation; and

loading the MB-Ack into the probe queue of the first processor for transmission to the first processor, the loaded MB-Ack pulling-in all previously ordered invalidate and probe commands in the probe queue of the first processor.

2. The method of claim 1 wherein the step of issuing the first set of memory reference operations comprises the step of loading the first set of memory reference operations into a first-in, first-out request queue of the system control logic.

3. The method of claim 2 wherein the step of ordering the first set of memory reference operations with respect to other memory reference operations issued by other processors of the system comprises the step of serializing each operation selected by an arbiter of the system control logic in accordance with an arbitration policy.

4. The method of claim 1 wherein the shared memory comprises a plurality of banks and wherein the step of issuing the first set of memory reference operations comprises the step of loading the first set of memory reference operations into a plurality of first-in, first-out request queues of the system control logic, each request queue associated with a respective processor and a respective shared memory bank.

5. The method of claim 4 wherein the step of ordering the MB operation at the ordering point after ordering of the first set of memory reference operations comprises the step of serializing the MB operation selected by an arbiter of the system control logic in accordance with an arbitration policy.

6. The method of claim 5 wherein the arbitration policy considers activity per shared memory bank.

7. The method of claim 5 wherein the arbitration policy prevents the MB operation from gaining access to the ordering point until all memory reference operations pending in the request queues associated with the first processor have accessed the ordering point.