rt-mutex.rst 3.5 KB

1234567891011121314151617181920212223242526272829303132333435363738394041424344454647484950515253545556575859606162636465666768697071727374757677
  1. ==================================
  2. RT-mutex subsystem with PI support
  3. ==================================
  4. RT-mutexes with priority inheritance are used to support PI-futexes,
  5. which enable pthread_mutex_t priority inheritance attributes
  6. (PTHREAD_PRIO_INHERIT). [See Documentation/locking/pi-futex.rst for more details
  7. about PI-futexes.]
  8. This technology was developed in the -rt tree and streamlined for
  9. pthread_mutex support.
  10. Basic principles:
  11. -----------------
  12. RT-mutexes extend the semantics of simple mutexes by the priority
  13. inheritance protocol.
  14. A low priority owner of a rt-mutex inherits the priority of a higher
  15. priority waiter until the rt-mutex is released. If the temporarily
  16. boosted owner blocks on a rt-mutex itself it propagates the priority
  17. boosting to the owner of the other rt_mutex it gets blocked on. The
  18. priority boosting is immediately removed once the rt_mutex has been
  19. unlocked.
  20. This approach allows us to shorten the block of high-prio tasks on
  21. mutexes which protect shared resources. Priority inheritance is not a
  22. magic bullet for poorly designed applications, but it allows
  23. well-designed applications to use userspace locks in critical parts of
  24. an high priority thread, without losing determinism.
  25. The enqueueing of the waiters into the rtmutex waiter tree is done in
  26. priority order. For same priorities FIFO order is chosen. For each
  27. rtmutex, only the top priority waiter is enqueued into the owner's
  28. priority waiters tree. This tree too queues in priority order. Whenever
  29. the top priority waiter of a task changes (for example it timed out or
  30. got a signal), the priority of the owner task is readjusted. The
  31. priority enqueueing is handled by "pi_waiters".
  32. RT-mutexes are optimized for fastpath operations and have no internal
  33. locking overhead when locking an uncontended mutex or unlocking a mutex
  34. without waiters. The optimized fastpath operations require cmpxchg
  35. support. [If that is not available then the rt-mutex internal spinlock
  36. is used]
  37. The state of the rt-mutex is tracked via the owner field of the rt-mutex
  38. structure:
  39. lock->owner holds the task_struct pointer of the owner. Bit 0 is used to
  40. keep track of the "lock has waiters" state:
  41. ============ ======= ================================================
  42. owner bit0 Notes
  43. ============ ======= ================================================
  44. NULL 0 lock is free (fast acquire possible)
  45. NULL 1 lock is free and has waiters and the top waiter
  46. is going to take the lock [1]_
  47. taskpointer 0 lock is held (fast release possible)
  48. taskpointer 1 lock is held and has waiters [2]_
  49. ============ ======= ================================================
  50. The fast atomic compare exchange based acquire and release is only
  51. possible when bit 0 of lock->owner is 0.
  52. .. [1] It also can be a transitional state when grabbing the lock
  53. with ->wait_lock is held. To prevent any fast path cmpxchg to the lock,
  54. we need to set the bit0 before looking at the lock, and the owner may
  55. be NULL in this small time, hence this can be a transitional state.
  56. .. [2] There is a small time when bit 0 is set but there are no
  57. waiters. This can happen when grabbing the lock in the slow path.
  58. To prevent a cmpxchg of the owner releasing the lock, we need to
  59. set this bit before looking at the lock.
  60. BTW, there is still technically a "Pending Owner", it's just not called
  61. that anymore. The pending owner happens to be the top_waiter of a lock
  62. that has no owner and has been woken up to grab the lock.