linux/Documentation/spinlocks.txt
<<
>>
Prefs
   1SPIN_LOCK_UNLOCKED and RW_LOCK_UNLOCKED defeat lockdep state tracking and
   2are hence deprecated.
   3
   4Please use DEFINE_SPINLOCK()/DEFINE_RWLOCK() or
   5__SPIN_LOCK_UNLOCKED()/__RW_LOCK_UNLOCKED() as appropriate for static
   6initialization.
   7
   8Most of the time, you can simply turn:
   9
  10        static spinlock_t xxx_lock = SPIN_LOCK_UNLOCKED;
  11
  12into:
  13
  14        static DEFINE_SPINLOCK(xxx_lock);
  15
  16Static structure member variables go from:
  17
  18        struct foo bar {
  19                .lock   =       SPIN_LOCK_UNLOCKED;
  20        };
  21
  22to:
  23
  24        struct foo bar {
  25                .lock   =       __SPIN_LOCK_UNLOCKED(bar.lock);
  26        };
  27
  28Declaration of static rw_locks undergo a similar transformation.
  29
  30Dynamic initialization, when necessary, may be performed as
  31demonstrated below.
  32
  33   spinlock_t xxx_lock;
  34   rwlock_t xxx_rw_lock;
  35
  36   static int __init xxx_init(void)
  37   {
  38        spin_lock_init(&xxx_lock);
  39        rwlock_init(&xxx_rw_lock);
  40        ...
  41   }
  42
  43   module_init(xxx_init);
  44
  45The following discussion is still valid, however, with the dynamic
  46initialization of spinlocks or with DEFINE_SPINLOCK, etc., used
  47instead of SPIN_LOCK_UNLOCKED.
  48
  49-----------------------
  50
  51On Fri, 2 Jan 1998, Doug Ledford wrote:
  52> 
  53> I'm working on making the aic7xxx driver more SMP friendly (as well as
  54> importing the latest FreeBSD sequencer code to have 7895 support) and wanted
  55> to get some info from you.  The goal here is to make the various routines
  56> SMP safe as well as UP safe during interrupts and other manipulating
  57> routines.  So far, I've added a spin_lock variable to things like my queue
  58> structs.  Now, from what I recall, there are some spin lock functions I can
  59> use to lock these spin locks from other use as opposed to a (nasty)
  60> save_flags(); cli(); stuff; restore_flags(); construct.  Where do I find
  61> these routines and go about making use of them?  Do they only lock on a
  62> per-processor basis or can they also lock say an interrupt routine from
  63> mucking with a queue if the queue routine was manipulating it when the
  64> interrupt occurred, or should I still use a cli(); based construct on that
  65> one?
  66
  67See <asm/spinlock.h>. The basic version is:
  68
  69   spinlock_t xxx_lock = SPIN_LOCK_UNLOCKED;
  70
  71
  72        unsigned long flags;
  73
  74        spin_lock_irqsave(&xxx_lock, flags);
  75        ... critical section here ..
  76        spin_unlock_irqrestore(&xxx_lock, flags);
  77
  78and the above is always safe. It will disable interrupts _locally_, but the
  79spinlock itself will guarantee the global lock, so it will guarantee that
  80there is only one thread-of-control within the region(s) protected by that
  81lock. 
  82
  83Note that it works well even under UP - the above sequence under UP
  84essentially is just the same as doing a
  85
  86        unsigned long flags;
  87
  88        save_flags(flags); cli();
  89         ... critical section ...
  90        restore_flags(flags);
  91
  92so the code does _not_ need to worry about UP vs SMP issues: the spinlocks
  93work correctly under both (and spinlocks are actually more efficient on
  94architectures that allow doing the "save_flags + cli" in one go because I
  95don't export that interface normally).
  96
  97NOTE NOTE NOTE! The reason the spinlock is so much faster than a global
  98interrupt lock under SMP is exactly because it disables interrupts only on
  99the local CPU. The spin-lock is safe only when you _also_ use the lock
 100itself to do locking across CPU's, which implies that EVERYTHING that
 101touches a shared variable has to agree about the spinlock they want to
 102use.
 103
 104The above is usually pretty simple (you usually need and want only one
 105spinlock for most things - using more than one spinlock can make things a
 106lot more complex and even slower and is usually worth it only for
 107sequences that you _know_ need to be split up: avoid it at all cost if you
 108aren't sure). HOWEVER, it _does_ mean that if you have some code that does
 109
 110        cli();
 111        .. critical section ..
 112        sti();
 113
 114and another sequence that does
 115
 116        spin_lock_irqsave(flags);
 117        .. critical section ..
 118        spin_unlock_irqrestore(flags);
 119
 120then they are NOT mutually exclusive, and the critical regions can happen
 121at the same time on two different CPU's. That's fine per se, but the
 122critical regions had better be critical for different things (ie they
 123can't stomp on each other). 
 124
 125The above is a problem mainly if you end up mixing code - for example the
 126routines in ll_rw_block() tend to use cli/sti to protect the atomicity of
 127their actions, and if a driver uses spinlocks instead then you should
 128think about issues like the above..
 129
 130This is really the only really hard part about spinlocks: once you start
 131using spinlocks they tend to expand to areas you might not have noticed
 132before, because you have to make sure the spinlocks correctly protect the
 133shared data structures _everywhere_ they are used. The spinlocks are most
 134easily added to places that are completely independent of other code (ie
 135internal driver data structures that nobody else ever touches, for
 136example). 
 137
 138----
 139
 140Lesson 2: reader-writer spinlocks.
 141
 142If your data accesses have a very natural pattern where you usually tend
 143to mostly read from the shared variables, the reader-writer locks
 144(rw_lock) versions of the spinlocks are often nicer. They allow multiple
 145readers to be in the same critical region at once, but if somebody wants
 146to change the variables it has to get an exclusive write lock. The
 147routines look the same as above:
 148
 149   rwlock_t xxx_lock = RW_LOCK_UNLOCKED;
 150
 151
 152        unsigned long flags;
 153
 154        read_lock_irqsave(&xxx_lock, flags);
 155        .. critical section that only reads the info ...
 156        read_unlock_irqrestore(&xxx_lock, flags);
 157
 158        write_lock_irqsave(&xxx_lock, flags);
 159        .. read and write exclusive access to the info ...
 160        write_unlock_irqrestore(&xxx_lock, flags);
 161
 162The above kind of lock is useful for complex data structures like linked
 163lists etc, especially when you know that most of the work is to just
 164traverse the list searching for entries without changing the list itself,
 165for example. Then you can use the read lock for that kind of list
 166traversal, which allows many concurrent readers. Anything that _changes_
 167the list will have to get the write lock. 
 168
 169Note: you cannot "upgrade" a read-lock to a write-lock, so if you at _any_
 170time need to do any changes (even if you don't do it every time), you have
 171to get the write-lock at the very beginning. I could fairly easily add a
 172primitive to create a "upgradeable" read-lock, but it hasn't been an issue
 173yet. Tell me if you'd want one. 
 174
 175----
 176
 177Lesson 3: spinlocks revisited.
 178
 179The single spin-lock primitives above are by no means the only ones. They
 180are the most safe ones, and the ones that work under all circumstances,
 181but partly _because_ they are safe they are also fairly slow. They are
 182much faster than a generic global cli/sti pair, but slower than they'd
 183need to be, because they do have to disable interrupts (which is just a
 184single instruction on a x86, but it's an expensive one - and on other
 185architectures it can be worse).
 186
 187If you have a case where you have to protect a data structure across
 188several CPU's and you want to use spinlocks you can potentially use
 189cheaper versions of the spinlocks. IFF you know that the spinlocks are
 190never used in interrupt handlers, you can use the non-irq versions:
 191
 192        spin_lock(&lock);
 193        ...
 194        spin_unlock(&lock);
 195
 196(and the equivalent read-write versions too, of course). The spinlock will
 197guarantee the same kind of exclusive access, and it will be much faster. 
 198This is useful if you know that the data in question is only ever
 199manipulated from a "process context", ie no interrupts involved. 
 200
 201The reasons you mustn't use these versions if you have interrupts that
 202play with the spinlock is that you can get deadlocks:
 203
 204        spin_lock(&lock);
 205        ...
 206                <- interrupt comes in:
 207                        spin_lock(&lock);
 208
 209where an interrupt tries to lock an already locked variable. This is ok if
 210the other interrupt happens on another CPU, but it is _not_ ok if the
 211interrupt happens on the same CPU that already holds the lock, because the
 212lock will obviously never be released (because the interrupt is waiting
 213for the lock, and the lock-holder is interrupted by the interrupt and will
 214not continue until the interrupt has been processed). 
 215
 216(This is also the reason why the irq-versions of the spinlocks only need
 217to disable the _local_ interrupts - it's ok to use spinlocks in interrupts
 218on other CPU's, because an interrupt on another CPU doesn't interrupt the
 219CPU that holds the lock, so the lock-holder can continue and eventually
 220releases the lock). 
 221
 222Note that you can be clever with read-write locks and interrupts. For
 223example, if you know that the interrupt only ever gets a read-lock, then
 224you can use a non-irq version of read locks everywhere - because they
 225don't block on each other (and thus there is no dead-lock wrt interrupts. 
 226But when you do the write-lock, you have to use the irq-safe version. 
 227
 228For an example of being clever with rw-locks, see the "waitqueue_lock" 
 229handling in kernel/sched.c - nothing ever _changes_ a wait-queue from
 230within an interrupt, they only read the queue in order to know whom to
 231wake up. So read-locks are safe (which is good: they are very common
 232indeed), while write-locks need to protect themselves against interrupts.
 233
 234                Linus
 235
 236
 237
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.