| Message ID | 1406797002-21015-1-git-send-email-ard.biesheuvel@linaro.org |
|---|---|
| State | New |
| Headers | show |
Hi Ard, On Thu, Jul 31, 2014 at 09:56:42AM +0100, Ard Biesheuvel wrote: > This changes the arm64 defconfig to build the ARMv8 Crypto Extensions > and NEON based crypto drivers as modules rather than as built-ins. The > reason is that the ARMv8 Crypto Extensions based modules will be probed > for automatically through udev upon detecting the availability of the > respective crypto instructions (AES, PMULL, SHA1, SHA256), and similarly, > the NEON alternatives will be probed for upon request (i.e., at first use > of the algorithm) if no hardware acceleration is available. > > Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org> > --- > arch/arm64/configs/defconfig | 14 +++++++------- > 1 file changed, 7 insertions(+), 7 deletions(-) > > diff --git a/arch/arm64/configs/defconfig b/arch/arm64/configs/defconfig > index c1071268c912..3310b4166dd0 100644 > --- a/arch/arm64/configs/defconfig > +++ b/arch/arm64/configs/defconfig > @@ -127,10 +127,10 @@ CONFIG_LOCKUP_DETECTOR=y > CONFIG_SECURITY=y > CONFIG_CRYPTO_ANSI_CPRNG=y > CONFIG_ARM64_CRYPTO=y > -CONFIG_CRYPTO_SHA1_ARM64_CE=y > -CONFIG_CRYPTO_SHA2_ARM64_CE=y > -CONFIG_CRYPTO_GHASH_ARM64_CE=y > -CONFIG_CRYPTO_AES_ARM64_CE=y > -CONFIG_CRYPTO_AES_ARM64_CE_CCM=y > -CONFIG_CRYPTO_AES_ARM64_CE_BLK=y > -CONFIG_CRYPTO_AES_ARM64_NEON_BLK=y > +CONFIG_CRYPTO_SHA1_ARM64_CE=m > +CONFIG_CRYPTO_SHA2_ARM64_CE=m > +CONFIG_CRYPTO_GHASH_ARM64_CE=m > +CONFIG_CRYPTO_AES_ARM64_CE=m > +CONFIG_CRYPTO_AES_ARM64_CE_CCM=m > +CONFIG_CRYPTO_AES_ARM64_CE_BLK=m > +CONFIG_CRYPTO_AES_ARM64_NEON_BLK=m Whilst I'm usually fairly open to defconfig updates if they make things more useful to people, I think building these as modules is going to be a pain for those not using udev (i.e. using busybox and/or simple filesystems for a VM). This seems more like something a distro might choose to do as opposed to something that is a good idea to have enabled by default. There could be an argument for a distroconfig, but I don't have high hopes that anybody would actually use it in practice. Will
On 31 July 2014 11:57, Will Deacon <will.deacon@arm.com> wrote: > Hi Ard, > > On Thu, Jul 31, 2014 at 09:56:42AM +0100, Ard Biesheuvel wrote: >> This changes the arm64 defconfig to build the ARMv8 Crypto Extensions >> and NEON based crypto drivers as modules rather than as built-ins. The >> reason is that the ARMv8 Crypto Extensions based modules will be probed >> for automatically through udev upon detecting the availability of the >> respective crypto instructions (AES, PMULL, SHA1, SHA256), and similarly, >> the NEON alternatives will be probed for upon request (i.e., at first use >> of the algorithm) if no hardware acceleration is available. >> >> Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org> >> --- >> arch/arm64/configs/defconfig | 14 +++++++------- >> 1 file changed, 7 insertions(+), 7 deletions(-) >> >> diff --git a/arch/arm64/configs/defconfig b/arch/arm64/configs/defconfig >> index c1071268c912..3310b4166dd0 100644 >> --- a/arch/arm64/configs/defconfig >> +++ b/arch/arm64/configs/defconfig >> @@ -127,10 +127,10 @@ CONFIG_LOCKUP_DETECTOR=y >> CONFIG_SECURITY=y >> CONFIG_CRYPTO_ANSI_CPRNG=y >> CONFIG_ARM64_CRYPTO=y >> -CONFIG_CRYPTO_SHA1_ARM64_CE=y >> -CONFIG_CRYPTO_SHA2_ARM64_CE=y >> -CONFIG_CRYPTO_GHASH_ARM64_CE=y >> -CONFIG_CRYPTO_AES_ARM64_CE=y >> -CONFIG_CRYPTO_AES_ARM64_CE_CCM=y >> -CONFIG_CRYPTO_AES_ARM64_CE_BLK=y >> -CONFIG_CRYPTO_AES_ARM64_NEON_BLK=y >> +CONFIG_CRYPTO_SHA1_ARM64_CE=m >> +CONFIG_CRYPTO_SHA2_ARM64_CE=m >> +CONFIG_CRYPTO_GHASH_ARM64_CE=m >> +CONFIG_CRYPTO_AES_ARM64_CE=m >> +CONFIG_CRYPTO_AES_ARM64_CE_CCM=m >> +CONFIG_CRYPTO_AES_ARM64_CE_BLK=m >> +CONFIG_CRYPTO_AES_ARM64_NEON_BLK=m > > Whilst I'm usually fairly open to defconfig updates if they make things more > useful to people, I think building these as modules is going to be a pain > for those not using udev (i.e. using busybox and/or simple filesystems for a > VM). This seems more like something a distro might choose to do as opposed > to something that is a good idea to have enabled by default. > > There could be an argument for a distroconfig, but I don't have high hopes > that anybody would actually use it in practice. > OK, fair enough
diff --git a/arch/arm64/configs/defconfig b/arch/arm64/configs/defconfig index c1071268c912..3310b4166dd0 100644 --- a/arch/arm64/configs/defconfig +++ b/arch/arm64/configs/defconfig @@ -127,10 +127,10 @@ CONFIG_LOCKUP_DETECTOR=y CONFIG_SECURITY=y CONFIG_CRYPTO_ANSI_CPRNG=y CONFIG_ARM64_CRYPTO=y -CONFIG_CRYPTO_SHA1_ARM64_CE=y -CONFIG_CRYPTO_SHA2_ARM64_CE=y -CONFIG_CRYPTO_GHASH_ARM64_CE=y -CONFIG_CRYPTO_AES_ARM64_CE=y -CONFIG_CRYPTO_AES_ARM64_CE_CCM=y -CONFIG_CRYPTO_AES_ARM64_CE_BLK=y -CONFIG_CRYPTO_AES_ARM64_NEON_BLK=y +CONFIG_CRYPTO_SHA1_ARM64_CE=m +CONFIG_CRYPTO_SHA2_ARM64_CE=m +CONFIG_CRYPTO_GHASH_ARM64_CE=m +CONFIG_CRYPTO_AES_ARM64_CE=m +CONFIG_CRYPTO_AES_ARM64_CE_CCM=m +CONFIG_CRYPTO_AES_ARM64_CE_BLK=m +CONFIG_CRYPTO_AES_ARM64_NEON_BLK=m
This changes the arm64 defconfig to build the ARMv8 Crypto Extensions and NEON based crypto drivers as modules rather than as built-ins. The reason is that the ARMv8 Crypto Extensions based modules will be probed for automatically through udev upon detecting the availability of the respective crypto instructions (AES, PMULL, SHA1, SHA256), and similarly, the NEON alternatives will be probed for upon request (i.e., at first use of the algorithm) if no hardware acceleration is available. Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org> --- arch/arm64/configs/defconfig | 14 +++++++------- 1 file changed, 7 insertions(+), 7 deletions(-)