zephyr/include/spinlock.h
Andy Ross f37e0c6e4d kernel/spinlock: Fix race in spinlock validation
The k_spin_lock() validation was setting the new owner of the spinlock
BEFORE the actual lock was taken, so it could race against other
processors trying the same thing.  Split the modification step out
into a separate function that can be called after we affirmatively
have the lock.

Signed-off-by: Andy Ross <andrew.j.ross@intel.com>
2019-03-13 19:15:20 +01:00

128 lines
3.0 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>
/* These stubs aren't provided by the mocking framework, and I can't
* find a proper place to put them as mocking seems not to have a
* proper "arch" layer.
*/
#ifdef ZTEST_UNITTEST
static inline int z_arch_irq_lock(void)
{
return 0;
}
static inline void z_arch_irq_unlock(int key)
{
ARG_UNUSED(key);
}
#endif
/* There's a spinlock validation framework available when asserts are
* enabled. It adds a relatively hefty overhead (about 3k or so) to
* kernel code size, don't use on platforms known to be small. (Note
* we're using the kconfig value here. This isn't defined for every
* board, but the default of zero works well as an "infinity"
* fallback. There is a DT_FLASH_SIZE parameter too, but that seems
* even more poorly supported.
*/
#if (CONFIG_FLASH_SIZE == 0) || (CONFIG_FLASH_SIZE > 32)
#if defined(CONFIG_ASSERT) && (CONFIG_MP_NUM_CPUS < 4)
#include <misc/__assert.h>
struct k_spinlock;
int z_spin_lock_valid(struct k_spinlock *l);
int z_spin_unlock_valid(struct k_spinlock *l);
void z_spin_lock_set_owner(struct k_spinlock *l);
#define SPIN_VALIDATE
#endif
#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)
{
ARG_UNUSED(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 = z_arch_irq_lock();
#ifdef SPIN_VALIDATE
__ASSERT(z_spin_lock_valid(l), "Recursive spinlock");
#endif
#ifdef CONFIG_SMP
while (!atomic_cas(&l->locked, 0, 1)) {
}
#endif
#ifdef SPIN_VALIDATE
z_spin_lock_set_owner(l);
#endif
return k;
}
static ALWAYS_INLINE void k_spin_unlock(struct k_spinlock *l,
k_spinlock_key_t key)
{
ARG_UNUSED(l);
#ifdef SPIN_VALIDATE
__ASSERT(z_spin_unlock_valid(l), "Not my spinlock!");
#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
z_arch_irq_unlock(key.key);
}
/* Internal function: releases the lock, but leaves local interrupts
* disabled
*/
static ALWAYS_INLINE void k_spin_release(struct k_spinlock *l)
{
ARG_UNUSED(l);
#ifdef SPIN_VALIDATE
__ASSERT(z_spin_unlock_valid(l), "Not my spinlock!");
#endif
#ifdef CONFIG_SMP
atomic_clear(&l->locked);
#endif
}
#endif /* ZEPHYR_INCLUDE_SPINLOCK_H_ */