LTSLEEP(9)

📦 version 11.0, in nixpkgs unstable

💤 LTSLEEP(9) — process context sleep and wakeup 💤

okay bestie, this is LTSLEEP(9) from the Kernel Developer's Manual, and it's giving kernel sleep/wakeup lore. no cap, we're gonna get through it together. 🫶

🌸 NAME 🌸

ltsleep, mtsleep, tsleep, wakeup — process context sleep and wakeup. literally the whole group chat: sleeping in process context and waking up again. slay. 💅

✨ SYNOPSIS ✨

the include and signatures, exactly as they are, bestie:

#include <sys/proc.h>

int
mtsleep(wchan_t ident, pri_t priority, const char *wmesg, int timo,
    kmutex_t *mtx);

int
tsleep(wchan_t ident, pri_t priority, const char *wmesg, int timo);

void
wakeup(wchan_t ident);

💤 DESCRIPTION 💤

🚨 obsolete alert, bestie 🚨 The interfaces described in this manual page are obsolete and will be removed from a future version of the system.

The ltsleep() interface has been obsoleted and removed from the system. not me showing up to the function party after it got deleted 💀.

Please see the condvar(9), mutex(9), and rwlock(9) manual pages for information on kernel synchronisation primitives. those are the current main characters for this kind of thing. ✨

These functions implement voluntary context switching. tsleep() and mtsleep() are used throughout the kernel whenever processing in the current context can not continue for any of the following reasons:

  • The current process needs to await the results of a pending I/O operation.
  • The current process needs resources (e.g., memory) which are temporarily unavailable.

The function wakeup() is used to notify sleeping processes of possible changes to the condition that caused them to go to sleep. Typically, an awakened process will — after it has acquired a context again — retry the action that blocked its operation to see if the “blocking” condition has cleared. lowkey iconic behavior, highkey the blueprint. 💖

the arguments, bestie

The tsleep() and mtsleep() functions take the following arguments:

  • ident — An identifier of the “wait channel” representing the resource for which the current process needs to wait. This typically is the virtual address of some kernel data-structure related to the resource for which the process is contending. The same identifier must be used in a call to wakeup() to get the process going again. ident should not be NULL. 🚫
  • priority — The process priority to be used when the process is awakened and put on the queue of runnable processes. This mechanism is used to optimize “throughput” of processes executing in kernel mode. If the flag PCATCH is OR'ed into priority the process checks for posted signals before and after sleeping.
  • wmesg — A pointer to a character string indicating the reason a process is sleeping. The kernel does not use the string, but makes it available (through the process structure field p_wmesg) for user level utilities such as ps(1).
  • timo — If non-zero, the process will sleep for at most timo/hz seconds. If this amount of time elapses and no wakeup(ident) has occurred, and no signal (if PCATCH was set) was posted, tsleep() will return EWOULDBLOCK.

The mtsleep() function takes an additional argument and flag:

  • mtx — A mutex(9) representing the lock protecting the data-structures. On entry mtsleep() will release the lock and re-acquire the lock on return.
  • priority — If the flag PNORELOCK is OR'ed into priority then mtsleep() will not re-acquire the lock.

The wakeup() function will mark all processes which are currently sleeping on the identifier ident as runnable. Eventually, each of the processes will resume execution in the kernel context, causing a return from tsleep() or mtsleep().

⚠️ Note that processes returning from sleep should always re-evaluate the conditions that blocked them, since a call to wakeup() merely signals a possible change to the blocking conditions. fr fr, do the vibe check again. 👀

🎀 RETURN VALUES 🎀

  • tsleep() and mtsleep() return 0 if they return as a result of a wakeup().
  • If tsleep() and mtsleep() return as a result of a signal, the return value is ERESTART if the signal has the SA_RESTART property (see sigaction(2)), and EINTR otherwise.
  • If tsleep() and mtsleep() return because of a timeout, the return value is EWOULDBLOCK.

🦋 MIGRATING TO CONDVAR 🦋

okay so the conversion from tsleep/wakeup into condvar(9) should not be done mechanically, i.e. “blindly”. Code logic should be understood before changing, and it may also need to be revisited for the change. Please also read the condvar(9) man page. this is a “read the room first” situation, bestie. 📖

The tsleep() and mtsleep(), and wakeup() pairs should generally be replaced by cv_wait(9) / cv_wait_sig(9) / cv_timedwait(9) / cv_timedwait_sig(9) and cv_signal(9) / cv_broadcast(9) pairs. The cv_wait*() variant to use can be determined from looking at the corresponding tsleep() usage.

There are two arguments of interest: timo and priority. The priority value may have OR'ed the flag PCATCH.

  • The PCATCH flag means that the blocking thread should be awoken on signal, and the sleep call should be replaced with cv_wait_sig(9).
  • The timo value, if it is not zero, indicates how long to sleep, and the sleep call should be replaced with cv_timedwait(9).
  • If both the PCATCH flag and a non-zero timo value are specified, then cv_timedwait_sig(9) should be used.

A mutex(9) (interlock) must be held across cv_wait() and cv_broadcast() calls, in order to protect state. Most old code will require the addition of locking, whereas some will require amending to remove PNORELOCK. iykyk, locking is the whole plot. 🫶

🌈 SEE ALSO 🌈

⭐ HISTORY ⭐

The sleep/wakeup process synchronization mechanism is very old. It appeared in a very early version of Unix. tsleep() appeared in 4.4BSD. ltsleep() appeared in NetBSD 1.5. history really said “we've been here forever” 💀✨

xoxo 💖 this page is from Debian, dated May 7, 2024, and its manual title is LTSLEEP(9). bye bestie 🫶✨

📦 where this page comes from