Re: [dpdk-dev] [PATCH 1/2] eal:Introduce rte_dma_wmb/rte_dma_rmb.

[Wang Dong <dong.wang.pro@hotmail.com>]

Hi Konstantin,
Yes, it's better to use rte_rmb/rte_wmb instead of their source code.
Actually, the warning is a bug, it lose a "}" of rte_dma_rmb(), I didn't 
test it on PowerPC, so the bug didn't be found out before I submit the 
patch, sorry.

I'll resubmit this patch.

Dong

> Hi Dong,
>
>> -----Original Message-----
>> From: dev [mailto:dev-bounces@dpdk.org] On Behalf Of WangDong
>> Sent: Sunday, June 28, 2015 4:23 PM
>> To: dev@dpdk.org
>> Subject: [dpdk-dev] [PATCH 1/2] eal:Introduce rte_dma_wmb/rte_dma_rmb.
>>
>> These macro can be used to replace current PMD's compiler memory barrier (volatile varible) and rte_wmb.
>> In x86, they implement to compiler memory barrier.
>> In power, they implement to processor memory barrier.
>>
>> ---
>>   .../common/include/arch/ppc_64/rte_atomic.h        |  4 ++++
>>   .../common/include/arch/x86/rte_atomic.h           |  4 ++++
>>   lib/librte_eal/common/include/generic/rte_atomic.h | 25 ++++++++++++++++++++++
>>   3 files changed, 33 insertions(+)
>>
>> diff --git a/lib/librte_eal/common/include/arch/ppc_64/rte_atomic.h b/lib/librte_eal/common/include/arch/ppc_64/rte_atomic.h
>> index fb7af2b..8f4129d 100644
>> --- a/lib/librte_eal/common/include/arch/ppc_64/rte_atomic.h
>> +++ b/lib/librte_eal/common/include/arch/ppc_64/rte_atomic.h
>> @@ -72,6 +72,10 @@ extern "C" {
>>    */
>>   #define	rte_rmb() {asm volatile("sync" : : : "memory"); }
>>
>> +#define rte_dma_wmb() {asm volatile("sync" : : : "memory"); }
>> +
>> +#define rte_dma_rmb() {asm volatile("sync" : : : "memory");
>
> As a nit, probably better:
>
> +#define rte_dma_wmb()  rte_rmb()
> +#define rte_dma_rmb()    rte_wmb()
>
> Here?
>
> Konstantin
>
>> +
>>   /*------------------------- 16 bit atomic operations -------------------------*/
>>   /* To be compatible with Power7, use GCC built-in functions for 16 bit
>>    * operations */
>> diff --git a/lib/librte_eal/common/include/arch/x86/rte_atomic.h b/lib/librte_eal/common/include/arch/x86/rte_atomic.h
>> index e93e8ee..7cfbe8f 100644
>> --- a/lib/librte_eal/common/include/arch/x86/rte_atomic.h
>> +++ b/lib/librte_eal/common/include/arch/x86/rte_atomic.h
>> @@ -53,6 +53,10 @@ extern "C" {
>>
>>   #define	rte_rmb() _mm_lfence()
>>
>> +#define rte_dma_wmb() rte_compiler_barrier()
>> +
>> +#define rte_dma_rmb() rte_compiler_barrier()
>> +
>>   /*------------------------- 16 bit atomic operations -------------------------*/
>>
>>   #ifndef RTE_FORCE_INTRINSICS
>> diff --git a/lib/librte_eal/common/include/generic/rte_atomic.h b/lib/librte_eal/common/include/generic/rte_atomic.h
>> index 6c7581a..a51eeee 100644
>> --- a/lib/librte_eal/common/include/generic/rte_atomic.h
>> +++ b/lib/librte_eal/common/include/generic/rte_atomic.h
>> @@ -72,6 +72,31 @@ static inline void rte_wmb(void);
>>    */
>>   static inline void rte_rmb(void);
>>
>> +/**
>> + * Write memory barrier for DMA.
>> + *
>> + * Be used in PMD, unlike rte_wmb() which use processor memory barrier,
>> + * this memory barrier focus on performance, if compiler memory barrier
>> + * is sufficient for guarantee memory ordering, this function will
>> + * use compiler memory barrier.
>> + *
>> + * This function is architecture dependent.
>> + */
>> +static inline void rte_dma_wmb(void);
>> +
>> +/**
>> + * Read memory barrier for DMA.
>> + *
>> + * Be used in PMD, unlike rte_rmb() which use processor memory barrier,
>> + * this memory barrier focus on performance, if compiler memory barrier
>> + * is sufficient for guarantee memory ordering, this function will
>> + * use compiler memory barrier.
>> + *
>> + * This function is architecture dependent.
>> + */
>> +static inline void rte_dma_rmb(void);
>> +
>> +
>>   #endif /* __DOXYGEN__ */
>>
>>   /**
>> --
>> 2.1.0
>