Home Explore Help
Register Sign In
phyus
/
linux-vybrid_public
8
0
Fork 0
Files Issues 0 Pull Requests 0 Wiki
Tree: f2e74eeac0
Branches Tags
linux-vybrid_pu...
/
kernel
/
sched_features.h

sched_features.h 2.1 KB
History Raw

1234567891011121314151617181920212223242526272829303132333435363738394041424344454647484950515253545556575859606162636465666768697071727374757677 
  1. /*
    2. 

  2.  * Only give sleepers 50% of their service deficit. This allows
    3. 

  3.  * them to run sooner, but does not allow tons of sleepers to
    4. 

  4.  * rip the spread apart.
    5. 

  5.  */
    6. 

  6. SCHED_FEAT(GENTLE_FAIR_SLEEPERS, 1)
    7. 

  7. 

  8. /*
    9. 

  9.  * Place new tasks ahead so that they do not starve already running
    10. 

  10.  * tasks
    11. 

  11.  */
    12. 

  12. SCHED_FEAT(START_DEBIT, 1)
    13. 

  13. 

  14. /*
    15. 

  15.  * Should wakeups try to preempt running tasks.
    16. 

  16.  */
    17. 

  17. SCHED_FEAT(WAKEUP_PREEMPT, 1)
    18. 

  18. 

  19. /*
    20. 

  20.  * When converting the wakeup granularity to virtual time, do it such
    21. 

  21.  * that heavier tasks preempting a lighter task have an edge.
    22. 

  22.  */
    23. 

  23. SCHED_FEAT(ASYM_GRAN, 1)
    24. 

  24. 

  25. /*
    26. 

  26.  * Use the SYNC wakeup hint, pipes and the likes use this to indicate
    27. 

  27.  * the remote end is likely to consume the data we just wrote, and
    28. 

  28.  * therefore has cache benefit from being placed on the same cpu, see
    29. 

  29.  * also AFFINE_WAKEUPS.
    30. 

  30.  */
    31. 

  31. SCHED_FEAT(SYNC_WAKEUPS, 1)
    32. 

  32. 

  33. /*
    34. 

  34.  * Based on load and program behaviour, see if it makes sense to place
    35. 

  35.  * a newly woken task on the same cpu as the task that woke it --
    36. 

  36.  * improve cache locality. Typically used with SYNC wakeups as
    37. 

  37.  * generated by pipes and the like, see also SYNC_WAKEUPS.
    38. 

  38.  */
    39. 

  39. SCHED_FEAT(AFFINE_WAKEUPS, 1)
    40. 

  40. 

  41. /*
    42. 

  42.  * Prefer to schedule the task we woke last (assuming it failed
    43. 

  43.  * wakeup-preemption), since its likely going to consume data we
    44. 

  44.  * touched, increases cache locality.
    45. 

  45.  */
    46. 

  46. SCHED_FEAT(NEXT_BUDDY, 0)
    47. 

  47. 

  48. /*
    49. 

  49.  * Prefer to schedule the task that ran last (when we did
    50. 

  50.  * wake-preempt) as that likely will touch the same data, increases
    51. 

  51.  * cache locality.
    52. 

  52.  */
    53. 

  53. SCHED_FEAT(LAST_BUDDY, 1)
    54. 

  54. 

  55. /*
    56. 

  56.  * Consider buddies to be cache hot, decreases the likelyness of a
    57. 

  57.  * cache buddy being migrated away, increases cache locality.
    58. 

  58.  */
    59. 

  59. SCHED_FEAT(CACHE_HOT_BUDDY, 1)
    60. 

  60. 

  61. /*
    62. 

  62.  * Use arch dependent cpu power functions
    63. 

  63.  */
    64. 

  64. SCHED_FEAT(ARCH_POWER, 0)
    65. 

  65. 

  66. SCHED_FEAT(HRTICK, 0)
    67. 

  67. SCHED_FEAT(DOUBLE_TICK, 0)
    68. 

  68. SCHED_FEAT(LB_BIAS, 1)
    69. 

  69. SCHED_FEAT(LB_SHARES_UPDATE, 1)
    70. 

  70. SCHED_FEAT(ASYM_EFF_LOAD, 1)
    71. 

  71. 

  72. /*
    73. 

  73.  * Spin-wait on mutex acquisition when the mutex owner is running on
    74. 

  74.  * another cpu -- assumes that when the owner is running, it will soon
    75. 

  75.  * release the lock. Decreases scheduling overhead.
    76. 

  76.  */
    77. 

  77. SCHED_FEAT(OWNER_SPIN, 1)
    78.