Documentation/preempt-locking.txt

   1 ===========================================================================
   2 Proper Locking Under a Preemptible Kernel: Keeping Kernel Code Preempt-Safe
   3 ===========================================================================
   4
   5 :Author: Robert Love <rml@tech9.net>
   6 :Last Updated: 28 Aug 2002
   7
   8
   9 Introduction
  10 ============
  11
  12
  13 A preemptible kernel creates new locking issues.  The issues are the same as
  14 those under SMP: concurrency and reentrancy.  Thankfully, the Linux preemptible
  15 kernel model leverages existing SMP locking mechanisms.  Thus, the kernel
  16 requires explicit additional locking for very few additional situations.
  17
  18 This document is for all kernel hackers.  Developing code in the kernel
  19 requires protecting these situations.
  20
  21
  22 RULE #1: Per-CPU data structures need explicit protection
  23 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  24
  25
  26 Two similar problems arise. An example code snippet::
  27
  28         struct this_needs_locking tux[NR_CPUS];
  29         tux[smp_processor_id()] = some_value;
  30         /* task is preempted here... */
  31         something = tux[smp_processor_id()];
  32
  33 First, since the data is per-CPU, it may not have explicit SMP locking, but
  34 require it otherwise.  Second, when a preempted task is finally rescheduled,
  35 the previous value of smp_processor_id may not equal the current.  You must
  36 protect these situations by disabling preemption around them.
  37
  38 You can also use put_cpu() and get_cpu(), which will disable preemption.
  39
  40
  41 RULE #2: CPU state must be protected.
  42 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  43
  44
  45 Under preemption, the state of the CPU must be protected.  This is arch-
  46 dependent, but includes CPU structures and state not preserved over a context
  47 switch.  For example, on x86, entering and exiting FPU mode is now a critical
  48 section that must occur while preemption is disabled.  Think what would happen
  49 if the kernel is executing a floating-point instruction and is then preempted.
  50 Remember, the kernel does not save FPU state except for user tasks.  Therefore,
  51 upon preemption, the FPU registers will be sold to the lowest bidder.  Thus,
  52 preemption must be disabled around such regions.
  53
  54 Note, some FPU functions are already explicitly preempt safe.  For example,
  55 kernel_fpu_begin and kernel_fpu_end will disable and enable preemption.
  56 However, fpu__restore() must be called with preemption disabled.
  57
  58
  59 RULE #3: Lock acquire and release must be performed by same task
  60 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  61
  62
  63 A lock acquired in one task must be released by the same task.  This
  64 means you can't do oddball things like acquire a lock and go off to
  65 play while another task releases it.  If you want to do something
  66 like this, acquire and release the task in the same code path and
  67 have the caller wait on an event by the other task.
  68
  69
  70 Solution
  71 ========
  72
  73
  74 Data protection under preemption is achieved by disabling preemption for the
  75 duration of the critical region.
  76
  77 ::
  78
  79   preempt_enable()              decrement the preempt counter
  80   preempt_disable()             increment the preempt counter
  81   preempt_enable_no_resched()   decrement, but do not immediately preempt
  82   preempt_check_resched()       if needed, reschedule
  83   preempt_count()               return the preempt counter
  84
  85 The functions are nestable.  In other words, you can call preempt_disable
  86 n-times in a code path, and preemption will not be reenabled until the n-th
  87 call to preempt_enable.  The preempt statements define to nothing if
  88 preemption is not enabled.
  89
  90 Note that you do not need to explicitly prevent preemption if you are holding
  91 any locks or interrupts are disabled, since preemption is implicitly disabled
  92 in those cases.
  93
  94 But keep in mind that 'irqs disabled' is a fundamentally unsafe way of
  95 disabling preemption - any spin_unlock() decreasing the preemption count
  96 to 0 might trigger a reschedule. A simple printk() might trigger a reschedule.
  97 So use this implicit preemption-disabling property only if you know that the
  98 affected codepath does not do any of this. Best policy is to use this only for
  99 small, atomic code that you wrote and which calls no complex functions.
 100
 101 Example::
 102
 103         cpucache_t *cc; /* this is per-CPU */
 104         preempt_disable();
 105         cc = cc_data(searchp);
 106         if (cc && cc->avail) {
 107                 __free_block(searchp, cc_entry(cc), cc->avail);
 108                 cc->avail = 0;
 109         }
 110         preempt_enable();
 111         return 0;
 112
 113 Notice how the preemption statements must encompass every reference of the
 114 critical variables.  Another example::
 115
 116         int buf[NR_CPUS];
 117         set_cpu_val(buf);
 118         if (buf[smp_processor_id()] == -1) printf(KERN_INFO "wee!\n");
 119         spin_lock(&buf_lock);
 120         /* ... */
 121
 122 This code is not preempt-safe, but see how easily we can fix it by simply
 123 moving the spin_lock up two lines.
 124
 125
 126 Preventing preemption using interrupt disabling
 127 ===============================================
 128
 129
 130 It is possible to prevent a preemption event using local_irq_disable and
 131 local_irq_save.  Note, when doing so, you must be very careful to not cause
 132 an event that would set need_resched and result in a preemption check.  When
 133 in doubt, rely on locking or explicit preemption disabling.
 134
 135 Note in 2.5 interrupt disabling is now only per-CPU (e.g. local).
 136
 137 An additional concern is proper usage of local_irq_disable and local_irq_save.
 138 These may be used to protect from preemption, however, on exit, if preemption
 139 may be enabled, a test to see if preemption is required should be done.  If
 140 these are called from the spin_lock and read/write lock macros, the right thing
 141 is done.  They may also be called within a spin-lock protected region, however,
 142 if they are ever called outside of this context, a test for preemption should
 143 be made. Do note that calls from interrupt context or bottom half/ tasklets
 144 are also protected by preemption locks and so may use the versions which do
 145 not check preemption.