4ff2dfce09
Something is going wrong with code generation here, potentially the inline assembly generated by _arch_irq_un/lock(), and these calls are not being inlined by gcc. So what should be a ~3 instruction sequence on most uniprocessor architectures is turning into 8-20 cycles worth of work to implement the API as written. Use an ALWAYS_INLINE, which is sort of ugly semantically but produces much better code. Signed-off-by: Andy Ross <andrew.j.ross@intel.com>
85 lines
1.8 KiB
C
85 lines
1.8 KiB
C
/*
|
|
* Copyright (c) 2018 Intel Corporation.
|
|
*
|
|
* SPDX-License-Identifier: Apache-2.0
|
|
*/
|
|
#ifndef ZEPHYR_INCLUDE_SPINLOCK_H_
|
|
#define ZEPHYR_INCLUDE_SPINLOCK_H_
|
|
|
|
#include <atomic.h>
|
|
|
|
#if defined(CONFIG_ASSERT) && (CONFIG_MP_NUM_CPUS < 4)
|
|
#include <kernel_structs.h>
|
|
#define SPIN_VALIDATE
|
|
#endif
|
|
|
|
struct k_spinlock_key {
|
|
int key;
|
|
};
|
|
|
|
typedef struct k_spinlock_key k_spinlock_key_t;
|
|
|
|
struct k_spinlock {
|
|
#ifdef CONFIG_SMP
|
|
atomic_t locked;
|
|
#endif
|
|
|
|
#ifdef SPIN_VALIDATE
|
|
/* Stores the thread that holds the lock with the locking CPU
|
|
* ID in the bottom two bits.
|
|
*/
|
|
size_t thread_cpu;
|
|
#endif
|
|
};
|
|
|
|
static ALWAYS_INLINE k_spinlock_key_t k_spin_lock(struct k_spinlock *l)
|
|
{
|
|
k_spinlock_key_t k;
|
|
|
|
/* Note that we need to use the underlying arch-specific lock
|
|
* implementation. The "irq_lock()" API in SMP context is
|
|
* actually a wrapper for a global spinlock!
|
|
*/
|
|
k.key = _arch_irq_lock();
|
|
|
|
#ifdef SPIN_VALIDATE
|
|
if (l->thread_cpu) {
|
|
__ASSERT((l->thread_cpu & 3) != _current_cpu->id,
|
|
"Recursive spinlock");
|
|
}
|
|
l->thread_cpu = _current_cpu->id | (u32_t)_current;
|
|
#endif
|
|
|
|
#ifdef CONFIG_SMP
|
|
while (!atomic_cas(&l->locked, 0, 1)) {
|
|
}
|
|
#endif
|
|
|
|
return k;
|
|
}
|
|
|
|
static ALWAYS_INLINE void k_spin_unlock(struct k_spinlock *l,
|
|
k_spinlock_key_t key)
|
|
{
|
|
#ifdef SPIN_VALIDATE
|
|
__ASSERT(l->thread_cpu == (_current_cpu->id | (u32_t)_current),
|
|
"Not my spinlock!");
|
|
l->thread_cpu = 0;
|
|
#endif
|
|
|
|
|
|
#ifdef CONFIG_SMP
|
|
/* Strictly we don't need atomic_clear() here (which is an
|
|
* exchange operation that returns the old value). We are always
|
|
* setting a zero and (because we hold the lock) know the existing
|
|
* state won't change due to a race. But some architectures need
|
|
* a memory barrier when used like this, and we don't have a
|
|
* Zephyr framework for that.
|
|
*/
|
|
atomic_clear(&l->locked);
|
|
#endif
|
|
_arch_irq_unlock(key.key);
|
|
}
|
|
|
|
#endif /* ZEPHYR_INCLUDE_SPINLOCK_H_ */
|