DragonFly On-Line Manual Pages
SPINLOCK(9) DragonFly Kernel Developer's Manual SPINLOCK(9)
spin_init, spin_lock, spin_lock_quick, spin_trylock, spin_uninit,
spin_unlock, spin_unlock_quick, spin_pool_lock, spin_pool_unlock -- core
spin_init(struct spinlock *mtx);
spin_uninit(struct spinlock *mtx);
spin_lock(struct spinlock *mtx);
spin_lock_quick(globaldata_t gd, struct spinlock *mtx);
spin_trylock(struct spinlock *mtx);
spin_unlock(struct spinlock *mtx);
spin_unlock_quick(globaldata_t gd, struct spinlock *mtx);
The spinlock structure and call API are defined in the <sys/spinlock.h>
and <sys/spinlock2.h> header files, respectively.
The spin_init() function initializes a new spinlock structure for use.
The structure is cleaned up with spin_uninit() when it is no longer
The spin_lock() function obtains an exclusive read-write spinlock. A
thread may hold any number of exclusive spinlocks but should always be
mindful of ordering deadlocks. The spin_trylock() function will return
TRUE if the spinlock was successfully obtained and FALSE if it wasn't.
If you have the current CPU's globaldata pointer in hand you can call
spin_lock_quick(), but most code will just call the normal version. A
spinlock used only for exclusive access has about the same overhead as a
mutex based on a locked bus cycle.
A previously obtained exclusive spinlock is released by calling either
spin_unlock() or spin_unlock_quick().
The spin_pool_lock() function locks a pool spinlock associated with a
given address. The spinlock's lifetime is not tied to any objects at the
address. The function returns the locked spinlock. The
spin_pool_unlock() function unlocks a pool spinlock associated with an
address. Pool spinlocks need not be initialized.
A thread may not hold any spinlock across a blocking condition or thread
switch. LWKT tokens should be used for situations where you want an
exclusive run-time lock that will survive a blocking condition or thread
switch. Tokens will be automatically unlocked when a thread switches
away and relocked when the thread is switched back in. If you want a
lock that survives a blocking condition or thread switch without being
released, use lockmgr(9) locks or LWKT reader/writer locks.
DragonFly's core spinlocks should only be used around small contained
sections of code. For example, to manage a reference count or to imple-
ment higher level locking mechanisms. Both the token code and the
lockmgr(9) code use exclusive spinlocks internally. Core spinlocks
should not be used around large chunks of code.
Holding one or more spinlocks will disable thread preemption by another
thread (e.g. preemption by an interrupt thread), but will not disable
FAST interrupts or IPIs. This means that a FAST interrupt can still
operate during a spinlock, and any threaded interrupt (which is basically
all interrupts except the clock interrupt) will still be scheduled for
later execution, but will not be able to preempt the current thread. If
you wish to disable FAST interrupts and IPIs you need to enter a critical
section prior to obtaining the spinlock.
Currently, FAST interrupts, including IPI messages, are not allowed to
acquire any spinlocks. It is possible to work around this if
mycpu->gd_spinlocks_wr is 0. If one or the other is not zero, the FAST
interrupt or IPI cannot acquire any spinlocks without risking a deadlock,
even if the spinlocks in question are not related.
A thread may hold any number of exclusive read-write spinlocks.
Spinlocks spin. A thread will not block, switch away, or lose its criti-
cal section while obtaining or releasing a spinlock. Spinlocks do not
use IPIs or other mechanisms. They are considered to be a very low level
If a spinlock can not be obtained after one second a warning will be
printed on the console. If a system panic occurs, spinlocks will succeed
after one second in order to allow the panic operation to proceed.
If you have a complex structure such as a vnode(9) which contains a token
or lockmgr(9) lock, it is legal to directly access the internal spinlock
embedded in those structures for other purposes as long as the spinlock
is not held when you issue the token or lockmgr(9) operation.
The uncontended path of the spinlock implementation is in
/sys/sys/spinlock2.h. The core of the spinlock implementation is in
crit_enter(9), lockmgr(9), serializer(9)
A spinlock implementation first appeared in DragonFly 1.3.
The original spinlock implementation was written by Jeffrey M. Hsu and
was later extended by Matthew Dillon. This manual page was written by
Matthew Dillon and Sascha Wildner.
DragonFly 3.5 October 9, 2011 DragonFly 3.5