On 2021/2/18 10:06, Herbert Xu wrote: > On Thu, Feb 18, 2021 at 10:01:58AM +0800, liulongfang wrote: >> >>>> diff --git a/drivers/crypto/hisilicon/sec2/sec_crypto.c b/drivers/crypto/hisilicon/sec2/sec_crypto.c >>>> index d2c4a2c..988faf7 100644 >>>> --- a/drivers/crypto/hisilicon/sec2/sec_crypto.c >>>> +++ b/drivers/crypto/hisilicon/sec2/sec_crypto.c >>>> @@ -7,6 +7,7 @@ >>>> #include <crypto/des.h> >>>> #include <crypto/hash.h> >>>> #include <crypto/internal/aead.h> >>>> +#include <crypto/internal/hash.h> >>> >>> Please explain what exactly in this file needs this header file. >>> >>> As it stands you could just be hiding real bugs. >>> >>> Thanks, >>> >> The crypto_alloc_shash() interface in the header file >> will be used in the function sec_aead_ctx_init(), >> If this header file is not added, calling the interface >> crypto_alloc_shash() during the initialization of the >> aead algorithm will return an error. > > This makes no sense whatsoever as crypto_alloc_shash is defiend > by crypto/hash.h and you've already included that. > > Cheers, > On this kernel version, those modules set to Y will not use the sha512 algorithm. And our SEC module selects it "select CRYPTO_SHA512", so it is compiled into ko: "sha512_generic.ko". Because we did not load the "sha512_generic.ko" when we loaded the ko of SEC, the sha512 algorithm test failed, Therefore, before using SEC, we need to load this ko first, so this patch is not required, please ignore it. Thanks Longfang.
