System and method for high performance execution of locked memory instructions in a system with distributed memory and a restrictive memory model

• US 20020199067A1
• Filed: 08/30/2002
• Published: 12/26/2002
• Est. Priority Date: 12/29/2000
• Status: Active Grant

• Current Assignee: Intel Corporation
• Original Assignee: Intel Corporation
• Inventors: Boatright, Bryan D., Thatcher, Larry Edward, Patel, Rajesh Bhikhubhai
• Granted Patent: US 6,611,900 B2
• Time in Patent Office: Days
• US Class Current: 711/145
• CPC Class Codes: G06F 12/0831 using a bus scheme, e.g. with bus monitoring or watching means
• CPC Class Codes: G06F 9/3004 to perform operations on memory
• CPC Class Codes: G06F 9/30087 Synchronisation or serialisation instructions

Abstract

The present invention relates to locked memory instructions, and more specifically to a system and method for the high performance execution of locked memory instructions in a system with distributed memory and a restrictive memory model. In accordance with an embodiment of the present invention, a method for executing locked-memory instructions includes decoding a locked-memory instruction, obtaining exclusive ownership of a cacheline to be used by a load-lock operation, setting a bit to indicate the load-lock operation'"'"'s ownership of the cacheline, and activating a snoop checking process. The method also includes modifying a load data value and storing the modified load data value. The method further includes determining that the cacheline is still exclusively owned, storing the load data value, determining that the cacheline is unsnooped, merging the modified load data value with the load data value, and releasing the locked-memory instruction to be retired.

Claims

1. A method for executing a locked-memory instruction, the method comprising:

obtaining exclusive ownership of a cacheline for a load-lock operation;

setting a bit to indicate the load-lock operation'"'"'s ownership of the cacheline;

storing a modified version of a load data value;

determining that the cacheline is still exclusively owned;

storing the load data value;

determining that the cacheline is unsnooped;

merging the modified version of the load data value with the stored load data value; and

releasing the locked-memory instruction to be retired.

2. The method of claim 1 further comprising:

decoding the locked-memory instruction; and

breaking the locked-memory instruction into the load-lock operation, a modify operation; and

a store-unlock operation.

3. The method of claim 1 further comprising

allocating a write combining buffer entry;

and

setting a lock register by storing an address in the lock register of a store-unlock operation that reserved the write combining buffer entry.

4. The method of claim 3 further comprising:

delaying a snoop checking process during a critical sequence in the execution of the locked-memory instruction.

5. The method of claim 4 wherein said delaying the snoop checking process during a critical sequence in the execution of the locked-memory instruction comprises:

delaying the snoop checking process while

allocating the write combining buffer entry in an exclusive state,

checking to determine whether a snoop of the executing instruction occurred, and

updating the write combining buffer entry with the stored load data value.

6. The method of claim 1 further comprising:

requesting the load data value be returned in an exclusive state, if the cacheline is in a non-exclusive state; and

determining that the cacheline is still exclusively owned.

7. The method of claim 1 further comprising:

flushing the locked-memory instruction and all younger dependent instructions, if the cacheline has been snooped; and

restarting the method by re-executing said obtaining exclusive ownership of the cacheline for the load-lock operation.

8. A machine-readable medium having stored thereon a locked-memory instruction, said locked memory instruction executable to perform a method comprising:

obtaining exclusive ownership of a cacheline for a load-lock operation;

setting a bit to indicate the load-lock operation'"'"'s ownership of the cacheline;

storing a modified version of a load data value;

determining that the cacheline is still exclusively owned;

storing the load data value;

determining that the cacheline is unsnooped;

merging the modified version of the load data value with the stored load data value; and

releasing the locked-memory instruction to be retired.

9. The machine-readable medium of claim 8, said method further comprising:

decoding the locked-memory instruction; and

breaking the locked-memory instruction into the load-lock operation, a modify operation, and a store-unlock operation.

10. The machine-readable medium of claim 8, said method further comprising:

allocating a write combining buffer entry; and

setting a lock register by storing an address in the lock register of a store-unlock operation that reserved the write combining buffer entry.

11. The machine-readable medium of claim 10, said method further comprising:

delaying a snoop checking process during a critical sequence in the execution of the locked-memory instruction.

13. The machine-readable medium of claim 8, said method further comprising:

requesting the load data value be returned in an exclusive state, if the cacheline is in a non-exclusive state; and

determining that the cacheline is still exclusively owned.

14. The machine-readable medium of claim 8, said method further comprising:

flushing the locked-memory instruction and all younger dependent instructions, if the cacheline has been snooped; and

restarting the method by re-executing said obtaining exclusive ownership of the cacheline for the load-lock operation.

16. The processor of claim 15 further comprising:

a second cache coupled to the first cache.

17. The processor of claim 15 wherein said processor is implemented as a plurality of processors in a multi-processor system such that each of said plurality of processors includes a separate set of hardware resources.

19. The processor of claim 18 further comprising:

a second cache coupled to the first cache to store an original version of the cacheline.

20. The processor of claim 18 wherein said processor is implemented as a plurality of processors in a multi-processor system such that each of said plurality of processors includes a separate set of hardware resources.

12. The machine-readable medium of claim 111 wherein said delaying the snoop checking process during a critical sequence in the execution of the locked-memory instruction comprises:

delaying the snoop checking process while

allocating the write combining buffer entry in an exclusive state,

checking to determine whether a snoop of the executing instruction occurred, and

updating the write combining buffer entry with the stored load data value.

15. A processor comprising:

a first cache;

a load buffer coupled to the first cache;

a control logic component coupled to the load buffer;

a store buffer;

a pause queue coupled to the store buffer;

a write combining buffer coupled to the pause queue and the first cache; and

a lock register coupled to the pause queue and the load buffer.

18. A processor comprising:

a first cache to store a local copy of a cacheline and to maintain exclusivity information associated with the local copy of the cacheline for use by a locked-memory instruction;

a load buffer coupled to the first cache, the load buffer to receive and queue a load-lock operation and an associated load data value;

a control logic component coupled to the load buffer, the control logic component to control retirement of the locked-memory instruction;

a store buffer to receive and queue a store-unlock operation and the load data value;

a pause queue coupled to the store buffer, the pause queue to store a modified version of the load data value and execute a store-unlock operation received from the store buffer;

a lock register coupled to the pause queue and the load buffer, the lock register to receive and store an address of the store-unlock operation; and

a write combining buffer coupled to the pause queue and the first cache, the write combining buffer to receive the modified version of the load data value from the pause queue and merge the modified version of the load data value with the load data value.