Re: [PATCH v2 2/6] crypto: arm64/sm3-ce - make dependent on sm3 library

[Tianjia Zhang]

Hi

On 12/31/21 3:05 PM, liulongfang wrote:
> On 2021/12/22 12:50, Tianjia Zhang Wrote:
> > SM3 generic library is stand-alone implementation, sm3-ce can depend
> > on the SM3 library instead of sm3-generic.
> >
> > Signed-off-by: Tianjia Zhang <tianjia.zhang@xxxxxxxxxxxxxxxxx>
> > ---
> >   arch/arm64/crypto/Kconfig       |  2 +-
> >   arch/arm64/crypto/sm3-ce-glue.c | 20 ++++++++++++++------
> >   2 files changed, 15 insertions(+), 7 deletions(-)
> >
> > diff --git a/arch/arm64/crypto/Kconfig b/arch/arm64/crypto/Kconfig
> > index addfa413650b..2a965aa0188d 100644
> > --- a/arch/arm64/crypto/Kconfig
> > +++ b/arch/arm64/crypto/Kconfig
> > @@ -45,7 +45,7 @@ config CRYPTO_SM3_ARM64_CE
> >   	tristate "SM3 digest algorithm (ARMv8.2 Crypto Extensions)"
> >   	depends on KERNEL_MODE_NEON
> >   	select CRYPTO_HASH
> > -	select CRYPTO_SM3
> > +	select CRYPTO_LIB_SM3
> >   
> >   config CRYPTO_SM4_ARM64_CE
> >   	tristate "SM4 symmetric cipher (ARMv8.2 Crypto Extensions)"
> > diff --git a/arch/arm64/crypto/sm3-ce-glue.c b/arch/arm64/crypto/sm3-ce-glue.c
> > index d71faca322f2..3198f31c9446 100644
> > --- a/arch/arm64/crypto/sm3-ce-glue.c
> > +++ b/arch/arm64/crypto/sm3-ce-glue.c
> > @@ -27,7 +27,7 @@ static int sm3_ce_update(struct shash_desc *desc, const u8 *data,
> >   			 unsigned int len)
> >   {
> >   	if (!crypto_simd_usable())
> > -		return crypto_sm3_update(desc, data, len);
> > +		return sm3_update(shash_desc_ctx(desc), data, len);
> >   
> >   	kernel_neon_begin();
> >   	sm3_base_do_update(desc, data, len, sm3_ce_transform);
> > @@ -39,7 +39,7 @@ static int sm3_ce_update(struct shash_desc *desc, const u8 *data,
> >   static int sm3_ce_final(struct shash_desc *desc, u8 *out)
> >   {
> >   	if (!crypto_simd_usable())
> > -		return crypto_sm3_finup(desc, NULL, 0, out);
> > +		return sm3_final(shash_desc_ctx(desc), out);
> >   
> >   	kernel_neon_begin();
> >   	sm3_base_do_finalize(desc, sm3_ce_transform);
> > @@ -51,14 +51,22 @@ static int sm3_ce_final(struct shash_desc *desc, u8 *out)
> >   static int sm3_ce_finup(struct shash_desc *desc, const u8 *data,
> >   			unsigned int len, u8 *out)
> >   {
> > -	if (!crypto_simd_usable())
> > -		return crypto_sm3_finup(desc, data, len, out);
> > +	if (!crypto_simd_usable()) {
> > +		struct sm3_state *sctx = shash_desc_ctx(desc);
> > +
> > +		if (len)
> > +			sm3_update(sctx, data, len);
> > +		sm3_final(sctx, out);
> > +		return 0;
> > +	}
> >   
> >   	kernel_neon_begin();
> > -	sm3_base_do_update(desc, data, len, sm3_ce_transform);
> > +	if (len)
> > +		sm3_base_do_update(desc, data, len, sm3_ce_transform);
> > +	sm3_base_do_finalize(desc, sm3_ce_transform);
> >   	kernel_neon_end();
> >   
> > -	return sm3_ce_final(desc, out);
> > +	return sm3_base_finish(desc, out);
> >   }
> >   
> >   static struct shash_alg sm3_alg = {
> > You have modified the implementation of SM3 algorithm, so what benefits will be gained
> after such modification?
> What flaws are solved or can performance be improved?
> Thanks.
> Longfang.

This modification does not bring obvious performance improvement, but
makes the code logic more reasonable in terms of architecture and
calling level. The calling relationship before modification is:
  sm3-ce -> sm3-generic -> sm3-lib,
after this modification is: sm3-ce -> sm3-lib.

Best regards,
Tianjia