Documentation/mutex-design.txt

   1 Generic Mutex Subsystem
   2
   3 started by Ingo Molnar <mingo@redhat.com>
   4
   5   "Why on earth do we need a new mutex subsystem, and what's wrong
   6    with semaphores?"
   7
   8 firstly, there's nothing wrong with semaphores. But if the simpler
   9 mutex semantics are sufficient for your code, then there are a couple
  10 of advantages of mutexes:
  11
  12  - 'struct mutex' is smaller on most architectures: .e.g on x86,
  13    'struct semaphore' is 20 bytes, 'struct mutex' is 16 bytes.
  14    A smaller structure size means less RAM footprint, and better
  15    CPU-cache utilization.
  16
  17  - tighter code. On x86 i get the following .text sizes when
  18    switching all mutex-alike semaphores in the kernel to the mutex
  19    subsystem:
  20
  21         text    data     bss     dec     hex filename
  22      3280380  868188  396860 4545428  455b94 vmlinux-semaphore
  23      3255329  865296  396732 4517357  44eded vmlinux-mutex
  24
  25    that's 25051 bytes of code saved, or a 0.76% win - off the hottest
  26    codepaths of the kernel. (The .data savings are 2892 bytes, or 0.33%)
  27    Smaller code means better icache footprint, which is one of the
  28    major optimization goals in the Linux kernel currently.
  29
  30  - the mutex subsystem is slightly faster and has better scalability for
  31    contended workloads. On an 8-way x86 system, running a mutex-based
  32    kernel and testing creat+unlink+close (of separate, per-task files)
  33    in /tmp with 16 parallel tasks, the average number of ops/sec is:
  34
  35     Semaphores:                        Mutexes:
  36
  37     $ ./test-mutex V 16 10             $ ./test-mutex V 16 10
  38     8 CPUs, running 16 tasks.          8 CPUs, running 16 tasks.
  39     checking VFS performance.          checking VFS performance.
  40     avg loops/sec:      34713          avg loops/sec:      84153
  41     CPU utilization:    63%            CPU utilization:    22%
  42
  43    i.e. in this workload, the mutex based kernel was 2.4 times faster
  44    than the semaphore based kernel, _and_ it also had 2.8 times less CPU
  45    utilization. (In terms of 'ops per CPU cycle', the semaphore kernel
  46    performed 551 ops/sec per 1% of CPU time used, while the mutex kernel
  47    performed 3825 ops/sec per 1% of CPU time used - it was 6.9 times
  48    more efficient.)
  49
  50    the scalability difference is visible even on a 2-way P4 HT box:
  51
  52     Semaphores:                        Mutexes:
  53
  54     $ ./test-mutex V 16 10             $ ./test-mutex V 16 10
  55     4 CPUs, running 16 tasks.          8 CPUs, running 16 tasks.
  56     checking VFS performance.          checking VFS performance.
  57     avg loops/sec:      127659         avg loops/sec:      181082
  58     CPU utilization:    100%           CPU utilization:    34%
  59
  60    (the straight performance advantage of mutexes is 41%, the per-cycle
  61     efficiency of mutexes is 4.1 times better.)
  62
  63  - there are no fastpath tradeoffs, the mutex fastpath is just as tight
  64    as the semaphore fastpath. On x86, the locking fastpath is 2
  65    instructions:
  66
  67     c0377ccb <mutex_lock>:
  68     c0377ccb:       f0 ff 08                lock decl (%eax)
  69     c0377cce:       78 0e                   js     c0377cde <.text.lock.mutex>
  70     c0377cd0:       c3                      ret
  71
  72    the unlocking fastpath is equally tight:
  73
  74     c0377cd1 <mutex_unlock>:
  75     c0377cd1:       f0 ff 00                lock incl (%eax)
  76     c0377cd4:       7e 0f                   jle    c0377ce5 <.text.lock.mutex+0x7>
  77     c0377cd6:       c3                      ret
  78
  79  - 'struct mutex' semantics are well-defined and are enforced if
  80    CONFIG_DEBUG_MUTEXES is turned on. Semaphores on the other hand have
  81    virtually no debugging code or instrumentation. The mutex subsystem
  82    checks and enforces the following rules:
  83
  84    * - only one task can hold the mutex at a time
  85    * - only the owner can unlock the mutex
  86    * - multiple unlocks are not permitted
  87    * - recursive locking is not permitted
  88    * - a mutex object must be initialized via the API
  89    * - a mutex object must not be initialized via memset or copying
  90    * - task may not exit with mutex held
  91    * - memory areas where held locks reside must not be freed
  92    * - held mutexes must not be reinitialized
  93    * - mutexes may not be used in irq contexts
  94
  95    furthermore, there are also convenience features in the debugging
  96    code:
  97
  98    * - uses symbolic names of mutexes, whenever they are printed in debug output
  99    * - point-of-acquire tracking, symbolic lookup of function names
 100    * - list of all locks held in the system, printout of them
 101    * - owner tracking
 102    * - detects self-recursing locks and prints out all relevant info
 103    * - detects multi-task circular deadlocks and prints out all affected
 104    *   locks and tasks (and only those tasks)
 105
 106 Disadvantages
 107 -------------
 108
 109 The stricter mutex API means you cannot use mutexes the same way you
 110 can use semaphores: e.g. they cannot be used from an interrupt context,
 111 nor can they be unlocked from a different context that which acquired
 112 it. [ I'm not aware of any other (e.g. performance) disadvantages from
 113 using mutexes at the moment, please let me know if you find any. ]
 114
 115 Implementation of mutexes
 116 -------------------------
 117
 118 'struct mutex' is the new mutex type, defined in include/linux/mutex.h
 119 and implemented in kernel/mutex.c. It is a counter-based mutex with a
 120 spinlock and a wait-list. The counter has 3 states: 1 for "unlocked",
 121 0 for "locked" and negative numbers (usually -1) for "locked, potential
 122 waiters queued".
 123
 124 the APIs of 'struct mutex' have been streamlined:
 125
 126  DEFINE_MUTEX(name);
 127
 128  mutex_init(mutex);
 129
 130  void mutex_lock(struct mutex *lock);
 131  int  mutex_lock_interruptible(struct mutex *lock);
 132  int  mutex_trylock(struct mutex *lock);
 133  void mutex_unlock(struct mutex *lock);
 134  int  mutex_is_locked(struct mutex *lock);
 135