1 /* SPDX-License-Identifier: GPL-2.0 */
6  * seqcount_t / seqlock_t - a reader-writer consistency mechanism with
7  * lockless readers (read-only retry loops), and no writer starvation.
12  * - Based on x86_64 vsyscall gettimeofday: Keith Owens, Andrea Arcangeli
13  * - Sequence counters with associated locks, (C) 2020 Linutronix GmbH
17 #include <linux/kcsan-checks.h>
28  * read begin/retry/end. For readers, typically there is a call to
32  * As a consequence, we take the following best-effort approach for raw usage
33  * via seqcount_t under KCSAN: upon beginning a seq-reader critical section,
47 	lockdep_init_map(&s->dep_map, name, key, 0);  in __seqcount_init()
48 	s->sequence = 0;  in __seqcount_init()
57  * seqcount_init() - runtime initializer for seqcount_t
58  * @s: Pointer to the seqcount_t instance
72 	seqcount_acquire_read(&l->dep_map, 0, 0, _RET_IP_);  in seqcount_lockdep_reader_access()
73 	seqcount_release(&l->dep_map, _RET_IP_);  in seqcount_lockdep_reader_access()
84  * SEQCNT_ZERO() - static initializer for seqcount_t
93  * serialization at initialization time. This enables lockdep to validate
94  * that the write side critical section is properly serialized.
97  * preemption protection is enforced in the write side function.
99  * Lockdep is never used in any for the raw write variants.
105  * typedef seqcount_LOCKNAME_t - sequence counter with LOCKNAME associated
107  * @lock:	Pointer to the associated lock
110  * LOCKNAME @lock. The lock is associated to the sequence counter in the
111  * static initializer or init function. This enables lockdep to validate
112  * that the write side critical section is properly serialized.
118  * seqcount_LOCKNAME_init() - runtime initializer for seqcount_LOCKNAME_t
119  * @s:		Pointer to the seqcount_LOCKNAME_t instance
120  * @lock:	Pointer to the associated lock
126 		seqcount_init(&____s->seqcount);			\
127 		__SEQ_LOCK(____s->lock = (_lock));			\
136  * SEQCOUNT_LOCKNAME()	- Instantiate seqcount_LOCKNAME_t and helpers
137  * seqprop_LOCKNAME_*()	- Property accessors for seqcount_LOCKNAME_t
148 	return &s->seqcount;						\
154 	return &s->seqcount;						\
160 	unsigned seq = smp_load_acquire(&s->seqcount.sequence);		\
166 		__SEQ_LOCK(lockbase##_lock(s->lock));			\
167 		__SEQ_LOCK(lockbase##_unlock(s->lock));			\
170 		 * Re-read the sequence counter since the (possibly	\
173 		seq = smp_load_acquire(&s->seqcount.sequence);		\
192 	__SEQ_LOCK(lockdep_assert_held(s->lock));			\
211 	return smp_load_acquire(&s->sequence);  in __seqprop_sequence()
228 SEQCOUNT_LOCKNAME(rwlock,       rwlock_t,        __SEQ_RT, read)  in SEQCOUNT_LOCKNAME()
233  * SEQCNT_LOCKNAME_ZERO - static initializer for seqcount_LOCKNAME_t  in SEQCOUNT_LOCKNAME()
235  * @lock:	Pointer to the associated LOCKNAME  in SEQCOUNT_LOCKNAME()
266  * __read_seqcount_begin() - begin a seqcount_t read section
267  * @s: Pointer to seqcount_t or any of the seqcount_LOCKNAME_t variants
269  * Return: count to be passed to read_seqcount_retry()
283  * raw_read_seqcount_begin() - begin a seqcount_t read section w/o lockdep
284  * @s: Pointer to seqcount_t or any of the seqcount_LOCKNAME_t variants
286  * Return: count to be passed to read_seqcount_retry()
291  * read_seqcount_begin() - begin a seqcount_t read critical section
292  * @s: Pointer to seqcount_t or any of the seqcount_LOCKNAME_t variants
294  * Return: count to be passed to read_seqcount_retry()
303  * raw_read_seqcount() - read the raw seqcount_t counter value
304  * @s: Pointer to seqcount_t or any of the seqcount_LOCKNAME_t variants
306  * raw_read_seqcount opens a read critical section of the given
311  * Return: count to be passed to read_seqcount_retry()
322  * raw_seqcount_try_begin() - begin a seqcount_t read critical section
324  * @s: Pointer to seqcount_t or any of the seqcount_LOCKNAME_t variants
325  * @start: count to be passed to read_seqcount_retry()
327  * Similar to raw_seqcount_begin(), except it enables eliding the critical
331  * Useful when counter stabilization is more or less equivalent to taking
334  * If true, start will be set to the (even) sequence count read.
336  * Return: true when a read critical section is started.
345  * raw_seqcount_begin() - begin a seqcount_t read critical section w/o
347  * @s: Pointer to seqcount_t or any of the seqcount_LOCKNAME_t variants
349  * raw_seqcount_begin opens a read critical section of the given
351  * for the count to stabilize. If a writer is active when it begins, it
352  * will fail the read_seqcount_retry() at the end of the read critical
355  * Use this only in special kernel hot paths where the read section is
359  * Return: count to be passed to read_seqcount_retry()
371  * __read_seqcount_retry() - end a seqcount_t read section w/o barrier
372  * @s: Pointer to seqcount_t or any of the seqcount_LOCKNAME_t variants
377  * provided before actually loading any of the variables that are to be
383  * Return: true if a read section retry is required, else false
391 	return unlikely(READ_ONCE(s->sequence) != start);
395  * read_seqcount_retry() - end a seqcount_t read critical section
396  * @s: Pointer to seqcount_t or any of the seqcount_LOCKNAME_t variants
399  * read_seqcount_retry closes the read critical section of given
403  * Return: true if a read section retry is required, else false
415  * raw_write_seqcount_begin() - start a seqcount_t write section w/o lockdep
416  * @s: Pointer to seqcount_t or any of the seqcount_LOCKNAME_t variants
431 	s->sequence++;  in do_raw_write_seqcount_begin()
436  * raw_write_seqcount_end() - end a seqcount_t write section w/o lockdep
437  * @s: Pointer to seqcount_t or any of the seqcount_LOCKNAME_t variants
452 	s->sequence++;  in do_raw_write_seqcount_end()
457  * write_seqcount_begin_nested() - start a seqcount_t write section with
459  * @s: Pointer to seqcount_t or any of the seqcount_LOCKNAME_t variants
462  * See Documentation/locking/lockdep-design.rst
477 	seqcount_acquire(&s->dep_map, subclass, 0, _RET_IP_);  in do_write_seqcount_begin_nested()
482  * write_seqcount_begin() - start a seqcount_t write side critical section
483  * @s: Pointer to seqcount_t or any of the seqcount_LOCKNAME_t variants
485  * Context: sequence counter write side sections must be serialized and
486  * non-preemptible. Preemption will be automatically disabled if and
487  * only if the seqcount write serialization lock is associated, and
507  * write_seqcount_end() - end a seqcount_t write side critical section
508  * @s: Pointer to seqcount_t or any of the seqcount_LOCKNAME_t variants
510  * Context: Preemption will be automatically re-enabled if and only if
511  * the seqcount write serialization lock is associated, and preemptible.
523 	seqcount_release(&s->dep_map, _RET_IP_);  in do_write_seqcount_end()
528  * raw_write_seqcount_barrier() - do a seqcount_t write barrier
529  * @s: Pointer to seqcount_t or any of the seqcount_LOCKNAME_t variants
531  * This can be used to provide an ordering guarantee instead of the usual
533  * the two back-to-back wmb()s.
536  * via WRITE_ONCE): a) to ensure the writes become visible to other threads
537  * atomically, avoiding compiler optimizations; b) to document which writes are
538  * meant to propagate to the reader critical section. This is necessary because
539  * neither writes before nor after the barrier are enclosed in a seq-writer
545  *	void read(void)
559  *      void write(void)
574 	s->sequence++;  in do_raw_write_seqcount_barrier()
576 	s->sequence++;  in do_raw_write_seqcount_barrier()
581  * write_seqcount_invalidate() - invalidate in-progress seqcount_t read
583  * @s: Pointer to seqcount_t or any of the seqcount_LOCKNAME_t variants
585  * After write_seqcount_invalidate, no seqcount_t read side operations
595 	s->sequence+=2;  in do_write_seqcount_invalidate()
602  * A sequence counter variant where the counter even/odd value is used to
603  * switch between two copies of protected data. This allows the read path,
604  * typically NMIs, to safely interrupt the write side critical section.
606  * As the write sections are fully preemptible, no special handling for
614  * SEQCNT_LATCH_ZERO() - static initializer for seqcount_latch_t
622  * seqcount_latch_init() - runtime initializer for seqcount_latch_t
623  * @s: Pointer to the seqcount_latch_t instance
625 #define seqcount_latch_init(s) seqcount_init(&(s)->seqcount)
628  * raw_read_seqcount_latch() - pick even/odd latch data copy
629  * @s: Pointer to seqcount_latch_t
635  * picking which data copy to read. The full counter must then be checked
642 	 * Due to the dependent load, a full smp_rmb() is not needed.  in raw_read_seqcount_latch()
644 	return READ_ONCE(s->seqcount.sequence);  in raw_read_seqcount_latch()
648  * read_seqcount_latch() - pick even/odd latch data copy
649  * @s: Pointer to seqcount_latch_t
655  * picking which data copy to read. The full counter must then be checked
665  * raw_read_seqcount_latch_retry() - end a seqcount_latch_t read section
666  * @s:		Pointer to seqcount_latch_t
669  * Return: true if a read section retry is required, else false
675 	return unlikely(READ_ONCE(s->seqcount.sequence) != start);  in raw_read_seqcount_latch_retry()
679  * read_seqcount_latch_retry() - end a seqcount_latch_t read section
680  * @s:		Pointer to seqcount_latch_t
683  * Return: true if a read section retry is required, else false
693  * raw_write_seqcount_latch() - redirect latch readers to even/odd copy
694  * @s: Pointer to seqcount_latch_t
699 	s->seqcount.sequence++;  in raw_write_seqcount_latch()
704  * write_seqcount_latch_begin() - redirect latch readers to odd copy
705  * @s: Pointer to seqcount_latch_t
708  * queries during non-atomic modifications. If you can guarantee queries never
709  * interrupt the modification -- e.g. the concurrency is strictly between CPUs
710  * -- you most likely do not need this.
713  * modifications to ensure queries observe either the old or the new state the
714  * latch allows the same for non-atomic updates. The trade-off is doubling the
715  * cost of storage; we have to maintain two copies of the entire data
719  * there is always one copy in a stable state, ready to give us an answer.
728  * Where a modification, which is assumed to be externally serialized, does the
733  *		write_seqcount_latch_begin(&latch->seq);
734  *		modify(latch->data[0], ...);
735  *		write_seqcount_latch(&latch->seq);
736  *		modify(latch->data[1], ...);
737  *		write_seqcount_latch_end(&latch->seq);
748  *			seq = read_seqcount_latch(&latch->seq);
751  *			entry = data_query(latch->data[idx], ...);
754  *		} while (read_seqcount_latch_retry(&latch->seq, seq));
759  * So during the modification, queries are first redirected to data[1]. Then we
760  * modify data[0]. When that is complete, we redirect queries back to data[0]
765  *	The non-requirement for atomic modifications does _NOT_ include
770  *	to miss an entire modification sequence, once it resumes it might
776  *	patterns to manage the lifetimes of the objects within.
785  * write_seqcount_latch() - redirect latch readers to even copy
786  * @s: Pointer to seqcount_latch_t
794  * write_seqcount_latch_end() - end a seqcount_latch_t write section
795  * @s:		Pointer to seqcount_latch_t
798  * latch-protected data have been updated.
812  * seqlock_init() - dynamic initializer for seqlock_t
813  * @sl: Pointer to the seqlock_t instance
817 		spin_lock_init(&(sl)->lock);				\
818 		seqcount_spinlock_init(&(sl)->seqcount, &(sl)->lock);	\
822  * DEFINE_SEQLOCK(sl) - Define a statically allocated seqlock_t
829  * read_seqbegin() - start a seqlock_t read side critical section
830  * @sl: Pointer to seqlock_t
832  * Return: count, to be passed to read_seqretry()
836 	return read_seqcount_begin(&sl->seqcount);  in read_seqbegin()
840  * read_seqretry() - end a seqlock_t read side section
841  * @sl: Pointer to seqlock_t
844  * read_seqretry closes the read side critical section of given seqlock_t.
848  * Return: true if a read section retry is required, else false
852 	return read_seqcount_retry(&sl->seqcount, start);  in read_seqretry()
856  * For all seqlock_t write side functions, use the internal
862  * write_seqlock() - start a seqlock_t write side critical section
863  * @sl: Pointer to seqlock_t
865  * write_seqlock opens a write side critical section for the given
867  * that sequential lock. All seqlock_t write side sections are thus
868  * automatically serialized and non-preemptible.
870  * Context: if the seqlock_t read section, or other write side critical
876 	spin_lock(&sl->lock);  in write_seqlock()
877 	do_write_seqcount_begin(&sl->seqcount.seqcount);  in write_seqlock()
881  * write_sequnlock() - end a seqlock_t write side critical section
882  * @sl: Pointer to seqlock_t
884  * write_sequnlock closes the (serialized and non-preemptible) write side
889 	do_write_seqcount_end(&sl->seqcount.seqcount);  in write_sequnlock()
890 	spin_unlock(&sl->lock);  in write_sequnlock()
894  * write_seqlock_bh() - start a softirqs-disabled seqlock_t write section
895  * @sl: Pointer to seqlock_t
897  * _bh variant of write_seqlock(). Use only if the read side section, or
898  * other write side sections, can be invoked from softirq contexts.
902 	spin_lock_bh(&sl->lock);  in write_seqlock_bh()
903 	do_write_seqcount_begin(&sl->seqcount.seqcount);  in write_seqlock_bh()
907  * write_sequnlock_bh() - end a softirqs-disabled seqlock_t write section
908  * @sl: Pointer to seqlock_t
910  * write_sequnlock_bh closes the serialized, non-preemptible, and
911  * softirqs-disabled, seqlock_t write side critical section opened with
916 	do_write_seqcount_end(&sl->seqcount.seqcount);  in write_sequnlock_bh()
917 	spin_unlock_bh(&sl->lock);  in write_sequnlock_bh()
921  * write_seqlock_irq() - start a non-interruptible seqlock_t write section
922  * @sl: Pointer to seqlock_t
924  * _irq variant of write_seqlock(). Use only if the read side section, or
925  * other write sections, can be invoked from hardirq contexts.
929 	spin_lock_irq(&sl->lock);  in write_seqlock_irq()
930 	do_write_seqcount_begin(&sl->seqcount.seqcount);  in write_seqlock_irq()
934  * write_sequnlock_irq() - end a non-interruptible seqlock_t write section
935  * @sl: Pointer to seqlock_t
937  * write_sequnlock_irq closes the serialized and non-interruptible
938  * seqlock_t write side section opened with write_seqlock_irq().
942 	do_write_seqcount_end(&sl->seqcount.seqcount);  in write_sequnlock_irq()
943 	spin_unlock_irq(&sl->lock);  in write_sequnlock_irq()
950 	spin_lock_irqsave(&sl->lock, flags);  in __write_seqlock_irqsave()
951 	do_write_seqcount_begin(&sl->seqcount.seqcount);  in __write_seqlock_irqsave()
956  * write_seqlock_irqsave() - start a non-interruptible seqlock_t write
958  * @lock:  Pointer to seqlock_t
959  * @flags: Stack-allocated storage for saving caller's local interrupt
960  *         state, to be passed to write_sequnlock_irqrestore().
962  * _irqsave variant of write_seqlock(). Use it only if the read side
963  * section, or other write sections, can be invoked from hardirq context.
969  * write_sequnlock_irqrestore() - end non-interruptible seqlock_t write
971  * @sl:    Pointer to seqlock_t
974  * write_sequnlock_irqrestore closes the serialized and non-interruptible
975  * seqlock_t write section previously opened with write_seqlock_irqsave().
980 	do_write_seqcount_end(&sl->seqcount.seqcount);  in write_sequnlock_irqrestore()
981 	spin_unlock_irqrestore(&sl->lock, flags);  in write_sequnlock_irqrestore()
985  * read_seqlock_excl() - begin a seqlock_t locking reader section
986  * @sl:	Pointer to seqlock_t
994  * Context: if the seqlock_t write section, *or other read sections*, can
998  * The opened read section must be closed with read_sequnlock_excl().
1002 	spin_lock(&sl->lock);  in read_seqlock_excl()
1006  * read_sequnlock_excl() - end a seqlock_t locking reader critical section
1007  * @sl: Pointer to seqlock_t
1011 	spin_unlock(&sl->lock);  in read_sequnlock_excl()
1015  * read_seqlock_excl_bh() - start a seqlock_t locking reader section with
1017  * @sl: Pointer to seqlock_t
1020  * seqlock_t write side section, *or other read sections*, can be invoked
1025 	spin_lock_bh(&sl->lock);  in read_seqlock_excl_bh()
1029  * read_sequnlock_excl_bh() - stop a seqlock_t softirq-disabled locking
1031  * @sl: Pointer to seqlock_t
1035 	spin_unlock_bh(&sl->lock);  in read_sequnlock_excl_bh()
1039  * read_seqlock_excl_irq() - start a non-interruptible seqlock_t locking
1041  * @sl: Pointer to seqlock_t
1044  * write side section, *or other read sections*, can be invoked from a
1049 	spin_lock_irq(&sl->lock);  in read_seqlock_excl_irq()
1053  * read_sequnlock_excl_irq() - end an interrupts-disabled seqlock_t
1055  * @sl: Pointer to seqlock_t
1059 	spin_unlock_irq(&sl->lock);  in read_sequnlock_excl_irq()
1066 	spin_lock_irqsave(&sl->lock, flags);  in __read_seqlock_excl_irqsave()
1071  * read_seqlock_excl_irqsave() - start a non-interruptible seqlock_t
1073  * @lock:  Pointer to seqlock_t
1074  * @flags: Stack-allocated storage for saving caller's local interrupt
1075  *         state, to be passed to read_sequnlock_excl_irqrestore().
1078  * write side section, *or other read sections*, can be invoked from a
1085  * read_sequnlock_excl_irqrestore() - end non-interruptible seqlock_t
1087  * @sl:    Pointer to seqlock_t
1093 	spin_unlock_irqrestore(&sl->lock, flags);  in read_sequnlock_excl_irqrestore()
1097  * read_seqbegin_or_lock() - begin a seqlock_t lockless or locking reader
1098  * @lock: Pointer to seqlock_t
1102  * as in read_seqlock_excl().  In the first call to this function, the
1103  * caller *must* initialize and pass an even value to @seq; this way, a
1104  * lockless read can be optimistically tried first.
1106  * read_seqbegin_or_lock is an API designed to optimistically try a normal
1107  * lockless seqlock_t read section first.  If an odd counter is found, the
1108  * lockless read trial has failed, and the next read iteration transforms
1111  * This is typically used to avoid seqlock_t lockless readers starvation
1112  * (too much retry loops) in the case of a sharp spike in write side
1115  * Context: if the seqlock_t write section, *or other read sections*, can
1123  * value must be checked with need_seqretry(). If the read section need to
1136  * need_seqretry() - validate seqlock_t "locking or lockless" read section
1137  * @lock: Pointer to seqlock_t
1140  * Return: true if a read section retry is required, false otherwise
1148  * done_seqretry() - end seqlock_t "locking or lockless" reader section
1149  * @lock: Pointer to seqlock_t
1152  * done_seqretry finishes the seqlock_t read side critical section started
1162  * read_seqbegin_or_lock_irqsave() - begin a seqlock_t lockless reader, or
1163  *                                   a non-interruptible locking reader
1164  * @lock: Pointer to seqlock_t
1168  * the seqlock_t write section, *or other read sections*, can be invoked
1175  *   1. The saved local interrupts state in case of a locking reader, to
1176  *      be passed to done_seqretry_irqrestore().
1195  * done_seqretry_irqrestore() - end a seqlock_t lockless reader, or a
1196  *				non-interruptible locking reader section
1197  * @lock:  Pointer to seqlock_t
1202  * This is the _irqrestore variant of done_seqretry(). The read section