linux/Documentation/volatile-considered-harmful.txt
<<
>>
Prefs
   1Why the "volatile" type class should not be used
   2------------------------------------------------
   3
   4C programmers have often taken volatile to mean that the variable could be
   5changed outside of the current thread of execution; as a result, they are
   6sometimes tempted to use it in kernel code when shared data structures are
   7being used.  In other words, they have been known to treat volatile types
   8as a sort of easy atomic variable, which they are not.  The use of volatile in
   9kernel code is almost never correct; this document describes why.
  10
  11The key point to understand with regard to volatile is that its purpose is
  12to suppress optimization, which is almost never what one really wants to
  13do.  In the kernel, one must protect shared data structures against
  14unwanted concurrent access, which is very much a different task.  The
  15process of protecting against unwanted concurrency will also avoid almost
  16all optimization-related problems in a more efficient way.
  17
  18Like volatile, the kernel primitives which make concurrent access to data
  19safe (spinlocks, mutexes, memory barriers, etc.) are designed to prevent
  20unwanted optimization.  If they are being used properly, there will be no
  21need to use volatile as well.  If volatile is still necessary, there is
  22almost certainly a bug in the code somewhere.  In properly-written kernel
  23code, volatile can only serve to slow things down.
  24
  25Consider a typical block of kernel code:
  26
  27    spin_lock(&the_lock);
  28    do_something_on(&shared_data);
  29    do_something_else_with(&shared_data);
  30    spin_unlock(&the_lock);
  31
  32If all the code follows the locking rules, the value of shared_data cannot
  33change unexpectedly while the_lock is held.  Any other code which might
  34want to play with that data will be waiting on the lock.  The spinlock
  35primitives act as memory barriers - they are explicitly written to do so -
  36meaning that data accesses will not be optimized across them.  So the
  37compiler might think it knows what will be in shared_data, but the
  38spin_lock() call, since it acts as a memory barrier, will force it to
  39forget anything it knows.  There will be no optimization problems with
  40accesses to that data.
  41
  42If shared_data were declared volatile, the locking would still be
  43necessary.  But the compiler would also be prevented from optimizing access
  44to shared_data _within_ the critical section, when we know that nobody else
  45can be working with it.  While the lock is held, shared_data is not
  46volatile.  When dealing with shared data, proper locking makes volatile
  47unnecessary - and potentially harmful.
  48
  49The volatile storage class was originally meant for memory-mapped I/O
  50registers.  Within the kernel, register accesses, too, should be protected
  51by locks, but one also does not want the compiler "optimizing" register
  52accesses within a critical section.  But, within the kernel, I/O memory
  53accesses are always done through accessor functions; accessing I/O memory
  54directly through pointers is frowned upon and does not work on all
  55architectures.  Those accessors are written to prevent unwanted
  56optimization, so, once again, volatile is unnecessary.
  57
  58Another situation where one might be tempted to use volatile is
  59when the processor is busy-waiting on the value of a variable.  The right
  60way to perform a busy wait is:
  61
  62    while (my_variable != what_i_want)
  63        cpu_relax();
  64
  65The cpu_relax() call can lower CPU power consumption or yield to a
  66hyperthreaded twin processor; it also happens to serve as a compiler
  67barrier, so, once again, volatile is unnecessary.  Of course, busy-
  68waiting is generally an anti-social act to begin with.
  69
  70There are still a few rare situations where volatile makes sense in the
  71kernel:
  72
  73  - The above-mentioned accessor functions might use volatile on
  74    architectures where direct I/O memory access does work.  Essentially,
  75    each accessor call becomes a little critical section on its own and
  76    ensures that the access happens as expected by the programmer.
  77
  78  - Inline assembly code which changes memory, but which has no other
  79    visible side effects, risks being deleted by GCC.  Adding the volatile
  80    keyword to asm statements will prevent this removal.
  81
  82  - The jiffies variable is special in that it can have a different value
  83    every time it is referenced, but it can be read without any special
  84    locking.  So jiffies can be volatile, but the addition of other
  85    variables of this type is strongly frowned upon.  Jiffies is considered
  86    to be a "stupid legacy" issue (Linus's words) in this regard; fixing it
  87    would be more trouble than it is worth.
  88
  89  - Pointers to data structures in coherent memory which might be modified
  90    by I/O devices can, sometimes, legitimately be volatile.  A ring buffer
  91    used by a network adapter, where that adapter changes pointers to
  92    indicate which descriptors have been processed, is an example of this
  93    type of situation.
  94
  95For most code, none of the above justifications for volatile apply.  As a
  96result, the use of volatile is likely to be seen as a bug and will bring
  97additional scrutiny to the code.  Developers who are tempted to use
  98volatile should take a step back and think about what they are truly trying
  99to accomplish.
 100
 101Patches to remove volatile variables are generally welcome - as long as
 102they come with a justification which shows that the concurrency issues have
 103been properly thought through.
 104
 105
 106NOTES
 107-----
 108
 109[1] http://lwn.net/Articles/233481/
 110[2] http://lwn.net/Articles/233482/
 111
 112CREDITS
 113-------
 114
 115Original impetus and research by Randy Dunlap
 116Written by Jonathan Corbet
 117Improvements via comments from Satyam Sharma, Johannes Stezenbach, Jesper
 118        Juhl, Heikki Orsila, H. Peter Anvin, Philipp Hahn, and Stefan
 119        Richter.
 120