Mirrors_Framework/nuttx
1
0
Fork 0
mirror of https://github.com/apache/nuttx.git synced 2026-05-30 05:16:47 +08:00
Code Issues Actions 2 Packages Projects Releases Wiki Activity

seqlock: Improve seqlock performance on non-SMP platforms.

Browse Source 
This commit implemented seqlock for non-SMP platforms, which achieves
1.62x performance improvement.

Signed-off-by: ouyangxiangzhen <ouyangxiangzhen@xiaomi.com>
This commit is contained in:
ouyangxiangzhen
2025-09-23 17:13:01 +08:00
committed by Alan C. Assis
parent a8724387e7
commit d576363fc4
1 changed files with 33 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files
include/nuttx/seqlock.h
+33 -4
View File
@@ -34,7 +34,11 @@
* Pre-processor Definitions
****************************************************************************/
#ifdef CONFIG_SMP
#define SEQLOCK_INITIALIZER { 1u, 1u }
#else
#define SEQLOCK_INITIALIZER { 0u }
#endif
/****************************************************************************
* Public Data Types
@@ -46,8 +50,12 @@
typedef struct seqclock
{
#ifdef CONFIG_SMP
atomic_t sequence;
uint32_t next_sequence;
#else
volatile uint32_t sequence;
#endif
} seqcount_t;
#undef EXTERN
@@ -79,8 +87,12 @@ extern "C"
static inline_function void seqlock_init(FAR seqcount_t *s)
{
#ifdef CONFIG_SMP
atomic_init(&s->sequence, 1u);
s->next_sequence = 1u;
#else
s->sequence = 0u;
#endif
}
/****************************************************************************
@@ -103,6 +115,7 @@ uint32_t read_seqbegin(FAR const seqcount_t *s)
{
uint32_t seq;
#ifdef CONFIG_SMP
do
{
/* Ensure no load operation is re-ordered before the acquire load. */
@@ -110,7 +123,10 @@ uint32_t read_seqbegin(FAR const seqcount_t *s)
seq = atomic_read_acquire(&s->sequence);
}
while (seq == 0u);
#else
seq = s->sequence;
SMP_RMB();
#endif
return seq;
}
@@ -133,11 +149,19 @@ uint32_t read_seqbegin(FAR const seqcount_t *s)
static inline_function
uint32_t read_seqretry(FAR const seqcount_t *s, uint32_t start)
{
uint32_t seq;
/* Ensure all load operations before are completed. */
SMP_RMB();
return predict_false(atomic_read(&s->sequence) != start);
#ifdef CONFIG_SMP
seq = atomic_read(&s->sequence);
#else
seq = s->sequence;
#endif
return predict_false(seq != start);
}
/****************************************************************************
@@ -158,12 +182,12 @@ uint32_t read_seqretry(FAR const seqcount_t *s, uint32_t start)
static inline_function
irqstate_t write_seqlock_irqsave(FAR seqcount_t *s)
{
uint32_t sequence;
irqstate_t flags = up_irq_save();
#ifdef CONFIG_SMP
for (; ; )
{
sequence = atomic_read(&s->sequence);
uint32_t sequence = atomic_read(&s->sequence);
if (predict_true(sequence != 0u))
{
@@ -177,6 +201,9 @@ irqstate_t write_seqlock_irqsave(FAR seqcount_t *s)
/* CPU Relax and retry. */
}
#else
s->sequence++;
#endif
return flags;
}
@@ -200,9 +227,11 @@ irqstate_t write_seqlock_irqsave(FAR seqcount_t *s)
static inline_function
void write_sequnlock_irqrestore(seqcount_t *s, irqstate_t flags)
{
#ifdef CONFIG_SMP
uint32_t next = s->next_sequence + 2;
s->next_sequence = next;
atomic_set_release(&s->sequence, next);
#endif
up_irq_restore(flags);
}