[dpdk-dev] [PATCH v5 01/15] eal/arm: atomic operations for ARM

Jerin Jacob jerin.jacob at caviumnetworks.com
Mon Nov 2 06:53:05 CET 2015


On Fri, Oct 30, 2015 at 01:25:28AM +0100, Jan Viktorin wrote:
> From: Vlastimil Kosar <kosar at rehivetech.com>
>
> This patch adds architecture specific atomic operation file
> for ARM architecture. It utilizes compiler intrinsics only.
>
> Signed-off-by: Vlastimil Kosar <kosar at rehivetech.com>
> Signed-off-by: Jan Viktorin <viktorin at rehivetech.com>
> ---
> v1 -> v2:
> * improve rte_wmb()
> * use __atomic_* or __sync_*? (may affect the required GCC version)
>
> v4:
> * checkpatch complaints about volatile keyword (but seems to be OK to me)
> * checkpatch complaints about do { ... } while (0) for single statement
>   with asm volatile (but I didn't find a way how to write it without
>   the checkpatch complaints)
> * checkpatch is now happy with whitespaces
> ---
>  .../common/include/arch/arm/rte_atomic.h           | 256 +++++++++++++++++++++
>  1 file changed, 256 insertions(+)
>  create mode 100644 lib/librte_eal/common/include/arch/arm/rte_atomic.h
>
> diff --git a/lib/librte_eal/common/include/arch/arm/rte_atomic.h b/lib/librte_eal/common/include/arch/arm/rte_atomic.h
> new file mode 100644
> index 0000000..ea1e485
> --- /dev/null
> +++ b/lib/librte_eal/common/include/arch/arm/rte_atomic.h
> @@ -0,0 +1,256 @@
> +/*-
> + *   BSD LICENSE
> + *
> + *   Copyright(c) 2015 RehiveTech. All rights reserved.
> + *
> + *   Redistribution and use in source and binary forms, with or without
> + *   modification, are permitted provided that the following conditions
> + *   are met:
> + *
> + *     * Redistributions of source code must retain the above copyright
> + *       notice, this list of conditions and the following disclaimer.
> + *     * Redistributions in binary form must reproduce the above copyright
> + *       notice, this list of conditions and the following disclaimer in
> + *       the documentation and/or other materials provided with the
> + *       distribution.
> + *     * Neither the name of RehiveTech nor the names of its
> + *       contributors may be used to endorse or promote products derived
> + *       from this software without specific prior written permission.
> + *
> + *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
> + *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
> + *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
> + *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
> + *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
> + *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
> + *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
> + *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
> + *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
> + *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
> + *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
> + */
> +
> +#ifndef _RTE_ATOMIC_ARM_H_
> +#define _RTE_ATOMIC_ARM_H_
> +
> +#ifdef __cplusplus
> +extern "C" {
> +#endif
> +
> +#include "generic/rte_atomic.h"
> +
> +/**
> + * General memory barrier.
> + *
> + * Guarantees that the LOAD and STORE operations generated before the
> + * barrier occur before the LOAD and STORE operations generated after.
> + */
> +#define	rte_mb()  __sync_synchronize()
> +
> +/**
> + * Write memory barrier.
> + *
> + * Guarantees that the STORE operations generated before the barrier
> + * occur before the STORE operations generated after.
> + */
> +#define	rte_wmb() do { asm volatile ("dmb st" : : : "memory"); } while (0)
> +
> +/**
> + * Read memory barrier.
> + *
> + * Guarantees that the LOAD operations generated before the barrier
> + * occur before the LOAD operations generated after.
> + */
> +#define	rte_rmb() __sync_synchronize()
> +

#define dmb(opt)        asm volatile("dmb " #opt : : : "memory")

static inline void rte_mb(void)
{
        dmb(ish);
}

static inline void rte_wmb(void)
{
        dmb(ishst);
}

static inline void rte_rmb(void)
{
        dmb(ishld);
}

For armv8, it make sense to have above definition for rte_*mb(). If does
n't make sense for armv7 then we need split this file rte_atomic_32/64.h



> +/*------------------------- 16 bit atomic operations -------------------------*/
> +
> +#ifndef RTE_FORCE_INTRINSICS
> +static inline int
> +rte_atomic16_cmpset(volatile uint16_t *dst, uint16_t exp, uint16_t src)
> +{
> +	return __atomic_compare_exchange(dst, &exp, &src, 0, __ATOMIC_ACQUIRE,
> +		__ATOMIC_ACQUIRE) ? 1 : 0;
> +}

IMO, it should be __ATOMIC_SEQ_CST be instead of __ATOMIC_ACQUIRE.
__ATOMIC_ACQUIRE works in conjunction with __ATOMIC_RELEASE.
AFAIK, DPDK atomic api expects full barrier. C11 memory model not yet
used. So why can't we use RTE_FORCE_INTRINSICS based generic
implementation. Same holds true for spinlock implementation too(i.e using
RTE_FORCE_INTRINSICS). Am I missing something here ?



> +
> +static inline int rte_atomic16_test_and_set(rte_atomic16_t *v)
> +{
> +	return rte_atomic16_cmpset((volatile uint16_t *)&v->cnt, 0, 1);
> +}
> +
> +static inline void
> +rte_atomic16_inc(rte_atomic16_t *v)
> +{
> +	__atomic_add_fetch(&v->cnt, 1, __ATOMIC_ACQUIRE);
> +}
> +
> +static inline void
> +rte_atomic16_dec(rte_atomic16_t *v)
> +{
> +	__atomic_sub_fetch(&v->cnt, 1, __ATOMIC_ACQUIRE);
> +}
> +
> +static inline int rte_atomic16_inc_and_test(rte_atomic16_t *v)
> +{
> +	return (__atomic_add_fetch(&v->cnt, 1, __ATOMIC_ACQUIRE) == 0);
> +}
> +
> +static inline int rte_atomic16_dec_and_test(rte_atomic16_t *v)
> +{
> +	return (__atomic_sub_fetch(&v->cnt, 1, __ATOMIC_ACQUIRE) == 0);
> +}
> +
> +/*------------------------- 32 bit atomic operations -------------------------*/
> +
> +static inline int
> +rte_atomic32_cmpset(volatile uint32_t *dst, uint32_t exp, uint32_t src)
> +{
> +	return __atomic_compare_exchange(dst, &exp, &src, 0, __ATOMIC_ACQUIRE,
> +		__ATOMIC_ACQUIRE) ? 1 : 0;
> +}
> +
> +static inline int rte_atomic32_test_and_set(rte_atomic32_t *v)
> +{
> +	return rte_atomic32_cmpset((volatile uint32_t *)&v->cnt, 0, 1);
> +}
> +
> +static inline void
> +rte_atomic32_inc(rte_atomic32_t *v)
> +{
> +	__atomic_add_fetch(&v->cnt, 1, __ATOMIC_ACQUIRE);
> +}
> +
> +static inline void
> +rte_atomic32_dec(rte_atomic32_t *v)
> +{
> +	__atomic_sub_fetch(&v->cnt, 1, __ATOMIC_ACQUIRE);
> +}
> +
> +static inline int rte_atomic32_inc_and_test(rte_atomic32_t *v)
> +{
> +	return (__atomic_add_fetch(&v->cnt, 1, __ATOMIC_ACQUIRE) == 0);
> +}
> +
> +static inline int rte_atomic32_dec_and_test(rte_atomic32_t *v)
> +{
> +	return (__atomic_sub_fetch(&v->cnt, 1, __ATOMIC_ACQUIRE) == 0);
> +}
> +
> +/*------------------------- 64 bit atomic operations -------------------------*/
> +
> +static inline int
> +rte_atomic64_cmpset(volatile uint64_t *dst, uint64_t exp, uint64_t src)
> +{
> +	return __atomic_compare_exchange(dst, &exp, &src, 0, __ATOMIC_ACQUIRE,
> +		__ATOMIC_ACQUIRE) ? 1 : 0;
> +}
> +
> +static inline void
> +rte_atomic64_init(rte_atomic64_t *v)
> +{
> +	int success = 0;
> +	uint64_t tmp;
> +
> +	while (success == 0) {
> +		tmp = v->cnt;
> +		success = rte_atomic64_cmpset(
> +				(volatile uint64_t *)&v->cnt, tmp, 0);
> +	}
> +}
> +
> +static inline int64_t
> +rte_atomic64_read(rte_atomic64_t *v)
> +{
> +	int success = 0;
> +	uint64_t tmp;
> +
> +	while (success == 0) {
> +		tmp = v->cnt;
> +		/* replace the value by itself */
> +		success = rte_atomic64_cmpset(
> +				(volatile uint64_t *) &v->cnt, tmp, tmp);
> +	}
> +	return tmp;
> +}

This will be overkill for arm64. Generic implementation has __LP64__
based check for 64bit platform


> +
> +static inline void
> +rte_atomic64_set(rte_atomic64_t *v, int64_t new_value)
> +{
> +	int success = 0;
> +	uint64_t tmp;
> +
> +	while (success == 0) {
> +		tmp = v->cnt;
> +		success = rte_atomic64_cmpset(
> +				(volatile uint64_t *)&v->cnt, tmp, new_value);
> +	}
> +}
> +
> +static inline void
> +rte_atomic64_add(rte_atomic64_t *v, int64_t inc)
> +{
> +	__atomic_fetch_add(&v->cnt, inc, __ATOMIC_ACQUIRE);
> +}
> +
> +static inline void
> +rte_atomic64_sub(rte_atomic64_t *v, int64_t dec)
> +{
> +	__atomic_fetch_sub(&v->cnt, dec, __ATOMIC_ACQUIRE);
> +}
> +

__atomic_fetch_* operations on 64bit works only when compiler support
(__GCC_ATOMIC_LLONG_LOCK_FREE >=2).

if DPDK API's expects full barrier not the C11 memory model based
__ATOMIC_ACQUIRE then better to use generic implementation.

> +static inline void
> +rte_atomic64_inc(rte_atomic64_t *v)
> +{
> +	__atomic_fetch_add(&v->cnt, 1, __ATOMIC_ACQUIRE);
> +}
> +
> +static inline void
> +rte_atomic64_dec(rte_atomic64_t *v)
> +{
> +	__atomic_fetch_sub(&v->cnt, 1, __ATOMIC_ACQUIRE);
> +}
> +
> +static inline int64_t
> +rte_atomic64_add_return(rte_atomic64_t *v, int64_t inc)
> +{
> +	return __atomic_add_fetch(&v->cnt, inc, __ATOMIC_ACQUIRE);
> +}
> +
> +static inline int64_t
> +rte_atomic64_sub_return(rte_atomic64_t *v, int64_t dec)
> +{
> +	return __atomic_sub_fetch(&v->cnt, dec, __ATOMIC_ACQUIRE);
> +}
> +
> +static inline int rte_atomic64_inc_and_test(rte_atomic64_t *v)
> +{
> +	return (__atomic_add_fetch(&v->cnt, 1, __ATOMIC_ACQUIRE) == 0);
> +}
> +
> +static inline int rte_atomic64_dec_and_test(rte_atomic64_t *v)
> +{
> +	return (__atomic_sub_fetch(&v->cnt, 1, __ATOMIC_ACQUIRE) == 0);
> +}
> +
> +static inline int rte_atomic64_test_and_set(rte_atomic64_t *v)
> +{
> +	return rte_atomic64_cmpset((volatile uint64_t *)&v->cnt, 0, 1);
> +}
> +
> +/**
> + * Atomically set a 64-bit counter to 0.
> + *
> + * @param v
> + *   A pointer to the atomic counter.
> + */
> +static inline void rte_atomic64_clear(rte_atomic64_t *v)
> +{
> +	rte_atomic64_set(v, 0);
> +}
> +#endif
> +
> +#ifdef __cplusplus
> +}
> +#endif
> +
> +#endif /* _RTE_ATOMIC_ARM_H_ */
> --
> 2.6.1
>


More information about the dev mailing list