2021-06-03 05:55:21

by Hangbin Liu

[permalink] [raw]
Subject: [PATCH] crypto: x86/curve25519 - fix cpu feature checking logic in mod_exit

In curve25519_mod_init() the curve25519_alg will be registered only when
(X86_FEATURE_BMI2 && X86_FEATURE_ADX). But in curve25519_mod_exit()
it still checks (X86_FEATURE_BMI2 || X86_FEATURE_ADX) when do crypto
unregister. This will trigger a BUG_ON in crypto_unregister_alg() as
alg->cra_refcnt is 0 if the cpu only supports one of X86_FEATURE_BMI2
and X86_FEATURE_ADX.

Fixes: 07b586fe0662 ("crypto: x86/curve25519 - replace with formally verified implementation")
Signed-off-by: Hangbin Liu <[email protected]>
---
arch/x86/crypto/curve25519-x86_64.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/arch/x86/crypto/curve25519-x86_64.c b/arch/x86/crypto/curve25519-x86_64.c
index 6706b6cb1d0f..38caf61cd5b7 100644
--- a/arch/x86/crypto/curve25519-x86_64.c
+++ b/arch/x86/crypto/curve25519-x86_64.c
@@ -1500,7 +1500,7 @@ static int __init curve25519_mod_init(void)
static void __exit curve25519_mod_exit(void)
{
if (IS_REACHABLE(CONFIG_CRYPTO_KPP) &&
- (boot_cpu_has(X86_FEATURE_BMI2) || boot_cpu_has(X86_FEATURE_ADX)))
+ static_branch_likely(&curve25519_use_bmi2_adx))
crypto_unregister_kpp(&curve25519_alg);
}

--
2.26.3


2021-06-03 10:25:40

by Jason A. Donenfeld

[permalink] [raw]
Subject: Re: [PATCH] crypto: x86/curve25519 - fix cpu feature checking logic in mod_exit

On Thu, Jun 3, 2021 at 7:53 AM Hangbin Liu <[email protected]> wrote:
>
> In curve25519_mod_init() the curve25519_alg will be registered only when
> (X86_FEATURE_BMI2 && X86_FEATURE_ADX). But in curve25519_mod_exit()
> it still checks (X86_FEATURE_BMI2 || X86_FEATURE_ADX) when do crypto
> unregister. This will trigger a BUG_ON in crypto_unregister_alg() as
> alg->cra_refcnt is 0 if the cpu only supports one of X86_FEATURE_BMI2
> and X86_FEATURE_ADX.
>
> Fixes: 07b586fe0662 ("crypto: x86/curve25519 - replace with formally verified implementation")
> Signed-off-by: Hangbin Liu <[email protected]>
> ---
> arch/x86/crypto/curve25519-x86_64.c | 2 +-
> 1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/arch/x86/crypto/curve25519-x86_64.c b/arch/x86/crypto/curve25519-x86_64.c
> index 6706b6cb1d0f..38caf61cd5b7 100644
> --- a/arch/x86/crypto/curve25519-x86_64.c
> +++ b/arch/x86/crypto/curve25519-x86_64.c
> @@ -1500,7 +1500,7 @@ static int __init curve25519_mod_init(void)
> static void __exit curve25519_mod_exit(void)
> {
> if (IS_REACHABLE(CONFIG_CRYPTO_KPP) &&
> - (boot_cpu_has(X86_FEATURE_BMI2) || boot_cpu_has(X86_FEATURE_ADX)))
> + static_branch_likely(&curve25519_use_bmi2_adx))
> crypto_unregister_kpp(&curve25519_alg);
> }

Looks like the error is actually that the `||` should be a `&&`. But
if you'd like to branch on that static key instead, that's fine.

Cc: [email protected]
Reviewed-by: Jason A. Donenfeld <[email protected]>

2021-06-03 14:08:47

by Hangbin Liu

[permalink] [raw]
Subject: Re: [PATCH] crypto: x86/curve25519 - fix cpu feature checking logic in mod_exit

On Thu, Jun 3, 2021 at 6:24 PM Jason A. Donenfeld <[email protected]> wrote:
> > if (IS_REACHABLE(CONFIG_CRYPTO_KPP) &&
> > - (boot_cpu_has(X86_FEATURE_BMI2) || boot_cpu_has(X86_FEATURE_ADX)))
> > + static_branch_likely(&curve25519_use_bmi2_adx))
> > crypto_unregister_kpp(&curve25519_alg);
> > }
>
> Looks like the error is actually that the `||` should be a `&&`. But
> if you'd like to branch on that static key instead, that's fine.

Yes, the code would be shorter by checking the static key :)

Thanks
hangbin

2021-06-11 07:24:18

by Herbert Xu

[permalink] [raw]
Subject: Re: [PATCH] crypto: x86/curve25519 - fix cpu feature checking logic in mod_exit

On Thu, Jun 03, 2021 at 01:53:40AM -0400, Hangbin Liu wrote:
> In curve25519_mod_init() the curve25519_alg will be registered only when
> (X86_FEATURE_BMI2 && X86_FEATURE_ADX). But in curve25519_mod_exit()
> it still checks (X86_FEATURE_BMI2 || X86_FEATURE_ADX) when do crypto
> unregister. This will trigger a BUG_ON in crypto_unregister_alg() as
> alg->cra_refcnt is 0 if the cpu only supports one of X86_FEATURE_BMI2
> and X86_FEATURE_ADX.
>
> Fixes: 07b586fe0662 ("crypto: x86/curve25519 - replace with formally verified implementation")
> Signed-off-by: Hangbin Liu <[email protected]>
> ---
> arch/x86/crypto/curve25519-x86_64.c | 2 +-
> 1 file changed, 1 insertion(+), 1 deletion(-)

Patch applied. Thanks.
--
Email: Herbert Xu <[email protected]>
Home Page: http://gondor.apana.org.au/~herbert/
PGP Key: http://gondor.apana.org.au/~herbert/pubkey.txt

2021-06-11 10:10:19

by Jason A. Donenfeld

[permalink] [raw]
Subject: Re: [PATCH] crypto: x86/curve25519 - fix cpu feature checking logic in mod_exit

Hi Herbert,

Is there a reason why in
https://git.kernel.org/pub/scm/linux/kernel/git/herbert/cryptodev-2.6.git/patch/?id=1b82435d17774f3eaab35dce239d354548aa9da2
you didn't mark it with the Cc: [email protected] line that I included above my
Reviewed-by? Netdev no longer has their own stable process. Do you
have something else in mind for this?

Jason

2021-06-11 12:47:06

by Herbert Xu

[permalink] [raw]
Subject: Re: [PATCH] crypto: x86/curve25519 - fix cpu feature checking logic in mod_exit

On Fri, Jun 11, 2021 at 12:07:43PM +0200, Jason A. Donenfeld wrote:
> Hi Herbert,
>
> Is there a reason why in
> https://git.kernel.org/pub/scm/linux/kernel/git/herbert/cryptodev-2.6.git/patch/?id=1b82435d17774f3eaab35dce239d354548aa9da2
> you didn't mark it with the Cc: [email protected] line that I included above my
> Reviewed-by? Netdev no longer has their own stable process. Do you
> have something else in mind for this?

Hi Jason:

This patch has a Fixes header set and it'll be automatically pushed
to stable.

Cheers,
--
Email: Herbert Xu <[email protected]>
Home Page: http://gondor.apana.org.au/~herbert/
PGP Key: http://gondor.apana.org.au/~herbert/pubkey.txt