1/* SPDX-License-Identifier: GPL-2.0 */
2#ifndef __LINUX_PREEMPT_H
3#define __LINUX_PREEMPT_H
4
5/*
6 * include/linux/preempt.h - macros for accessing and manipulating
7 * preempt_count (used for kernel preemption, interrupt count, etc.)
8 */
9
10#include <linux/linkage.h>
11#include <linux/list.h>
12
13/*
14 * We put the hardirq and softirq counter into the preemption
15 * counter. The bitmask has the following meaning:
16 *
17 * - bits 0-7 are the preemption count (max preemption depth: 256)
18 * - bits 8-15 are the softirq count (max # of softirqs: 256)
19 *
20 * The hardirq count could in theory be the same as the number of
21 * interrupts in the system, but we run all interrupt handlers with
22 * interrupts disabled, so we cannot have nesting interrupts. Though
23 * there are a few palaeontologic drivers which reenable interrupts in
24 * the handler, so we need more than one bit here.
25 *
26 * PREEMPT_MASK: 0x000000ff
27 * SOFTIRQ_MASK: 0x0000ff00
28 * HARDIRQ_MASK: 0x000f0000
29 * NMI_MASK: 0x00100000
30 * PREEMPT_NEED_RESCHED: 0x80000000
31 */
32#define PREEMPT_BITS 8
33#define SOFTIRQ_BITS 8
34#define HARDIRQ_BITS 4
35#define NMI_BITS 1
36
37#define PREEMPT_SHIFT 0
38#define SOFTIRQ_SHIFT (PREEMPT_SHIFT + PREEMPT_BITS)
39#define HARDIRQ_SHIFT (SOFTIRQ_SHIFT + SOFTIRQ_BITS)
40#define NMI_SHIFT (HARDIRQ_SHIFT + HARDIRQ_BITS)
41
42#define __IRQ_MASK(x) ((1UL << (x))-1)
43
44#define PREEMPT_MASK (__IRQ_MASK(PREEMPT_BITS) << PREEMPT_SHIFT)
45#define SOFTIRQ_MASK (__IRQ_MASK(SOFTIRQ_BITS) << SOFTIRQ_SHIFT)
46#define HARDIRQ_MASK (__IRQ_MASK(HARDIRQ_BITS) << HARDIRQ_SHIFT)
47#define NMI_MASK (__IRQ_MASK(NMI_BITS) << NMI_SHIFT)
48
49#define PREEMPT_OFFSET (1UL << PREEMPT_SHIFT)
50#define SOFTIRQ_OFFSET (1UL << SOFTIRQ_SHIFT)
51#define HARDIRQ_OFFSET (1UL << HARDIRQ_SHIFT)
52#define NMI_OFFSET (1UL << NMI_SHIFT)
53
54#define SOFTIRQ_DISABLE_OFFSET (2 * SOFTIRQ_OFFSET)
55
56/* We use the MSB mostly because its available */
57#define PREEMPT_NEED_RESCHED 0x80000000
58
59#define PREEMPT_DISABLED (PREEMPT_DISABLE_OFFSET + PREEMPT_ENABLED)
60
61/*
62 * Disable preemption until the scheduler is running -- use an unconditional
63 * value so that it also works on !PREEMPT_COUNT kernels.
64 *
65 * Reset by start_kernel()->sched_init()->init_idle()->init_idle_preempt_count().
66 */
67#define INIT_PREEMPT_COUNT PREEMPT_OFFSET
68
69/*
70 * Initial preempt_count value; reflects the preempt_count schedule invariant
71 * which states that during context switches:
72 *
73 * preempt_count() == 2*PREEMPT_DISABLE_OFFSET
74 *
75 * Note: PREEMPT_DISABLE_OFFSET is 0 for !PREEMPT_COUNT kernels.
76 * Note: See finish_task_switch().
77 */
78#define FORK_PREEMPT_COUNT (2*PREEMPT_DISABLE_OFFSET + PREEMPT_ENABLED)
79
80/* preempt_count() and related functions, depends on PREEMPT_NEED_RESCHED */
81#include <asm/preempt.h>
82
83#define hardirq_count() (preempt_count() & HARDIRQ_MASK)
84#define softirq_count() (preempt_count() & SOFTIRQ_MASK)
85#define irq_count() (preempt_count() & (HARDIRQ_MASK | SOFTIRQ_MASK \
86 | NMI_MASK))
87
88/*
89 * Are we doing bottom half or hardware interrupt processing?
90 *
91 * in_irq() - We're in (hard) IRQ context
92 * in_softirq() - We have BH disabled, or are processing softirqs
93 * in_interrupt() - We're in NMI,IRQ,SoftIRQ context or have BH disabled
94 * in_serving_softirq() - We're in softirq context
95 * in_nmi() - We're in NMI context
96 * in_task() - We're in task context
97 *
98 * Note: due to the BH disabled confusion: in_softirq(),in_interrupt() really
99 * should not be used in new code.
100 */
101#define in_irq() (hardirq_count())
102#define in_softirq() (softirq_count())
103#define in_interrupt() (irq_count())
104#define in_serving_softirq() (softirq_count() & SOFTIRQ_OFFSET)
105#define in_nmi() (preempt_count() & NMI_MASK)
106#define in_task() (!(preempt_count() & \
107 (NMI_MASK | HARDIRQ_MASK | SOFTIRQ_OFFSET)))
108
109/*
110 * The preempt_count offset after preempt_disable();
111 */
112#if defined(CONFIG_PREEMPT_COUNT)
113# define PREEMPT_DISABLE_OFFSET PREEMPT_OFFSET
114#else
115# define PREEMPT_DISABLE_OFFSET 0
116#endif
117
118/*
119 * The preempt_count offset after spin_lock()
120 */
121#define PREEMPT_LOCK_OFFSET PREEMPT_DISABLE_OFFSET
122
123/*
124 * The preempt_count offset needed for things like:
125 *
126 * spin_lock_bh()
127 *
128 * Which need to disable both preemption (CONFIG_PREEMPT_COUNT) and
129 * softirqs, such that unlock sequences of:
130 *
131 * spin_unlock();
132 * local_bh_enable();
133 *
134 * Work as expected.
135 */
136#define SOFTIRQ_LOCK_OFFSET (SOFTIRQ_DISABLE_OFFSET + PREEMPT_LOCK_OFFSET)
137
138/*
139 * Are we running in atomic context? WARNING: this macro cannot
140 * always detect atomic context; in particular, it cannot know about
141 * held spinlocks in non-preemptible kernels. Thus it should not be
142 * used in the general case to determine whether sleeping is possible.
143 * Do not use in_atomic() in driver code.
144 */
145#define in_atomic() (preempt_count() != 0)
146
147/*
148 * Check whether we were atomic before we did preempt_disable():
149 * (used by the scheduler)
150 */
151#define in_atomic_preempt_off() (preempt_count() != PREEMPT_DISABLE_OFFSET)
152
153#if defined(CONFIG_DEBUG_PREEMPT) || defined(CONFIG_TRACE_PREEMPT_TOGGLE)
154extern void preempt_count_add(int val);
155extern void preempt_count_sub(int val);
156#define preempt_count_dec_and_test() \
157 ({ preempt_count_sub(1); should_resched(0); })
158#else
159#define preempt_count_add(val) __preempt_count_add(val)
160#define preempt_count_sub(val) __preempt_count_sub(val)
161#define preempt_count_dec_and_test() __preempt_count_dec_and_test()
162#endif
163
164#define __preempt_count_inc() __preempt_count_add(1)
165#define __preempt_count_dec() __preempt_count_sub(1)
166
167#define preempt_count_inc() preempt_count_add(1)
168#define preempt_count_dec() preempt_count_sub(1)
169
170#ifdef CONFIG_PREEMPT_COUNT
171
172#define preempt_disable() \
173do { \
174 preempt_count_inc(); \
175 barrier(); \
176} while (0)
177
178#define sched_preempt_enable_no_resched() \
179do { \
180 barrier(); \
181 preempt_count_dec(); \
182} while (0)
183
184#define preempt_enable_no_resched() sched_preempt_enable_no_resched()
185
186#define preemptible() (preempt_count() == 0 && !irqs_disabled())
187
188#ifdef CONFIG_PREEMPT
189#define preempt_enable() \
190do { \
191 barrier(); \
192 if (unlikely(preempt_count_dec_and_test())) \
193 __preempt_schedule(); \
194} while (0)
195
196#define preempt_enable_notrace() \
197do { \
198 barrier(); \
199 if (unlikely(__preempt_count_dec_and_test())) \
200 __preempt_schedule_notrace(); \
201} while (0)
202
203#define preempt_check_resched() \
204do { \
205 if (should_resched(0)) \
206 __preempt_schedule(); \
207} while (0)
208
209#else /* !CONFIG_PREEMPT */
210#define preempt_enable() \
211do { \
212 barrier(); \
213 preempt_count_dec(); \
214} while (0)
215
216#define preempt_enable_notrace() \
217do { \
218 barrier(); \
219 __preempt_count_dec(); \
220} while (0)
221
222#define preempt_check_resched() do { } while (0)
223#endif /* CONFIG_PREEMPT */
224
225#define preempt_disable_notrace() \
226do { \
227 __preempt_count_inc(); \
228 barrier(); \
229} while (0)
230
231#define preempt_enable_no_resched_notrace() \
232do { \
233 barrier(); \
234 __preempt_count_dec(); \
235} while (0)
236
237#else /* !CONFIG_PREEMPT_COUNT */
238
239/*
240 * Even if we don't have any preemption, we need preempt disable/enable
241 * to be barriers, so that we don't have things like get_user/put_user
242 * that can cause faults and scheduling migrate into our preempt-protected
243 * region.
244 */
245#define preempt_disable() barrier()
246#define sched_preempt_enable_no_resched() barrier()
247#define preempt_enable_no_resched() barrier()
248#define preempt_enable() barrier()
249#define preempt_check_resched() do { } while (0)
250
251#define preempt_disable_notrace() barrier()
252#define preempt_enable_no_resched_notrace() barrier()
253#define preempt_enable_notrace() barrier()
254#define preemptible() 0
255
256#endif /* CONFIG_PREEMPT_COUNT */
257
258#ifdef MODULE
259/*
260 * Modules have no business playing preemption tricks.
261 */
262#undef sched_preempt_enable_no_resched
263#undef preempt_enable_no_resched
264#undef preempt_enable_no_resched_notrace
265#undef preempt_check_resched
266#endif
267
268#define preempt_set_need_resched() \
269do { \
270 set_preempt_need_resched(); \
271} while (0)
272#define preempt_fold_need_resched() \
273do { \
274 if (tif_need_resched()) \
275 set_preempt_need_resched(); \
276} while (0)
277
278#ifdef CONFIG_PREEMPT_NOTIFIERS
279
280struct preempt_notifier;
281
282/**
283 * preempt_ops - notifiers called when a task is preempted and rescheduled
284 * @sched_in: we're about to be rescheduled:
285 * notifier: struct preempt_notifier for the task being scheduled
286 * cpu: cpu we're scheduled on
287 * @sched_out: we've just been preempted
288 * notifier: struct preempt_notifier for the task being preempted
289 * next: the task that's kicking us out
290 *
291 * Please note that sched_in and out are called under different
292 * contexts. sched_out is called with rq lock held and irq disabled
293 * while sched_in is called without rq lock and irq enabled. This
294 * difference is intentional and depended upon by its users.
295 */
296struct preempt_ops {
297 void (*sched_in)(struct preempt_notifier *notifier, int cpu);
298 void (*sched_out)(struct preempt_notifier *notifier,
299 struct task_struct *next);
300};
301
302/**
303 * preempt_notifier - key for installing preemption notifiers
304 * @link: internal use
305 * @ops: defines the notifier functions to be called
306 *
307 * Usually used in conjunction with container_of().
308 */
309struct preempt_notifier {
310 struct hlist_node link;
311 struct preempt_ops *ops;
312};
313
314void preempt_notifier_inc(void);
315void preempt_notifier_dec(void);
316void preempt_notifier_register(struct preempt_notifier *notifier);
317void preempt_notifier_unregister(struct preempt_notifier *notifier);
318
319static inline void preempt_notifier_init(struct preempt_notifier *notifier,
320 struct preempt_ops *ops)
321{
322 INIT_HLIST_NODE(&notifier->link);
323 notifier->ops = ops;
324}
325
326#endif
327
328#endif /* __LINUX_PREEMPT_H */
329