Re: [dpdk-dev] 回复: [dpdk-stable] [PATCH v3] crypto/qat: fix uninitilized compiler warning

[Ferruh Yigit <ferruh.yigit@intel.com>]

On 5/20/2021 6:44 AM, Feifei Wang wrote:
> Hi, Ferruh
> 
> Thanks for your comments.
> Please see below.
> 
>> -----邮件原件-----
>> 发件人: Ferruh Yigit <ferruh.yigit@intel.com>
>> 发送时间: 2021年5月19日 16:12
>> 收件人: Feifei Wang <Feifei.Wang2@arm.com>; John Griffin
>> <john.griffin@intel.com>; Fiona Trahe <fiona.trahe@intel.com>; Deepak
>> Kumar Jain <deepak.k.jain@intel.com>; Jerin Jacob
>> <jerin.jacob@caviumnetworks.com>; Herbert Guan
>> <Herbert.Guan@arm.com>
>> 抄送: dev@dpdk.org; david.marchand@redhat.com; nd <nd@arm.com>;
>> stable@dpdk.org; Ruifeng Wang <Ruifeng.Wang@arm.com>
>> 主题: Re: [dpdk-stable] [PATCH v3] crypto/qat: fix uninitilized compiler
>> warning
>>
>> On 5/17/2021 10:07 AM, Feifei Wang wrote:
>>> In Arm platform, when "RTE_ARCH_ARM64_MEMCPY" is set as true,
>> compiler
>>> will report variable uninitilized warning:
>>>
>>> ../drivers/crypto/qat/qat_sym_session.c:
>>> In function ‘partial_hash_compute’:
>>> ../lib/eal/include/generic/rte_byteorder.h:241:24: warning:
>>> ‘<U35a0>’ may be used uninitialized in this function
>>> 	[-Wmaybe-uninitialized]
>>> 	241 | #define rte_bswap32(x) __builtin_bswap32(x)
>>> 	...
>>>
>>> This is because "digest" will be initialized by "rte_memcpy" function
>>> rather than "memcpy" if "RTE_ARCH_ARM64_MEMCPY" is set as true.
>>> However,
>>
>> How 'digest' is initialized by 'rte_memcpy'?
> 
> Firstly, 'digest' is initialized by rte_memcpy in partial_hash_sha_x function :
> 'partial_hash_compute' -> 'partial_hash_sha_x' -> 'rte_memcpy'.
> 
> If "RTE_ARCH_ARM64_MEMCPY = false", rte_memcpy will be defined as
> 'memcpy' to initialize 'digest' in  lib\eal\arm\include\rte_memcpy_64.h: 364, 
> and compiler can identify this.
> 
> However, if "RTE_ARCH_ARM64_MEMCPY = true", rte_memcpy will be a inline 
> function, and finally it will initialize 'digest' with two steps by invoking rte_mov16:
> rte_memcpy -> rte_memcpy_ge16_lt_128 ->
> step 1: rte_mov16(dst,src )
>  step 2: rte_mov16(dst - 16 + n, src - 16 + n)
> And the compiler cannot identify this multi-step initialization, then it will report warning.
>  

OK, I got what you mean, thanks for clarification.

>>
>>> compiler cannot know it is initialized by the function.
>>>
>>> To fix this, use "calloc" to initialize "digest".
>>>
>>> Fixes: cd7fc8a84b48 ("eal/arm64: optimize memcpy")
>>> Cc: stable@dpdk.org
>>>
>>> Signed-off-by: Feifei Wang <feifei.wang2@arm.com>
>>> Reviewed-by: Ruifeng Wang <ruifeng.wang@arm.com>
>>> ---
>>> v2: add check and free for memory dynamic allocation (David Marchand)
>>> v3: fix compiler error
>>>
>>>  drivers/crypto/qat/qat_sym_session.c | 27 ++++++++++++++++++---------
>>>  1 file changed, 18 insertions(+), 9 deletions(-)
>>>
>>> diff --git a/drivers/crypto/qat/qat_sym_session.c
>>> b/drivers/crypto/qat/qat_sym_session.c
>>> index 231b1640da..105a10957a 100644
>>> --- a/drivers/crypto/qat/qat_sym_session.c
>>> +++ b/drivers/crypto/qat/qat_sym_session.c
>>> @@ -1190,8 +1190,7 @@ static int partial_hash_compute(enum
>> icp_qat_hw_auth_algo hash_alg,
>>>  			uint8_t *data_out)
>>>  {
>>>  	int digest_size;
>>> -	uint8_t digest[qat_hash_get_digest_size(
>>> -			ICP_QAT_HW_AUTH_ALGO_DELIMITER)];
>>> +	uint8_t *digest;
>>
>> Will a memset 'digest' work too? Although not sure which one is better.
> Thanks for your meaningful comments, I try to use memset and it is ok to solve this warning.
> I will update this in the next version. 
>