Snoop filter for filtering snoop requests

• US 7,373,462 B2
• Filed: 03/29/2005
• Issued: 05/13/2008
• Est. Priority Date: 03/29/2005
• Status: Expired due to Fees

• Current Assignee: Intel Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Salapura, Valentina, Vranas, Pavlos M., Blumrich, Matthias A., Chen, Dong, Gara, Alan G., Giampapa, Mark E., Heidelberger, Philip, Hoenicke, Dirk I., Ohmacht, Martin
• Primary Examiner(s): Thai, Tuan V.
• Application Number:
  US11/093,152
• Publication Number:
  US 20060224838A1
• Time in Patent Office: 1,141 Days
• Field of Search: 711/100, 711/117, 711/118, 711/131, 711/144, 711/146, 711/154
• US Class Current: 711/146
• CPC Class Codes: G06F 12/0822 Copy directories local copy tags for implementing a bus snooping protocol G06F12/0831
• CPC Class Codes: G06F 12/0831 using a bus scheme, e.g. with bus monitoring or watching means
• CPC Class Codes: G06F 2212/507 using speculative control
• CPC Class Codes: Y02D 10/00 Energy efficient computing, e.g. low power processors, power management or thermal management

Abstract

A method and apparatus for supporting cache coherency in a multiprocessor computing environment having multiple processing units, each processing unit having one or more local cache memories associated and operatively connected therewith. The method comprises providing a snoop filter device associated with each processing unit, each snoop filter device having a plurality of dedicated input ports for receiving snoop requests from dedicated memory writing sources in the multiprocessor computing environment. Each snoop filter device includes a plurality of parallel operating port snoop filters in correspondence with the plurality of dedicated input ports, each port snoop filter implementing one or more parallel operating sub-filter elements that are adapted to concurrently filter snoop requests received from respective dedicated memory writing sources and forward a subset of those requests to its associated processing unit.

Claims

1. A snoop filtering method for supporting cache coherency in a computing environment having multiple processing units, each processing unit implementing a corresponding snoop filter apparatus, said method comprising:

receiving, at a plurality of inputs, a snoop request from a plurality of memory writing sources in said computing environment, each input corresponding to a memory writing source of said plurality of memory writing sources;

providing at each of said plurality of inputs a snoop filter means for concurrently filtering received respective snoop requests from a corresponding memory writing source; and

,

each snoop filter means implementing one or more parallel operating sub-filter elements for simultaneously processing received snoop requests and forwarding a subset thereof to said associated processing unit,

whereby as a result of said concurrent filtering, a number of snoop requests forwarded to a processing unit is reduced thereby increasing performance of said computing environment.

2. The snoop filtering method as claimed in claim 1, wherein each said one or more parallel operating sub-filter elements generates a signal indicating whether a snoop request is to be forwarded to said associated processing unit or not forwarded, said method further comprising:

responding to each signal generated from said sub-filter element for deciding whether a snoop request is to be forwarded or discarded.

3. The snoop filtering method as claimed in claim 2, wherein one of said one or more parallel operating sub-filter elements implements steps of:

determining whether an address of a received snoop request is within an address range comprising a minimum range address and a maximum range address, and one of;

discarding the snoop request if it is outside said address range;

or forwarding the snoop snoop request if it is within said address range.

4. The snoop filtering method as claimed in claim 2, wherein one of said one or more parallel operating sub-filter elements implements steps of:

determining whether an address of a received snoop request is within an address range comprising a minimum range address and a maximum range address, and one of;

forwarding the snoop request if it is iutside said address range;

or discarding the snoop request if it is with said address range.

5. The snoop filtering method as claimed in claim 2, wherein one of said one or more parallel operating sub-filter elements in a snoop filter means implements steps of:

tracking snoop requests received at said snoop filter means and recording an addresses corresponding to snoop requests received;

comparing a received snoop request address against all addresses recorded at said snoop filter means, and, one of forwarding said received snoop request to said associated processing unit when an address does not match a recorded address, or discarding said snoop request when an address matches a recorded address.

6. The snoop filtering method as claimed in claim 5, wherein said snoop requests addresses are recorded in a snoop cache device comprising a plurality of cache line entries, each entry having an address tag field associated with the snooped cache line address, and a valid line vector for encoding snoop requests for a sequence of consecutive cache lines, said comparing step comprise sing the steps of comparing an address tag field of a received snoop request against all tag fields in the snoop cache, and, comparing a valid line vector of the snoop request against the valid line vector of the snoop cache line.

7. The snoop filtering method as claimed in claim 6, further comprising the step of:

adding a new snoop cache line entry to said snoop cache when the snoop cache line request address tag does not match the address tags stored in the snoop cache.

8. The snoop filtering method as claimed in claim 6, further comprising the step of:

setting a bit location in the valid line vector in the selected snoop cache line corresponding to a bit location of said valid line vector of the snoop request when the snoop request address tag matches an address tag stored in the snoop cache and the valid line vector of that snoop request does not match a valid line vector that snoop cache line.

9. The snoop filtering method as claimed in claim 6, further comprising the step of:

adding a snoop request address from said dedicated memory writer into the snoop cache when said means responsive to each signal generated from said sub-filter element for deciding whether a snoop request is to be forwarded or discarded indicates that said request is to be forwarded.

10. The snoop filtering method as claimed in claim 2, wherein each said processing unit has one or more cache memories associated therewith, said method further comprising the step of tracking cache line addresses of data that have been loaded into a cache memory level of its associated processor and recording said cache line addresses in a memory storage device.

11. The snoop filtering method as claimed in claim 10, wherein said memory storage means comprises a plurality of stream register sets, each stream register set comprising one or more stream registers, each stream register including a base register and a corresponding mask register, wherein said base register tracks address bits common to all cache lines represented by the stream register;

and, said corresponding mask register tracks bits representing differences between a base register address and subsequent load/store addresses to the cache memories of said processing units.

12. The snoop filtering method as claimed in claim 11, further comprising the steps of:

comparing an address of the received snoop request against addresses stored in said memory storage means; and

, one of

forwarding said received snoop request to said processor in response to matching an address in said memory storage means, or otherwise discarding said snoop request.

13. The snoop filtering method as claimed in claim 12, wherein one of said one or more parallel operating sub-filter elements in a snoop filter means implements steps of:

converting a received snoop address into a line address by removing lower-order bits corresponding to an offset within a cache line; and

,

comparing all base register contents against the snoop request address combined with the contents of the corresponding mask register to obtain a result signal indicating possibility of data located in said cache memory at that particular cache line address.

14. The snoop filtering method as claimed in claim 13, further comprising the step of:

tracking whether every cache line has been replaced in said cache memory level since a prior cache wrap detection state.

15. The snoop filtering method as claimed in claim 1, wherein a snoop request is received directly from a dedicated memory writing source, said dedicated memory writing source comprising one of said multiple processing units.

16. The snoop filtering method as claimed in claim 15, wherein each said processing unit has one or more cache memories associated therewith, said method further comprising the steps of:

checking a cache line address of each process memory load request that misses in a cache memory against all entries in all snoop caches associated with all snoop request sources and, if said address comparison results in one or more matches,

removing the cache line address from all snoop caches for which it matched.

17. The snoop filtering method as claimed in claim 1, wherein a snoop request is received directly from a dedicated memory writing source, said dedicated memory writing source comprising a Direct Memory Access (DMA) engine.

18. The snoop filtering method as claimed in claim 1, further comprising steps of:

providing a queue means associated with each snoop filter means, each said queue means adapted for;

receiving said subset of snoop requests;

enqueing said subset of snoop requests; and

forwarding said subset of snoop requests to the associated processing unit.

19. The snoop filtering method as claimed in claim 18, further comprising the step of:

arbitrating between all of the snoop filter queue means and serializing all snoop requests forwarded from each of said plurality of processor snoop filter queue means for forwarding to said associated processing unit.

20. The snoop filtering method as claimed in claim 1, further comprising the step of:

disabling operation of one or more parallel operating sub-filter elements in a respective snoop filter means.

21. A computer program embodied on a computer-readable medium for supporting cache coherency in a computing environment having multiple processing units, implementing the method steps according to claim 1.