Copyright (c) 2000-2004 John H. Baldwin <jhb@FreeBSD.org> All rights reserved. Redistribution and use in source and binary forms, with or without modification, are permitted provided that the following conditions are met: 1. Redistributions of source code must retain the above copyright notice, this list of conditions and the following disclaimer. 2. Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the ...
NAMEinit_sleepqueues sleepq_abort sleepq_add sleepq_alloc sleepq_broadcast sleepq_calc_signal_retval sleepq_catch_signals sleepq_free sleepq_lock sleepq_lookup sleepq_release sleepq_remove sleepq_signal sleepq_set_timeout sleepq_set_timeout_sbt sleepq_sleepcnt sleepq_timedwait sleepq_timedwait_sig sleepq_type sleepq_wait sleepq_wait_sig - manage the queues of sleeping threads
SYNOPSISIn sys/param.h In sys/sleepqueue.h Ft void Fn init_sleepqueues void Ft int Fn sleepq_abort struct thread *td Ft void Fn sleepq_add void *wchan struct lock_object *lock const char *wmesg int flags int queue Ft struct sleepqueue * Fn sleepq_alloc void Ft int Fn sleepq_broadcast void *wchan int flags int pri int queue Ft int Fn sleepq_calc_signal_retval int sig Ft int Fn sleepq_catch_signals void *wchan Ft void Fn sleepq_free struct sleepqueue *sq Ft struct sleepqueue * Fn sleepq_lookup void *wchan Ft void Fn sleepq_lock void *wchan Ft void Fn sleepq_release void *wchan Ft void Fn sleepq_remove struct thread *td void *wchan Ft int Fn sleepq_signal void *wchan int flags int pri int queue Ft void Fn sleepq_set_timeout void *wchan int timo Ft void Fn sleepq_set_timeout_sbt void *wchan sbintime_t sbt sbintime_t pr int flags Ft u_int Fn sleepq_sleepcnt void *wchan int queue Ft int Fn sleepq_timedwait void *wchan Ft int Fn sleepq_timedwait_sig void *wchan int signal_caught Ft int Fn sleepq_type void *wchan Ft void Fn sleepq_wait void *wchan Ft int Fn sleepq_wait_sig void *wchan
DESCRIPTIONSleep queues provide a mechanism for suspending execution of a thread until some condition is met. Each queue is associated with a specific wait channel when it is active, and only one queue may be associated with a wait channel at any given point in time. The implementation of each wait channel splits its sleepqueue into 2 sub-queues in order to enable some optimizations on threads' wakeups. An active queue holds a list of threads that are blocked on the associated wait channel. Threads that are not blocked on a wait channel have an associated inactive sleep queue. When a thread blocks on a wait channel it donates its inactive sleep queue to the wait channel. When a thread is resumed, the wait channel that it was blocked on gives it an inactive sleep queue for later use.
The Fn sleepq_alloc function allocates an inactive sleep queue and is used to assign a sleep queue to a thread during thread creation. The Fn sleepq_free function frees the resources associated with an inactive sleep queue and is used to free a queue during thread destruction.
Active sleep queues are stored in a hash table hashed on the addresses pointed to by wait channels. Each bucket in the hash table contains a sleep queue chain. A sleep queue chain contains a spin mutex and a list of sleep queues that hash to that specific chain. Active sleep queues are protected by their chain's spin mutex. The Fn init_sleepqueues function initializes the hash table of sleep queue chains.
The Fn sleepq_lock function locks the sleep queue chain associated with wait channel Fa wchan .
The Fn sleepq_lookup returns a pointer to the currently active sleep queue for that wait channel associated with Fa wchan or NULL if there is no active sleep queue associated with argument Fa wchan . It requires the sleep queue chain associated with Fa wchan to have been locked by a prior call to Fn sleepq_lock .
The Fn sleepq_release function unlocks the sleep queue chain associated with Fn wchan and is primarily useful when aborting a pending sleep request before one of the wait functions is called.
The Fn sleepq_add function places the current thread on the sleep queue associated with the wait channel Fa wchan . The sleep queue chain associated with argument Fa wchan must be locked by a prior call to Fn sleepq_lock when this function is called. If a lock is specified via the Fa lock argument, and if the kernel was compiled with options INVARIANTS then the sleep queue code will perform extra checks to ensure that the lock is used by all threads sleeping on Fa wchan . The Fa wmesg parameter should be a short description of Fa wchan . The Fa flags parameter is a bitmask consisting of the type of sleep queue being slept on and zero or more optional flags. The Fa queue parameter specifies the sub-queue, in which the contending thread will be inserted.
There are currently three types of sleep queues:
- A sleep queue used to implement condition variables.
- A sleep queue used to implement sleep(9), wakeup(9) and wakeup_one9.
- A sleep queue used to implement pause(9).
There are currently two optional flag:
- The current thread is entering an interruptible sleep.
- When thread is entering an interruptible sleep, do not stop it upon arrival of stop action, like SIGSTOP Wake it up instead.
A timeout on the sleep may be specified by calling Fn sleepq_set_timeout after Fn sleepq_add . The Fa wchan parameter should be the same value from the preceding call to Fn sleepq_add , and the sleep queue chain associated with Fa wchan must have been locked by a prior call to Fn sleepq_lock . The Fa timo parameter should specify the timeout value in ticks.
Fn sleepq_set_timeout_sbt function takes Fa sbt argument instead of Fa timo . It allows to specify relative or absolute wakeup time with higher resolution in form of Vt sbintime_t . The parameter Fa pr allows to specify wanted absolute event precision. The parameter Fa flags allows to pass additional Fn callout_reset_sbt flags.
The current thread may be marked interruptible by calling Fn sleepq_catch_signals with Fa wchan set to the wait channel. This function returns a signal number if there are any pending signals for the current thread and 0 if there is not a pending signal. The sleep queue chain associated with argument Fa wchan should have been locked by a prior call to Fn sleepq_lock .
Once the thread is ready to suspend, one of the wait functions is called to put the current thread to sleep until it is awakened and to context switch to another thread. The Fn sleepq_wait function is used for non-interruptible sleeps that do not have a timeout. The Fn sleepq_timedwait function is used for non-interruptible sleeps that have had a timeout set via Fn sleepq_set_timeout . The Fn sleepq_wait_sig function is used for interruptible sleeps that do not have a timeout. The Fn sleepq_timedwait_sig function is used for interruptible sleeps that do have a timeout set. The Fa wchan argument to all of the wait functions is the wait channel being slept on. The sleep queue chain associated with argument Fa wchan needs to have been locked with a prior call to Fn sleepq_lock . The Fa signal_caught parameter to Fn sleepq_timedwait_sig specifies if a previous call to Fn sleepq_catch_signals found a pending signal.
When the thread is resumed, the wait functions return a non-zero value if the thread was awakened due to an interrupt other than a signal or a timeout. If the sleep timed out, then Er EWOULDBLOCK is returned. If the sleep was interrupted by something other than a signal, then some other return value will be returned. If zero is returned after resuming from an interruptible sleep, then Fn sleepq_calc_signal_retval should be called to determine if the sleep was interrupted by a signal. If so, Fn sleepq_calc_signal_retval returns Er ERESTART if the interrupting signal is restartable and Er EINTR otherwise. If the sleep was not interrupted by a signal, Fn sleepq_calc_signal_retval will return 0.
A sleeping thread is normally resumed by the Fn sleepq_broadcast and Fn sleepq_signal functions. The Fn sleepq_signal function awakens the highest priority thread sleeping on a wait channel while Fn sleepq_broadcast awakens all of the threads sleeping on a wait channel. The Fa wchan argument specifics which wait channel to awaken. The Fa flags argument must match the sleep queue type contained in the Fa flags argument passed to Fn sleepq_add by the threads sleeping on the wait channel. If the Fa pri argument does not equal -1, then each thread that is awakened will have its priority raised to Fa pri if it has a lower priority. The sleep queue chain associated with argument Fa wchan must be locked by a prior call to Fn sleepq_lock before calling any of these functions. The Fa queue argument specifies the sub-queue, from which threads need to be woken up.
A thread in an interruptible sleep can be interrupted by another thread via the Fn sleepq_abort function. The Fa td argument specifies the thread to interrupt. An individual thread can also be awakened from sleeping on a specific wait channel via the Fn sleepq_remove function. The Fa td argument specifies the thread to awaken and the Fa wchan argument specifies the wait channel to awaken it from. If the thread Fa td is not blocked on the wait channel Fa wchan then this function will not do anything, even if the thread is asleep on a different wait channel. This function should only be used if one of the other functions above is not sufficient. One possible use is waking up a specific thread from a widely shared sleep channel.
The Fn sleepq_sleepcnt function offer a simple way to retrieve the number of threads sleeping for the specified Fa queue , given a Fa wchan .
The Fn sleepq_type function returns the type of Fa wchan associated to a sleepqueue.
The Fn sleepq_abort , Fn sleepq_broadcast , and Fn sleepq_signal functions all return a boolean value. If the return value is true, then at least one thread was resumed that is currently swapped out. The caller is responsible for awakening the scheduler process so that the resumed thread will be swapped back in. This is done by calling the Fn kick_proc0 function after releasing the sleep queue chain lock via a call to Fn sleepq_release .
The sleep queue interface is currently used to implement the sleep(9) and condvar(9) interfaces. Almost all other code in the kernel should use one of those interfaces rather than manipulating sleep queues directly.