Re: [PATCH 2/5] crypto: crypto_user_stat: convert all stats from u32 to u64

[Eric Biggers]

Hi Corentin,

On Mon, Nov 05, 2018 at 12:51:11PM +0000, Corentin Labbe wrote:
> All the 32-bit fields need to be 64-bit.  In some cases, UINT32_MAX crypto
> operations can be done in seconds.
> 
> Reported-by: Eric Biggers <ebiggers@xxxxxxxxxx>
> Signed-off-by: Corentin Labbe <clabbe@xxxxxxxxxxxx>
> ---
>  crypto/algapi.c                 |  10 +--
>  crypto/crypto_user_stat.c       | 114 +++++++++++++++-----------------
>  include/crypto/acompress.h      |   8 +--
>  include/crypto/aead.h           |   8 +--
>  include/crypto/akcipher.h       |  16 ++---
>  include/crypto/hash.h           |   6 +-
>  include/crypto/kpp.h            |  12 ++--
>  include/crypto/rng.h            |   8 +--
>  include/crypto/skcipher.h       |   8 +--
>  include/linux/crypto.h          |  46 ++++++-------
>  include/uapi/linux/cryptouser.h |  38 +++++------
>  11 files changed, 133 insertions(+), 141 deletions(-)
> 
> diff --git a/crypto/algapi.c b/crypto/algapi.c
> index f5396c88e8cd..42fe316f80ee 100644
> --- a/crypto/algapi.c
> +++ b/crypto/algapi.c
> @@ -259,13 +259,13 @@ static struct crypto_larval *__crypto_register_alg(struct crypto_alg *alg)
>  	list_add(&larval->alg.cra_list, &crypto_alg_list);
>  
>  #ifdef CONFIG_CRYPTO_STATS
> -	atomic_set(&alg->encrypt_cnt, 0);
> -	atomic_set(&alg->decrypt_cnt, 0);
> +	atomic64_set(&alg->encrypt_cnt, 0);
> +	atomic64_set(&alg->decrypt_cnt, 0);
>  	atomic64_set(&alg->encrypt_tlen, 0);
>  	atomic64_set(&alg->decrypt_tlen, 0);
> -	atomic_set(&alg->verify_cnt, 0);
> -	atomic_set(&alg->cipher_err_cnt, 0);
> -	atomic_set(&alg->sign_cnt, 0);
> +	atomic64_set(&alg->verify_cnt, 0);
> +	atomic64_set(&alg->cipher_err_cnt, 0);
> +	atomic64_set(&alg->sign_cnt, 0);
>  #endif
>  
>  out:
> diff --git a/crypto/crypto_user_stat.c b/crypto/crypto_user_stat.c
> index a6fb2e6f618d..352569f378a0 100644
> --- a/crypto/crypto_user_stat.c
> +++ b/crypto/crypto_user_stat.c
> @@ -35,22 +35,21 @@ static int crypto_report_aead(struct sk_buff *skb, struct crypto_alg *alg)
>  {
>  	struct crypto_stat raead;
>  	u64 v64;
> -	u32 v32;
>  
>  	memset(&raead, 0, sizeof(raead));
>  
>  	strscpy(raead.type, "aead", sizeof(raead.type));
>  
> -	v32 = atomic_read(&alg->encrypt_cnt);
> -	raead.stat_encrypt_cnt = v32;
> +	v64 = atomic64_read(&alg->encrypt_cnt);
> +	raead.stat_encrypt_cnt = v64;
>  	v64 = atomic64_read(&alg->encrypt_tlen);
>  	raead.stat_encrypt_tlen = v64;
> -	v32 = atomic_read(&alg->decrypt_cnt);
> -	raead.stat_decrypt_cnt = v32;
> +	v64 = atomic64_read(&alg->decrypt_cnt);
> +	raead.stat_decrypt_cnt = v64;
>  	v64 = atomic64_read(&alg->decrypt_tlen);
>  	raead.stat_decrypt_tlen = v64;
> -	v32 = atomic_read(&alg->aead_err_cnt);
> -	raead.stat_aead_err_cnt = v32;
> +	v64 = atomic64_read(&alg->aead_err_cnt);
> +	raead.stat_aead_err_cnt = v64;
>  
>  	return nla_put(skb, CRYPTOCFGA_STAT_AEAD, sizeof(raead), &raead);
>  }

Why not assign the result of atomic64_read() directly?
I don't see the point of the 'v64' variable.

- Eric