preempt-locking.txt 5.2 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135
  1. Proper Locking Under a Preemptible Kernel:
  2. Keeping Kernel Code Preempt-Safe
  3. Robert Love <rml@tech9.net>
  4. Last Updated: 28 Aug 2002
  5. INTRODUCTION
  6. A preemptible kernel creates new locking issues. The issues are the same as
  7. those under SMP: concurrency and reentrancy. Thankfully, the Linux preemptible
  8. kernel model leverages existing SMP locking mechanisms. Thus, the kernel
  9. requires explicit additional locking for very few additional situations.
  10. This document is for all kernel hackers. Developing code in the kernel
  11. requires protecting these situations.
  12. RULE #1: Per-CPU data structures need explicit protection
  13. Two similar problems arise. An example code snippet:
  14. struct this_needs_locking tux[NR_CPUS];
  15. tux[smp_processor_id()] = some_value;
  16. /* task is preempted here... */
  17. something = tux[smp_processor_id()];
  18. First, since the data is per-CPU, it may not have explicit SMP locking, but
  19. require it otherwise. Second, when a preempted task is finally rescheduled,
  20. the previous value of smp_processor_id may not equal the current. You must
  21. protect these situations by disabling preemption around them.
  22. You can also use put_cpu() and get_cpu(), which will disable preemption.
  23. RULE #2: CPU state must be protected.
  24. Under preemption, the state of the CPU must be protected. This is arch-
  25. dependent, but includes CPU structures and state not preserved over a context
  26. switch. For example, on x86, entering and exiting FPU mode is now a critical
  27. section that must occur while preemption is disabled. Think what would happen
  28. if the kernel is executing a floating-point instruction and is then preempted.
  29. Remember, the kernel does not save FPU state except for user tasks. Therefore,
  30. upon preemption, the FPU registers will be sold to the lowest bidder. Thus,
  31. preemption must be disabled around such regions.
  32. Note, some FPU functions are already explicitly preempt safe. For example,
  33. kernel_fpu_begin and kernel_fpu_end will disable and enable preemption.
  34. However, math_state_restore must be called with preemption disabled.
  35. RULE #3: Lock acquire and release must be performed by same task
  36. A lock acquired in one task must be released by the same task. This
  37. means you can't do oddball things like acquire a lock and go off to
  38. play while another task releases it. If you want to do something
  39. like this, acquire and release the task in the same code path and
  40. have the caller wait on an event by the other task.
  41. SOLUTION
  42. Data protection under preemption is achieved by disabling preemption for the
  43. duration of the critical region.
  44. preempt_enable() decrement the preempt counter
  45. preempt_disable() increment the preempt counter
  46. preempt_enable_no_resched() decrement, but do not immediately preempt
  47. preempt_check_resched() if needed, reschedule
  48. preempt_count() return the preempt counter
  49. The functions are nestable. In other words, you can call preempt_disable
  50. n-times in a code path, and preemption will not be reenabled until the n-th
  51. call to preempt_enable. The preempt statements define to nothing if
  52. preemption is not enabled.
  53. Note that you do not need to explicitly prevent preemption if you are holding
  54. any locks or interrupts are disabled, since preemption is implicitly disabled
  55. in those cases.
  56. But keep in mind that 'irqs disabled' is a fundamentally unsafe way of
  57. disabling preemption - any spin_unlock() decreasing the preemption count
  58. to 0 might trigger a reschedule. A simple printk() might trigger a reschedule.
  59. So use this implicit preemption-disabling property only if you know that the
  60. affected codepath does not do any of this. Best policy is to use this only for
  61. small, atomic code that you wrote and which calls no complex functions.
  62. Example:
  63. cpucache_t *cc; /* this is per-CPU */
  64. preempt_disable();
  65. cc = cc_data(searchp);
  66. if (cc && cc->avail) {
  67. __free_block(searchp, cc_entry(cc), cc->avail);
  68. cc->avail = 0;
  69. }
  70. preempt_enable();
  71. return 0;
  72. Notice how the preemption statements must encompass every reference of the
  73. critical variables. Another example:
  74. int buf[NR_CPUS];
  75. set_cpu_val(buf);
  76. if (buf[smp_processor_id()] == -1) printf(KERN_INFO "wee!\n");
  77. spin_lock(&buf_lock);
  78. /* ... */
  79. This code is not preempt-safe, but see how easily we can fix it by simply
  80. moving the spin_lock up two lines.
  81. PREVENTING PREEMPTION USING INTERRUPT DISABLING
  82. It is possible to prevent a preemption event using local_irq_disable and
  83. local_irq_save. Note, when doing so, you must be very careful to not cause
  84. an event that would set need_resched and result in a preemption check. When
  85. in doubt, rely on locking or explicit preemption disabling.
  86. Note in 2.5 interrupt disabling is now only per-CPU (e.g. local).
  87. An additional concern is proper usage of local_irq_disable and local_irq_save.
  88. These may be used to protect from preemption, however, on exit, if preemption
  89. may be enabled, a test to see if preemption is required should be done. If
  90. these are called from the spin_lock and read/write lock macros, the right thing
  91. is done. They may also be called within a spin-lock protected region, however,
  92. if they are ever called outside of this context, a test for preemption should
  93. be made. Do note that calls from interrupt context or bottom half/ tasklets
  94. are also protected by preemption locks and so may use the versions which do
  95. not check preemption.