Granted Patent B2

US 7,558,939 · App. 11/075,041 · Granted Jul 7, 2009

Abstract

Claims (147)

1. A hardware three-tiered translation lookaside buffer (TLB) apparatus in a multithreading microprocessor that concurrently fetches and executes instructions of a plurality of threads, for providing virtual-to-physical address translation information for a memory page in response to an instruction cache virtual fetch address for a selected one of the plurality of threads, the apparatus comprising:

a third-level TLB, for caching virtual-to-physical address translation information for a plurality of memory pages for the plurality of threads;

a second-level TLB, coupled to said third-level TLB, having a plurality of entries for caching said virtual-to-physical address translation information for a subset of said plurality of memory pages cached in said third-level TLB; and

a first-level TLB for each of a respective one of the plurality of threads, each said first-level TLB having at least one entry for caching virtual-to-physical address translation information only for said respective one of the plurality of threads.

2. The apparatus as recited in claim 1 , wherein each of said first-level TLBs further comprises:

storage, for storing replacement information indicating which of said first-level TLB at least one entry and said second-level TLB entries are caching most recently used virtual-to-physical address translation information for said respective one of the plurality of threads.

3. The apparatus as recited in claim 2 , wherein said replacement information comprises:

a single pointer, indicating which one, if any, of said second-level TLB entries is caching most recently used virtual-to-physical address translation information for said respective one of the plurality of threads.

4. The apparatus as recited in claim 2 , wherein said replacement information comprises:

a single pointer, indicating which one of said first-level TLB at least one entry and said second-level TLB entries is caching most recently used virtual-to-physical address translation information for said respective one of the plurality of threads.

5. The apparatus as recited in claim 2 , wherein said replacement information comprises:

a plurality of pointers, indicating which of said first-level TLB at least one entry and said second-level TLB entries are caching most recently used virtual-to-physical address translation information for said respective one of the plurality of threads.

6. The apparatus as recited in claim 2 , wherein said replacement information comprises pseudo-most-recently-used information for exactly indicating which one of said first-level TLB at least one entry and said second-level TLB entries is caching most recently used virtual-to-physical address translation information for said respective one of the plurality of threads and for approximately indicating which one or more of said first-level TLB at least one entry and said second-level TLB entries are caching next most recently used virtual-to-physical address translation information for said respective one of the plurality of threads.

7. The apparatus as recited in claim 2 , further comprising:

control logic, coupled to said first-, second-, and third-level TLBs, wherein if the virtual fetch address misses in both said second-level TLB and said respective first-level TLB of said one of the plurality of threads associated with the virtual fetch address, said control logic is configured to evict said virtual-to-physical address translation information from one of said second-level TLB entries, and to refill said one of said second-level TLB entries with virtual-to-physical address translation information from said third-level TLB.

8. The apparatus as recited in claim 7 , wherein if the virtual fetch address misses in both said respective first-level TLB and said second-level TLB, said control logic is configured to copy said evicted information to one of said at least one entry of each of said first-level TLBs whose said replacement information indicates said evicted one of said second-level TLB entries is caching most recently used virtual-to-physical address translation information for said respective one of the plurality of threads.

9. The apparatus as recited in claim 8 , wherein for each of said first-level TLBs to which said control logic copies said evicted information, said control logic is further configured to update said replacement information of said first-level TLB to indicate said evicted information is now being cached in said first-level TLB rather than said second-level TLB.

10. The apparatus as recited in claim 8 , wherein said control logic is configured to exclude said respective first-level TLB of the selected one of the plurality of threads when copying said evicted information.

11. The apparatus as recited in claim 10 , wherein each of said first-level TLBs has a single entry.

12. The apparatus as recited in claim 7 , wherein if the virtual fetch address hits an entry in said second-level TLB, said control logic is configured to update said replacement information to indicate said hit second-level TLB entry is most recently used.

13. The apparatus as recited in claim 7 , wherein if the virtual fetch address hits an entry in said first-level TLB, said control logic is configured to update said replacement information to indicate said hit first-level TLB entry is most recently used.

14. The apparatus as recited in claim 7 , wherein if the virtual fetch address hits in said respective first-level TLB, said control logic is configured to update said replacement information to indicate none of said second-level TLB entries.

15. The apparatus as recited in claim 14 , wherein if the virtual fetch address hits in said respective first-level TLB, said control logic is configured to update said replacement information by clearing said replacement information to indicate none of said second-level TLB entries.

16. The apparatus as recited in claim 15 , wherein said plurality of second-level TLB entries equals a positive integer N, wherein each of said replacement information comprises N bits which are set to indicate respective ones of said N plurality of second-level TLB entries, wherein said control logic is configured to clear all said N bits of said replacement information to indicate none of said second-level TLB entries.

17. The apparatus as recited in claim 7 , wherein if the virtual fetch address misses in said second-level TLB, said respective first-level TLB, and said third-level TLB, said control logic is configured to generate an exception to request system software to refill said one of said second-level TLB plurality of entries with virtual-to-physical address translation information.

18. The apparatus as recited in claim 7 , wherein if the virtual fetch address misses in said second-level TLB, said respective first-level TLB, and said third-level TLB, said control logic is configured to refill said one of said second-level TLB plurality of entries with virtual-to-physical address translation information from a page table walk.

19. The apparatus of claim 2 , wherein each of said first-level TLB at least one entry comprises a plurality of entries.

20. The apparatus as recited in claim 19 , wherein said replacement information comprises:

a single pointer, indicating which one of said first-level TLB at least one entry and said second-level TLB entries is caching most recently used virtual-to-physical address translation information for said respective one of the plurality of threads.

21. The apparatus of claim 20 , further comprising:

control logic, coupled to said first-level TLBs, wherein if said single pointer indicates one of said second-level TLB plurality of entries is caching said most recently used virtual-to-physical address translation information for said respective one of the plurality of threads, and if said most recently used translation information is being evicted from said second-level TLB, said control logic is configured to copy said evicted information to one of said plurality of entries of each of said first-level TLBs whose said replacement information indicates said evicted one of said second-level TLB entries is caching said most recently used virtual-to-physical address translation information.

22. The apparatus of claim 21 , wherein said control logic is configured to select the oldest one of said plurality of entries of each of said first-level TLBs for copying said eviction instruction thereto.

23. The apparatus of claim 21 , wherein said control logic is configured to select for copying to said one of said plurality of entries of each of said first-level TLBs that is next in round-robin order.

24. The apparatus of claim 21 , wherein said control logic is configured to select for copying to said one of said plurality of entries of each of said first-level TLBs that is least recently used among said first-level TLB plurality of entries.

25. The apparatus of claim 21 , wherein said control logic is configured to select for copying to said one of said plurality of entries of each of said first-level TLBs randomly.

26. The apparatus as recited in claim 1 , further comprising:

selection logic, coupled to said first-level TLBs for selecting said respective first-level TLB of the selected one of the plurality of threads associated with the virtual fetch address.

27. The apparatus as recited in claim 26 , wherein said selection logic is configured to select said respective first-level TLB in a first clock cycle prior to a second clock cycle in which the virtual fetch address is looked up in said second-level TLB.

28. The apparatus as recited in claim 26 , further comprising:

second selection logic, coupled to said first-level TLBs and said second-level TLB, for selecting one of said second-level TLB entries or said at least one entry of said respective first-level TLB in which the virtual fetch address hits.

29. The apparatus as recited in claim 28 , wherein said selection logic comprises a 4:1 mux.

30. The apparatus as recited in claim 1 , wherein each of said first-level TLBs is configured to provide said information cached in said at least one entry in a single cycle of said clock.

31. The apparatus as recited in claim 1 , wherein each of said first-level TLBs is configured to provide said information cached in said at least one entry in a fraction of a cycle of said clock.

32. The apparatus as recited in claim 1 , wherein each of said first-level TLBs has one entry.

33. The apparatus as recited in claim 1 , wherein said second-level TLB plurality of entries is 3.

34. The apparatus as recited in claim 1 , wherein said third-level TLB includes at least 4 times as many entries as said second-level TLB plurality of entries.

35. The apparatus as recited in claim 1 , wherein said third-level TLB includes 64 entries.

36. The apparatus of claim 1 , wherein said second-level TLB provides said information in a first time and said third-level TLB provides said information in a second time, wherein said second time is longer than said first time.

37. The apparatus of claim 1 , wherein said second-level TLB provides said information in a first time and said third-level TLB provides said information in a second time, wherein said second time is at least twice said first time.

38. The apparatus of claim 36 , wherein said first time comprises a fraction of a clock cycle of the microprocessor, wherein said second time comprises a single clock cycle of the microprocessor.

39. The apparatus of claim 36 , wherein said first time comprises a single clock cycle of the microprocessor, wherein said second time comprises a plurality of clock cycles of the microprocessor.

40. A method for translating an instruction cache virtual fetch page address to a physical address in a multithreading microprocessor that concurrently fetches and executes instructions of a plurality of threads, the method comprising:

caching virtual-to-physical address translation information in a three-tiered translation lookaside buffer (TLB) system, the system including a macro-TLB that caches virtual-to-physical address translation information for all of the plurality of threads, a micro-TLB that caches a subset of the translation information of the macro-TLB for all of the plurality of threads, and a plurality of nano-TLBs, each nano-TLB associated with a respective one of the plurality of threads, each nano-TLB for caching translation information only for the associated thread;

selecting one of the plurality of threads for providing a virtual fetch address;

selecting one of the plurality of nano-TLBs associated with the one of the plurality of threads selected to provide the virtual fetch address;

looking up the virtual address in the macro-TLB, micro-TLB, and selected nano-TLB; and

providing to the instruction cache a physical address translated from the virtual address provided by one of the macro-, micro-, or selected nano-TLB in which the virtual address hits.

41. The method as recited in claim 40 , wherein the micro-TLB and nano-TLBs are accessible in a shorter time than the macro-TLB.

42. The method as recited in claim 40 , wherein said looking up the virtual address comprises:

looking up the virtual address in the micro-TLB and in the selected nano-TLB; and

looking up the virtual address in the macro-TLB, if the virtual address misses in the micro-TLB and in the selected nano-TLB.

43. The method as recited in claim 40 , further comprising:

generating an exception, if the virtual address misses in the macro-TLB, the micro-TLB, and the selected nano-TLB.

44. The method as recited in claim 40 , further comprising:

maintaining replacement information for each of the plurality of nano-TLBs, in response to said looking up the virtual address in the TLB system, wherein the replacement information indicates which one or more entries in the micro-TLB and the plurality of nano-TLBs are caching most recently used translation information for the respective one of the plurality of threads.

45. The method as recited in claim 44 , wherein said maintaining the replacement information comprises:

updating the replacement information to indicate one of the plurality of nano-TLBs entries if the virtual fetch address hits in the one of the plurality of nano-TLBs entries.

46. The method as recited in claim 44 , wherein said maintaining the replacement information comprises:

updating the replacement information to indicate none of the one or more micro-TLB entries if the virtual fetch address hits in the nano-TLB.

47. The method as recited in claim 44 , wherein said maintaining the replacement information comprises:

updating the replacement information to indicate one of the micro-TLB entries if the most recently used translation information is refilled into the one of the micro-TLB entries.

48. The method as recited in claim 44 , wherein said maintaining the replacement information comprises:

updating the replacement information to indicate one of the micro-TLB entries if the virtual address hits in the one of the micro-TLB entries.

49. The method as recited in claim 44 , further comprising:

selecting one of the micro-TLB entries for eviction of the translation information cached therein, if the virtual address misses in the micro-TLB and selected nano-TLB, in response to said looking up; and

for each of the plurality of threads:

determining whether the replacement information indicates the entry in the micro-TLB selected for eviction is caching most recently used translation information for the thread; and

copying the evicted translation information to the selected nano-TLB, if the replacement information indicates the entry in the micro-TLB selected for eviction is caching most recently used translation information for the thread.

50. The method as recited in claim 49 , further comprising for each of the plurality of threads:

updating the replacement information to indicate the nano-TLB rather than the micro-TLB is caching the most recently used translation information for the thread, if said copying is performed.

51. The method as recited in claim 49 , further comprising:

refilling the micro-TLB entry selected for eviction with address translation information from the macro-TLB, after evicting the selected one of the micro-TLB entries.

52. The method as recited in claim 51 , further comprising:

updating the replacement information in the one of the plurality of nano-TLBs associated with the one of the plurality of threads selected to provide the virtual fetch address to indicate the refilled entry in the micro-TLB is caching the most recently used translation information for the thread.

53. The method as recited in claim 40 , wherein each of the nano-TLBs has one entry.

54. The method as recited in claim 40 , wherein each of the nano-TLBs has a plurality of entries.

55. The method as recited in claim 40 , wherein the macro-TLB has 3 entries.

56. A method for translating an instruction cache virtual fetch page address to a physical address in a multithreading microprocessor that concurrently fetches and executes instructions of a plurality of threads, the method comprising:

caching virtual-to-physical address translation information in a three-tiered translation lookaside buffer (TLB) system, the system including a macro-TLB that caches virtual-to-physical address translation information for all of the plurality of threads, a micro-TLB that caches a subset of the translation information of the macro-TLB for all of the plurality of threads, and a plurality of nano-TLBs, each nano-TLB associated with a respective one of the plurality of threads, each nano-TLB for caching translation information only for the associated thread;

selecting one of the plurality of threads for providing a virtual fetch address;

selecting one of the plurality of nano-TLBs associated with the one of the plurality of threads selected to provide the virtual fetch address;

looking up the virtual address in the macro-TLB, micro-TLB, and selected nano-TLB;

providing to the instruction cache a physical address translated from the virtual address provided by one of the macro-, micro-, or selected nano-TLBs in which the virtual address hits;

maintaining replacement information for each of the plurality of nano-TLBs, in response to said looking up the virtual address in the TLB system, wherein the replacement information indicates which one or more entries in the micro-TLB and the nano-TLB are caching most recently used translation information for the respective one of the plurality of threads;

selecting one of the micro-TLB entries for eviction of the translation information cached therein, if the virtual address misses in the micro-TLB and selected nano-TLB, in response to said looking up; and

for each of the plurality of threads:

determining whether the replacement information indicates the entry in the micro-TLB selected for eviction is caching most recently used translation information for the thread;

copying the evicted translation information to the associated nano-TLB, if the replacement information indicates the entry in the micro-TLB selected for eviction is caching most recently used translation information for the thread, wherein said copying is performed for each of the plurality of nano-TLBs other than the one of the plurality of nano-TLBs associated with the one of the plurality of threads selected to provide the virtual fetch address.

57. A multithreading microprocessor for concurrently fetching and executing instructions of a plurality of threads and supporting a virtual memory system, comprising:

an instruction cache, for caching instructions of the plurality of threads;

a virtual fetch address, for each of the plurality of threads, for accessing said instruction cache; and

an instruction fetcher, coupled to said instruction cache, comprising:

a fetch scheduler, configured to select each clock cycle said virtual fetch address of one of the plurality of threads for fetching from said instruction cache; and

a macro-TLB, a micro-TLB, and a plurality of nano-TLBs, each nano-TLB associated with a respective one of the plurality of threads, said macro-TLB configured to cache physical address translations for memory pages for the plurality of threads, said micro-TLB having a plurality of entries configured to cache said translations for a subset of said memory pages cached in said macro-TLB, said plurality of nano-TLBs configured to cache a physical address translation of at least one memory page for said respective one of the plurality of threads.

58. The microprocessor as recited in claim 57 , wherein said instruction fetcher further comprises:

selection logic, coupled to said plurality of nano-TLBs, configured to select one of said plurality of nano-TLBs associated with said respective one of the plurality of threads;

wherein each of said plurality of nano-TLBs further comprises:

replacement information, for indicating which one or more entries of said micro-TLB and said plurality of nano-TLBs are caching most recently used physical address translations for said respective one of the plurality of threads.

59. The microprocessor as recited in claim 58 , wherein said instruction fetcher further comprises:

control logic, coupled to said nano-, micro-, and macro-TLBs, wherein if said virtual fetch address misses in both said micro-TLB and said selected nano-TLB, said control logic is configured to evict said physical address translation from one of said one or more micro-TLB entries, and to refill said one of said micro-TLB entries with a physical address translation from said macro-TLB.

60. The microprocessor as recited in claim 59 , wherein if said virtual fetch address misses in both said selected nano-TLB and said micro-TLB, said control logic is configured to copy said evicted physical address translation to said each of said plurality of nano-TLBs whose said replacement information indicates said evicted one of said micro-TLB plurality of entries.

61. The microprocessor as recited in claim 57 , wherein said micro-TLB and said selected nano-TLB are configured to provide said physical address translation in less time than said macro-TLB.

62. The microprocessor as recited in claim 57 , wherein each of said plurality of nano-TLBs is configured to provide said physical address translation in a single cycle of the microprocessor clock.

63. The microprocessor as recited in claim 57 , wherein each of said plurality of nano-TLBs is configured to provide said physical address translation in a fraction of a cycle of the microprocessor clock.

64. The microprocessor as recited in claim 57 , wherein said micro-TLB is configured to provide said physical address translation in a single cycle of the microprocessor clock.

65. The microprocessor as recited in claim 57 , wherein said micro-TLB is configured to provide said physical address translation in a fraction of a cycle of the microprocessor clock.

66. A multithreading microprocessor for concurrently fetching and executing instructions of a plurality of threads and supporting a virtual memory system, comprising:

an instruction cache, for caching instructions of the plurality of threads;

a virtual fetch address, for each of the plurality of threads, for accessing said instruction cache; and

an instruction fetcher, coupled to said instruction cache, comprising:

a fetch scheduler, configured to select each clock cycle said virtual fetch address of one of the plurality of threads for fetching from said instruction cache;

a macro-TLB, a micro-TLB, and a plurality of nano-TLBs, each nano-TLB associated with a respective one of the plurality of threads, said macro-TLB configured to cache physical address translations for memory pages for the plurality of threads, said micro-TLB having a plurality of entries configured to cache said translations for a subset of said memory pages cached in said macro-TLB, said plurality of nano-TLBs configured to cache a physical address translation of at least one memory page for said respective one of the plurality of threads;

selection logic, coupled to said plurality of nano-TLBs, configured to select one of said plurality of nano-TLBs associated with said respective one of the plurality of threads, wherein each of said plurality of nano-TLBs further comprises replacement information, for indicating which one or more entries of said micro-TLB and said plurality of nano-TLBs are caching most recently used physical address translations for said respective one of the plurality of threads; and

control logic, coupled to said nano-, micro-, and macro-TLBs, wherein if said virtual fetch address misses in both said micro-TLB and said selected nano-TLB, said control logic is configured to evict said physical address translation from one of said micro-TLB entries, and to refill said one of said micro-TLB entries with a physical address translation from said macro-TLB, to copy said evicted physical address translation to said each of said plurality of nano-TLBs whose said replacement information indicates said evicted one of said micro-TLB plurality of entries, and to update said replacement information of said selected nano-TLB to indicate said refilled one of said micro-TLB entries.

67. The microprocessor as recited in claim 66 , wherein each nano-TLB that has replacement information indicating said evicted one of said micro-TLB entries excludes said selected nano-TLB.

68. The microprocessor as recited in claim 66 , wherein if said virtual fetch address hits in said micro-TLB, said control logic is configured to update said replacement information of said selected nano-TLB to indicate said hitting one of said micro-TLB entries.

69. The microprocessor as recited in claim 66 , wherein if said virtual fetch address hits in said selected nano-TLB, said control logic is configured to update said replacement information of said selected nano-TLB to indicate an entry in said selected nano-TLB in which said virtual fetch address hits.

70. The microprocessor as recited in claim 66 , wherein if said virtual fetch address hits in said selected nano-TLB, said control logic is configured to update said replacement information of said selected nano-TLB to indicate none of said micro-TLB entries.

71. A computer program product for use with a computing device, the computer program product comprising:

a computer readable storage medium, having computer readable program code embodied in said medium for causing a three-tiered translation lookaside buffer (TLB) apparatus in a multithreading microprocessor that concurrently fetches and executes instructions of a plurality of threads, for providing virtual-to-physical address translation information for a memory page in response to an instruction cache virtual fetch address for one of the plurality of threads, said computer readable program code comprising:

first computer readable program code for providing a third-level TLB, for caching virtual-to-physical address translation information for a plurality of memory pages for the plurality of threads;

second computer readable program code for providing a second-level TLB, coupled to said third-level TLB, having a plurality of entries for caching said virtual-to-physical address translation information for a subset of said plurality of memory pages cached in said third-level TLB; and

third computer readable program code for providing a first-level TLB for each of a respective one of the plurality of threads, each said first-level TLB having at least one entry for caching virtual-to-physical address translation information only for said respective one of the plurality of threads.

72. The computer program product of claim 71 , wherein said computer readable program code further comprises:

fourth computer readable program code for providing storage, for storing replacement information indicating which of said first-level TLB at least one entry and said second-level TLB entries are caching most recently used virtual-to-physical address translation information for said respective one of the plurality of threads.

73. A hardware three-tiered translation lookaside buffer (TLB) apparatus in a multithreading microprocessor that concurrently fetches and executes instructions of a plurality of threads, the apparatus comprising:

a macro-TLB, for caching virtual-to-physical address translation information for a plurality of memory pages for the plurality of threads;

a micro-TLB, having a plurality of entries for caching said virtual-to-physical address translation information for a subset of said plurality of memory pages cached in said macro-TLB;

a nano-TLB for each of a respective one of the plurality of threads, each having at least one entry for caching virtual-to-physical address translation information only for said respective one of the plurality of threads, each of said nano-TLBs including a pointer for indicating one, if any, of said micro-TLB entries; and

control logic, coupled to said macro-, micro-, and nano-TLBs, configured to update said pointers in response to lookups, evictions and refills of said micro-TLB and said nano-TLBs, such that for each of the plurality of threads, either address translation information for a memory page from which an instruction was most recently fetched is cached in said respective nano-TLB and said pointer indicates none of said micro-TLB entries, or said pointer indicates one of said micro-TLB entries caching said most recent address translation information.

74. The computer program product of claim 72 , wherein said computer readable program code further comprises:

fifth program code for providing control logic, for evicting said virtual-to-physical address translation information from one of said second-level TLB entries, and refilling said one of said second-level TLB entries with virtual-to-physical address translation information from said third-level TLB, if the virtual fetch address misses in both said second-level TLB and said respective first-level TLB of said one of the plurality of threads associated with the virtual fetch address.

75. The computer program product of claim 72 , wherein each of said first-level TLB at least one entry comprises a plurality of entries.

76. The computer program product of claim 75 , wherein said replacement information comprises:

a single pointer, indicating which one of said first-level TLB at least one entry and said second-level TLB entries is caching most recently used virtual-to-physical address translation information for said respective one of the plurality of threads.

77. The computer program product of claim 76 , wherein said computer readable program code further comprises:

sixth program code for providing control logic, for copying said evicted information to one of said plurality of entries of each of said first-level TLBs whose said replacement information indicates said evicted one of said second-level TLB entries is caching said most recently used virtual-to-physical address translation information, if said single pointer indicates one of said second-level TLB plurality of entries is caching said most recently used virtual-to-physical address translation information for said respective one of the plurality of threads, and if said most recently used translation information is being evicted from said second-level TLB.

Assignments (11)