Re: [dpdk-dev] [PATCH v2] mbuf: use C11 atomics for refcnt operations

[Olivier Matz <olivier.matz@6wind.com>]

Hi,

On Tue, Jul 07, 2020 at 06:10:33PM +0800, Phil Yang wrote:
> Use C11 atomics with explicit ordering instead of rte_atomic ops which
> enforce unnecessary barriers on aarch64.
> 
> Signed-off-by: Phil Yang <phil.yang@arm.com>
> Reviewed-by: Ruifeng Wang <ruifeng.wang@arm.com>
> ---
> v2:
> Fix ABI issue: revert the rte_mbuf_ext_shared_info struct refcnt field
> to refcnt_atomic.
> 
>  lib/librte_mbuf/rte_mbuf.c      |  1 -
>  lib/librte_mbuf/rte_mbuf.h      | 19 ++++++++++---------
>  lib/librte_mbuf/rte_mbuf_core.h | 11 +++--------
>  3 files changed, 13 insertions(+), 18 deletions(-)
> 
> diff --git a/lib/librte_mbuf/rte_mbuf.c b/lib/librte_mbuf/rte_mbuf.c
> index ae91ae2..8a456e5 100644
> --- a/lib/librte_mbuf/rte_mbuf.c
> +++ b/lib/librte_mbuf/rte_mbuf.c
> @@ -22,7 +22,6 @@
>  #include <rte_eal.h>
>  #include <rte_per_lcore.h>
>  #include <rte_lcore.h>
> -#include <rte_atomic.h>
>  #include <rte_branch_prediction.h>
>  #include <rte_mempool.h>
>  #include <rte_mbuf.h>
> diff --git a/lib/librte_mbuf/rte_mbuf.h b/lib/librte_mbuf/rte_mbuf.h
> index f8e492e..4a7a98c 100644
> --- a/lib/librte_mbuf/rte_mbuf.h
> +++ b/lib/librte_mbuf/rte_mbuf.h
> @@ -37,7 +37,6 @@
>  #include <rte_config.h>
>  #include <rte_mempool.h>
>  #include <rte_memory.h>
> -#include <rte_atomic.h>
>  #include <rte_prefetch.h>
>  #include <rte_branch_prediction.h>
>  #include <rte_byteorder.h>
> @@ -365,7 +364,7 @@ rte_pktmbuf_priv_flags(struct rte_mempool *mp)
>  static inline uint16_t
>  rte_mbuf_refcnt_read(const struct rte_mbuf *m)
>  {
> -	return (uint16_t)(rte_atomic16_read(&m->refcnt_atomic));
> +	return __atomic_load_n(&m->refcnt, __ATOMIC_RELAXED);
>  }
>  
>  /**
> @@ -378,14 +377,15 @@ rte_mbuf_refcnt_read(const struct rte_mbuf *m)
>  static inline void
>  rte_mbuf_refcnt_set(struct rte_mbuf *m, uint16_t new_value)
>  {
> -	rte_atomic16_set(&m->refcnt_atomic, (int16_t)new_value);
> +	__atomic_store_n(&m->refcnt, new_value, __ATOMIC_RELAXED);
>  }
>  
>  /* internal */
>  static inline uint16_t
>  __rte_mbuf_refcnt_update(struct rte_mbuf *m, int16_t value)
>  {
> -	return (uint16_t)(rte_atomic16_add_return(&m->refcnt_atomic, value));
> +	return (uint16_t)(__atomic_add_fetch((int16_t *)&m->refcnt, value,
> +					__ATOMIC_ACQ_REL));
>  }
>  
>  /**
> @@ -466,7 +466,7 @@ rte_mbuf_refcnt_set(struct rte_mbuf *m, uint16_t new_value)
>  static inline uint16_t
>  rte_mbuf_ext_refcnt_read(const struct rte_mbuf_ext_shared_info *shinfo)
>  {
> -	return (uint16_t)(rte_atomic16_read(&shinfo->refcnt_atomic));
> +	return __atomic_load_n(&shinfo->refcnt_atomic, __ATOMIC_RELAXED);
>  }
>  
>  /**
> @@ -481,7 +481,7 @@ static inline void
>  rte_mbuf_ext_refcnt_set(struct rte_mbuf_ext_shared_info *shinfo,
>  	uint16_t new_value)
>  {
> -	rte_atomic16_set(&shinfo->refcnt_atomic, (int16_t)new_value);
> +	__atomic_store_n(&shinfo->refcnt_atomic, new_value, __ATOMIC_RELAXED);
>  }
>  
>  /**
> @@ -505,7 +505,8 @@ rte_mbuf_ext_refcnt_update(struct rte_mbuf_ext_shared_info *shinfo,
>  		return (uint16_t)value;
>  	}
>  
> -	return (uint16_t)rte_atomic16_add_return(&shinfo->refcnt_atomic, value);
> +	return (uint16_t)(__atomic_add_fetch((int16_t *)&shinfo->refcnt_atomic,
> +					    value, __ATOMIC_ACQ_REL));
>  }
>  
>  /** Mbuf prefetch */
> @@ -1304,8 +1305,8 @@ static inline int __rte_pktmbuf_pinned_extbuf_decref(struct rte_mbuf *m)
>  	 * Direct usage of add primitive to avoid
>  	 * duplication of comparing with one.
>  	 */
> -	if (likely(rte_atomic16_add_return
> -			(&shinfo->refcnt_atomic, -1)))
> +	if (likely(__atomic_add_fetch((int *)&shinfo->refcnt_atomic, -1,
> +				     __ATOMIC_ACQ_REL)))
>  		return 1;
>  
>  	/* Reinitialize counter before mbuf freeing. */
> diff --git a/lib/librte_mbuf/rte_mbuf_core.h b/lib/librte_mbuf/rte_mbuf_core.h
> index 16600f1..806313a 100644
> --- a/lib/librte_mbuf/rte_mbuf_core.h
> +++ b/lib/librte_mbuf/rte_mbuf_core.h
> @@ -18,7 +18,6 @@
>  
>  #include <stdint.h>
>  #include <rte_compat.h>
> -#include <generic/rte_atomic.h>
>  
>  #ifdef __cplusplus
>  extern "C" {
> @@ -495,12 +494,8 @@ struct rte_mbuf {
>  	 * or non-atomic) is controlled by the CONFIG_RTE_MBUF_REFCNT_ATOMIC
>  	 * config option.
>  	 */
> -	RTE_STD_C11
> -	union {
> -		rte_atomic16_t refcnt_atomic; /**< Atomically accessed refcnt */
> -		/** Non-atomically accessed refcnt */
> -		uint16_t refcnt;
> -	};
> +	uint16_t refcnt;
> +

It seems this patch does 2 things:
- remove refcnt_atomic
- use C11 atomics

The first change is an API break. I think it should be announced in a deprecation
notice. The one about atomic does not talk about it.

So I suggest to keep refcnt_atomic until next version.


>  	uint16_t nb_segs;         /**< Number of segments. */
>  
>  	/** Input port (16 bits to support more than 256 virtual ports).
> @@ -679,7 +674,7 @@ typedef void (*rte_mbuf_extbuf_free_callback_t)(void *addr, void *opaque);
>  struct rte_mbuf_ext_shared_info {
>  	rte_mbuf_extbuf_free_callback_t free_cb; /**< Free callback function */
>  	void *fcb_opaque;                        /**< Free callback argument */
> -	rte_atomic16_t refcnt_atomic;        /**< Atomically accessed refcnt */
> +	uint16_t refcnt_atomic;              /**< Atomically accessed refcnt */
>  };
>  
>  /**< Maximum number of nb_segs allowed. */
> -- 
> 2.7.4
>