include/linux/preempt.h

1 /* SPDX-License-Identifier: GPL-2.0 */
6  * include/linux/preempt.h - macros for accessing and manipulating
18  * - bits 0-7 are the preemption count (max preemption depth: 256)
19  * - bits 8-15 are the softirq count (max # of softirqs: 256)
43 #define __IRQ_MASK(x)	((1UL << (x))-1)
60  * Disable preemption until the scheduler is running -- use an unconditional
63  * Reset by start_kernel()->sched_init()->init_idle()->init_idle_preempt_count().
69  * which states that during context switches:
82  * interrupt_context_level - return interrupt context level
84  * Returns the current interrupt context level.
85  *  0 - normal context
86  *  1 - softirq context
87  *  2 - hardirq context
88  *  3 - NMI context
111 # define softirq_count()	(current->softirq_disable_cnt & SOFTIRQ_MASK)
119  * Macros to retrieve the current execution context:
121  * in_nmi()		- We're in NMI context
122  * in_hardirq()		- We're in hard IRQ context
123  * in_serving_softirq()	- We're in softirq context
124  * in_task()		- We're in task context
137  * in_irq()       - Obsolete version of in_hardirq()
138  * in_softirq()   - We have BH disabled, or are processing softirqs
139  * in_interrupt() - We're in NMI,IRQ,SoftIRQ context or have BH disabled
180  * Are we running in atomic context?  WARNING: this macro cannot
181  * always detect atomic context; in particular, it cannot know about
182  * held spinlocks in non-preemptible kernels.  Thus it should not be
283  * that can cause faults and scheduling migrate into our preempt-protected
324  * preempt_ops - notifiers called when a task is preempted and rescheduled
344  * preempt_notifier - key for installing preemption notifiers
364 	notifier->link.next = NULL;  in preempt_notifier_init()
365 	notifier->link.pprev = NULL;  in preempt_notifier_init()
366 	notifier->ops = ops;  in preempt_notifier_init()
374  * Migrate-Disable and why it is undesired.
382  * Per this argument, the change from preempt_disable() to migrate_disable()
385  * - a higher priority tasks gains reduced wake-up latency; with preempt_disable()
388  * - a lower priority tasks; which under preempt_disable() could've instantly
402  * migration. This turns out to break a bunch of per-cpu usage. To this end,
406  * This is a 'temporary' work-around at best. The correct solution is getting
408  * per-cpu locking or short preempt-disable regions.
423  * Note: even non-work-conserving schedulers like semi-partitioned depends on
425  *       work-conserving schedulers.
439  * preempt_disable_nested - Disable preemption inside a normally preempt disabled section
442  * section which has preemption disabled implicitly on non-PREEMPT_RT
444  *  - holding a spinlock/rwlock
445  *  - soft interrupt context
446  *  - regular interrupt handlers
449  * interrupt context and regular interrupt handlers are preemptible and
452  * PREEMPT_RT. For non-PREEMPT_RT kernels this is a NOP.
456  *  - seqcount write side critical sections where the seqcount is not
460  *  - RMW per CPU variable updates like vmstat.
472  * preempt_enable_nested - Undo the effect of preempt_disable_nested()