openwrt/staging/blogic.git
5 years agocrypto: ecrdsa - add EC-RDSA (GOST 34.10) algorithm
Vitaly Chikunov [Thu, 11 Apr 2019 15:51:20 +0000 (18:51 +0300)]
crypto: ecrdsa - add EC-RDSA (GOST 34.10) algorithm

Add Elliptic Curve Russian Digital Signature Algorithm (GOST R
34.10-2012, RFC 7091, ISO/IEC 14888-3) is one of the Russian (and since
2018 the CIS countries) cryptographic standard algorithms (called GOST
algorithms). Only signature verification is supported, with intent to be
used in the IMA.

Summary of the changes:

* crypto/Kconfig:
  - EC-RDSA is added into Public-key cryptography section.

* crypto/Makefile:
  - ecrdsa objects are added.

* crypto/asymmetric_keys/x509_cert_parser.c:
  - Recognize EC-RDSA and Streebog OIDs.

* include/linux/oid_registry.h:
  - EC-RDSA OIDs are added to the enum. Also, a two currently not
    implemented curve OIDs are added for possible extension later (to
    not change numbering and grouping).

* crypto/ecc.c:
  - Kenneth MacKay copyright date is updated to 2014, because
    vli_mmod_slow, ecc_point_add, ecc_point_mult_shamir are based on his
    code from micro-ecc.
  - Functions needed for ecrdsa are EXPORT_SYMBOL'ed.
  - New functions:
    vli_is_negative - helper to determine sign of vli;
    vli_from_be64 - unpack big-endian array into vli (used for
      a signature);
    vli_from_le64 - unpack little-endian array into vli (used for
      a public key);
    vli_uadd, vli_usub - add/sub u64 value to/from vli (used for
      increment/decrement);
    mul_64_64 - optimized to use __int128 where appropriate, this speeds
      up point multiplication (and as a consequence signature
      verification) by the factor of 1.5-2;
    vli_umult - multiply vli by a small value (speeds up point
      multiplication by another factor of 1.5-2, depending on vli sizes);
    vli_mmod_special - module reduction for some form of Pseudo-Mersenne
      primes (used for the curves A);
    vli_mmod_special2 - module reduction for another form of
      Pseudo-Mersenne primes (used for the curves B);
    vli_mmod_barrett - module reduction using pre-computed value (used
      for the curve C);
    vli_mmod_slow - more general module reduction which is much slower
     (used when the modulus is subgroup order);
    vli_mod_mult_slow - modular multiplication;
    ecc_point_add - add two points;
    ecc_point_mult_shamir - add two points multiplied by scalars in one
      combined multiplication (this gives speed up by another factor 2 in
      compare to two separate multiplications).
    ecc_is_pubkey_valid_partial - additional samity check is added.
  - Updated vli_mmod_fast with non-strict heuristic to call optimal
      module reduction function depending on the prime value;
  - All computations for the previously defined (two NIST) curves should
    not unaffected.

* crypto/ecc.h:
  - Newly exported functions are documented.

* crypto/ecrdsa_defs.h
  - Five curves are defined.

* crypto/ecrdsa.c:
  - Signature verification is implemented.

* crypto/ecrdsa_params.asn1, crypto/ecrdsa_pub_key.asn1:
  - Templates for BER decoder for EC-RDSA parameters and public key.

Cc: linux-integrity@vger.kernel.org
Signed-off-by: Vitaly Chikunov <vt@altlinux.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: ecc - make ecc into separate module
Vitaly Chikunov [Thu, 11 Apr 2019 15:51:19 +0000 (18:51 +0300)]
crypto: ecc - make ecc into separate module

ecc.c have algorithms that could be used togeter by ecdh and ecrdsa.
Make it separate module. Add CRYPTO_ECC into Kconfig. EXPORT_SYMBOL and
document to what seems appropriate. Move structs ecc_point and ecc_curve
from ecc_curve_defs.h into ecc.h.

No code changes.

Signed-off-by: Vitaly Chikunov <vt@altlinux.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: Kconfig - create Public-key cryptography section
Vitaly Chikunov [Thu, 11 Apr 2019 15:51:18 +0000 (18:51 +0300)]
crypto: Kconfig - create Public-key cryptography section

Group RSA, DH, and ECDH into Public-key cryptography config section.

Signed-off-by: Vitaly Chikunov <vt@altlinux.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agoX.509: parse public key parameters from x509 for akcipher
Vitaly Chikunov [Thu, 11 Apr 2019 15:51:17 +0000 (18:51 +0300)]
X.509: parse public key parameters from x509 for akcipher

Some public key algorithms (like EC-DSA) keep in parameters field
important data such as digest and curve OIDs (possibly more for
different EC-DSA variants). Thus, just setting a public key (as
for RSA) is not enough.

Append parameters into the key stream for akcipher_set_{pub,priv}_key.
Appended data is: (u32) algo OID, (u32) parameters length, parameters
data.

This does not affect current akcipher API nor RSA ciphers (they could
ignore it). Idea of appending parameters to the key stream is by Herbert
Xu.

Cc: David Howells <dhowells@redhat.com>
Cc: Denis Kenzior <denkenz@gmail.com>
Cc: keyrings@vger.kernel.org
Signed-off-by: Vitaly Chikunov <vt@altlinux.org>
Reviewed-by: Denis Kenzior <denkenz@gmail.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agoKEYS: do not kmemdup digest in {public,tpm}_key_verify_signature
Vitaly Chikunov [Thu, 11 Apr 2019 15:51:16 +0000 (18:51 +0300)]
KEYS: do not kmemdup digest in {public,tpm}_key_verify_signature

Treat (struct public_key_signature)'s digest same as its signature (s).
Since digest should be already in the kmalloc'd memory do not kmemdup
digest value before calling {public,tpm}_key_verify_signature.

Patch is split from the previous as suggested by Herbert Xu.

Suggested-by: David Howells <dhowells@redhat.com>
Cc: David Howells <dhowells@redhat.com>
Cc: keyrings@vger.kernel.org
Signed-off-by: Vitaly Chikunov <vt@altlinux.org>
Reviewed-by: Denis Kenzior <denkenz@gmail.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: akcipher - new verify API for public key algorithms
Vitaly Chikunov [Thu, 11 Apr 2019 15:51:15 +0000 (18:51 +0300)]
crypto: akcipher - new verify API for public key algorithms

Previous akcipher .verify() just `decrypts' (using RSA encrypt which is
using public key) signature to uncover message hash, which was then
compared in upper level public_key_verify_signature() with the expected
hash value, which itself was never passed into verify().

This approach was incompatible with EC-DSA family of algorithms,
because, to verify a signature EC-DSA algorithm also needs a hash value
as input; then it's used (together with a signature divided into halves
`r||s') to produce a witness value, which is then compared with `r' to
determine if the signature is correct. Thus, for EC-DSA, nor
requirements of .verify() itself, nor its output expectations in
public_key_verify_signature() wasn't sufficient.

Make improved .verify() call which gets hash value as input and produce
complete signature check without any output besides status.

Now for the top level verification only crypto_akcipher_verify() needs
to be called and its return value inspected.

Make sure that `digest' is in kmalloc'd memory (in place of `output`) in
{public,tpm}_key_verify_signature() as insisted by Herbert Xu, and will
be changed in the following commit.

Cc: David Howells <dhowells@redhat.com>
Cc: keyrings@vger.kernel.org
Signed-off-by: Vitaly Chikunov <vt@altlinux.org>
Reviewed-by: Denis Kenzior <denkenz@gmail.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: rsa - unimplement sign/verify for raw RSA backends
Vitaly Chikunov [Thu, 11 Apr 2019 15:51:14 +0000 (18:51 +0300)]
crypto: rsa - unimplement sign/verify for raw RSA backends

In preparation for new akcipher verify call remove sign/verify callbacks
from RSA backends and make PKCS1 driver call encrypt/decrypt instead.

This also complies with the well-known idea that raw RSA should never be
used for sign/verify. It only should be used with proper padding scheme
such as PKCS1 driver provides.

Cc: Giovanni Cabiddu <giovanni.cabiddu@intel.com>
Cc: qat-linux@intel.com
Cc: Tom Lendacky <thomas.lendacky@amd.com>
Cc: Gary Hook <gary.hook@amd.com>
Cc: Horia Geantă <horia.geanta@nxp.com>
Cc: Aymen Sghaier <aymen.sghaier@nxp.com>
Signed-off-by: Vitaly Chikunov <vt@altlinux.org>
Reviewed-by: Horia Geantă <horia.geanta@nxp.com>
Acked-by: Gary R Hook <gary.hook@amd.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: akcipher - default implementations for request callbacks
Vitaly Chikunov [Thu, 11 Apr 2019 15:51:13 +0000 (18:51 +0300)]
crypto: akcipher - default implementations for request callbacks

Because with the introduction of EC-RDSA and change in workings of RSA
in regard to sign/verify, akcipher could have not all callbacks defined,
check the presence of callbacks in crypto_register_akcipher() and
provide default implementation if the callback is not implemented.

This is suggested by Herbert Xu instead of checking the presence of the
callback on every request.

Signed-off-by: Vitaly Chikunov <vt@altlinux.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: ux500 - Forbid 2-key 3DES in FIPS mode
Herbert Xu [Thu, 11 Apr 2019 08:51:22 +0000 (16:51 +0800)]
crypto: ux500 - Forbid 2-key 3DES in FIPS mode

This patch forbids the use of 2-key 3DES (K1 == K3) in FIPS mode.

It also removes the registration of the non-standard des/des3
ablkcipher algorithms.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: talitos - Forbid 2-key 3DES in FIPS mode
Herbert Xu [Thu, 11 Apr 2019 08:51:21 +0000 (16:51 +0800)]
crypto: talitos - Forbid 2-key 3DES in FIPS mode

This patch forbids the use of 2-key 3DES (K1 == K3) in FIPS mode.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: sun4i-ss - Forbid 2-key 3DES in FIPS mode
Herbert Xu [Thu, 11 Apr 2019 08:51:19 +0000 (16:51 +0800)]
crypto: sun4i-ss - Forbid 2-key 3DES in FIPS mode

This patch forbids the use of 2-key 3DES (K1 == K3) in FIPS mode.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Acked-by: Corentin Labbe <clabbe.montjoie@gmail.com>
Tested-by: Corentin Labbe <clabbe.montjoie@gmail.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: stm32 - Forbid 2-key 3DES in FIPS mode
Herbert Xu [Thu, 11 Apr 2019 08:51:18 +0000 (16:51 +0800)]
crypto: stm32 - Forbid 2-key 3DES in FIPS mode

This patch forbids the use of 2-key 3DES (K1 == K3) in FIPS mode.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Acked-by: Lionel Debieve<lionel.debieve@st.com>
Tested-by: Lionel Debieve<lionel.debieve@st.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: rockchip - Forbid 2-key 3DES in FIPS mode
Herbert Xu [Thu, 11 Apr 2019 08:51:17 +0000 (16:51 +0800)]
crypto: rockchip - Forbid 2-key 3DES in FIPS mode

This patch forbids the use of 2-key 3DES (K1 == K3) in FIPS mode.

It also removes a couple of unnecessary key length checks that
are already performed by the crypto API.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: qce - Forbid 2-key 3DES in FIPS mode
Herbert Xu [Thu, 11 Apr 2019 08:51:16 +0000 (16:51 +0800)]
crypto: qce - Forbid 2-key 3DES in FIPS mode

This patch forbids the use of 2-key 3DES (K1 == K3) in FIPS mode.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: picoxcell - Forbid 2-key 3DES in FIPS mode
Herbert Xu [Thu, 11 Apr 2019 08:51:15 +0000 (16:51 +0800)]
crypto: picoxcell - Forbid 2-key 3DES in FIPS mode

This patch forbids the use of 2-key 3DES (K1 == K3) in FIPS mode.

It also removes an unnecessary key length checks that are already
performed by the crypto API.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: omap - Forbid 2-key 3DES in FIPS mode
Herbert Xu [Thu, 11 Apr 2019 08:51:14 +0000 (16:51 +0800)]
crypto: omap - Forbid 2-key 3DES in FIPS mode

This patch forbids the use of 2-key 3DES (K1 == K3) in FIPS mode.

It also removes a couple of unnecessary key length checks that
are already performed by the crypto API.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: n2 - Forbid 2-key 3DES in FIPS mode
Herbert Xu [Thu, 11 Apr 2019 08:51:13 +0000 (16:51 +0800)]
crypto: n2 - Forbid 2-key 3DES in FIPS mode

This patch forbids the use of 2-key 3DES (K1 == K3) in FIPS mode.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: marvell - Forbid 2-key 3DES in FIPS mode
Herbert Xu [Thu, 11 Apr 2019 08:51:12 +0000 (16:51 +0800)]
crypto: marvell - Forbid 2-key 3DES in FIPS mode

This patch forbids the use of 2-key 3DES (K1 == K3) in FIPS mode.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: ixp4xx - Forbid 2-key 3DES in FIPS mode
Herbert Xu [Thu, 11 Apr 2019 08:51:11 +0000 (16:51 +0800)]
crypto: ixp4xx - Forbid 2-key 3DES in FIPS mode

This patch forbids the use of 2-key 3DES (K1 == K3) in FIPS mode.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: inside-secure - Forbid 2-key 3DES in FIPS mode
Herbert Xu [Thu, 11 Apr 2019 08:51:10 +0000 (16:51 +0800)]
crypto: inside-secure - Forbid 2-key 3DES in FIPS mode

This patch forbids the use of 2-key 3DES (K1 == K3) in FIPS mode.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: hisilicon - Forbid 2-key 3DES in FIPS mode
Herbert Xu [Thu, 11 Apr 2019 08:51:08 +0000 (16:51 +0800)]
crypto: hisilicon - Forbid 2-key 3DES in FIPS mode

This patch forbids the use of 2-key 3DES (K1 == K3) in FIPS mode.

It also removes a couple of unnecessary key length checks that
are already performed by the crypto API.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: hifn_795x - Forbid 2-key 3DES in FIPS mode
Herbert Xu [Thu, 11 Apr 2019 08:51:07 +0000 (16:51 +0800)]
crypto: hifn_795x - Forbid 2-key 3DES in FIPS mode

This patch forbids the use of 2-key 3DES (K1 == K3) in FIPS mode.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: ccree - Forbid 2-key 3DES in FIPS mode
Herbert Xu [Thu, 11 Apr 2019 08:51:06 +0000 (16:51 +0800)]
crypto: ccree - Forbid 2-key 3DES in FIPS mode

This patch forbids the use of 2-key 3DES (K1 == K3) in FIPS mode.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Acked-by: Gilad Ben-Yossef <gilad@benyossef.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: ccp - Forbid 2-key 3DES in FIPS mode
Herbert Xu [Thu, 11 Apr 2019 08:51:05 +0000 (16:51 +0800)]
crypto: ccp - Forbid 2-key 3DES in FIPS mode

This patch forbids the use of 2-key 3DES (K1 == K3) in FIPS mode.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: nitrox - Forbid 2-key 3DES in FIPS mode
Herbert Xu [Thu, 11 Apr 2019 08:51:04 +0000 (16:51 +0800)]
crypto: nitrox - Forbid 2-key 3DES in FIPS mode

This patch forbids the use of 2-key 3DES (K1 == K3) in FIPS mode.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: cavium - Forbid 2-key 3DES in FIPS mode
Herbert Xu [Thu, 11 Apr 2019 08:51:03 +0000 (16:51 +0800)]
crypto: cavium - Forbid 2-key 3DES in FIPS mode

This patch forbids the use of 2-key 3DES (K1 == K3) in FIPS mode.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: caam - Forbid 2-key 3DES in FIPS mode
Herbert Xu [Thu, 11 Apr 2019 08:51:02 +0000 (16:51 +0800)]
crypto: caam - Forbid 2-key 3DES in FIPS mode

This patch forbids the use of 2-key 3DES (K1 == K3) in FIPS mode.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Reviewed-by: Horia Geantă <horia.geanta@nxp.com>
Tested-by: Iuliana Prodan <iuliana.prodan@nxp.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: bcm - Forbid 2-key 3DES in FIPS mode
Herbert Xu [Thu, 11 Apr 2019 08:51:01 +0000 (16:51 +0800)]
crypto: bcm - Forbid 2-key 3DES in FIPS mode

This patch forbids the use of 2-key 3DES (K1 == K3) in FIPS mode.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: atmel - Forbid 2-key 3DES in FIPS mode
Herbert Xu [Thu, 11 Apr 2019 08:51:00 +0000 (16:51 +0800)]
crypto: atmel - Forbid 2-key 3DES in FIPS mode

This patch forbids the use of 2-key 3DES (K1 == K3) in FIPS mode.

This patch also removes the bogus CFB 3DES modes that only work
with a short 3DES key not otherwise allowed by the crypto API.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: sparc - Forbid 2-key 3DES in FIPS mode
Herbert Xu [Thu, 11 Apr 2019 08:50:58 +0000 (16:50 +0800)]
crypto: sparc - Forbid 2-key 3DES in FIPS mode

This patch forbids the use of 2-key 3DES (K1 == K3) in FIPS mode.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: s390 - Forbid 2-key 3DES in FIPS mode
Herbert Xu [Thu, 11 Apr 2019 08:50:57 +0000 (16:50 +0800)]
crypto: s390 - Forbid 2-key 3DES in FIPS mode

This patch forbids the use of 2-key 3DES (K1 == K3) in FIPS mode.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: des_generic - Forbid 2-key in 3DES and add helpers
Herbert Xu [Thu, 11 Apr 2019 08:50:56 +0000 (16:50 +0800)]
crypto: des_generic - Forbid 2-key in 3DES and add helpers

This patch adds a requirement to the generic 3DES implementation
such that 2-key 3DES (K1 == K3) is no longer allowed in FIPS mode.

We will also provide helpers that may be used by drivers that
implement 3DES to make the same check.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: vmx - return correct error code on failed setkey
Eric Biggers [Wed, 10 Apr 2019 06:46:35 +0000 (23:46 -0700)]
crypto: vmx - return correct error code on failed setkey

In the VMX implementations of AES and AES modes, return -EINVAL when an
invalid key length is provided, rather than some unusual error code
determined via a series of additions.  This makes the behavior match the
other AES implementations in the kernel's crypto API.

Cc: Daniel Axtens <dja@axtens.net>
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: arm64/aes-neonbs - don't access already-freed walk.iv
Eric Biggers [Wed, 10 Apr 2019 06:46:32 +0000 (23:46 -0700)]
crypto: arm64/aes-neonbs - don't access already-freed walk.iv

If the user-provided IV needs to be aligned to the algorithm's
alignmask, then skcipher_walk_virt() copies the IV into a new aligned
buffer walk.iv.  But skcipher_walk_virt() can fail afterwards, and then
if the caller unconditionally accesses walk.iv, it's a use-after-free.

xts-aes-neonbs doesn't set an alignmask, so currently it isn't affected
by this despite unconditionally accessing walk.iv.  However this is more
subtle than desired, and unconditionally accessing walk.iv has caused a
real problem in other algorithms.  Thus, update xts-aes-neonbs to start
checking the return value of skcipher_walk_virt().

Fixes: 1abee99eafab ("crypto: arm64/aes - reimplement bit-sliced ARM/NEON implementation for arm64")
Cc: <stable@vger.kernel.org> # v4.11+
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: arm/aes-neonbs - don't access already-freed walk.iv
Eric Biggers [Wed, 10 Apr 2019 06:46:31 +0000 (23:46 -0700)]
crypto: arm/aes-neonbs - don't access already-freed walk.iv

If the user-provided IV needs to be aligned to the algorithm's
alignmask, then skcipher_walk_virt() copies the IV into a new aligned
buffer walk.iv.  But skcipher_walk_virt() can fail afterwards, and then
if the caller unconditionally accesses walk.iv, it's a use-after-free.

arm32 xts-aes-neonbs doesn't set an alignmask, so currently it isn't
affected by this despite unconditionally accessing walk.iv.  However
this is more subtle than desired, and it was actually broken prior to
the alignmask being removed by commit cc477bf64573 ("crypto: arm/aes -
replace bit-sliced OpenSSL NEON code").  Thus, update xts-aes-neonbs to
start checking the return value of skcipher_walk_virt().

Fixes: e4e7f10bfc40 ("ARM: add support for bit sliced AES using NEON instructions")
Cc: <stable@vger.kernel.org> # v3.13+
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: salsa20 - don't access already-freed walk.iv
Eric Biggers [Wed, 10 Apr 2019 06:46:30 +0000 (23:46 -0700)]
crypto: salsa20 - don't access already-freed walk.iv

If the user-provided IV needs to be aligned to the algorithm's
alignmask, then skcipher_walk_virt() copies the IV into a new aligned
buffer walk.iv.  But skcipher_walk_virt() can fail afterwards, and then
if the caller unconditionally accesses walk.iv, it's a use-after-free.

salsa20-generic doesn't set an alignmask, so currently it isn't affected
by this despite unconditionally accessing walk.iv.  However this is more
subtle than desired, and it was actually broken prior to the alignmask
being removed by commit b62b3db76f73 ("crypto: salsa20-generic - cleanup
and convert to skcipher API").

Since salsa20-generic does not update the IV and does not need any IV
alignment, update it to use req->iv instead of walk.iv.

Fixes: 2407d60872dd ("[CRYPTO] salsa20: Salsa20 stream cipher")
Cc: stable@vger.kernel.org
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: lrw - don't access already-freed walk.iv
Eric Biggers [Wed, 10 Apr 2019 06:46:29 +0000 (23:46 -0700)]
crypto: lrw - don't access already-freed walk.iv

If the user-provided IV needs to be aligned to the algorithm's
alignmask, then skcipher_walk_virt() copies the IV into a new aligned
buffer walk.iv.  But skcipher_walk_virt() can fail afterwards, and then
if the caller unconditionally accesses walk.iv, it's a use-after-free.

Fix this in the LRW template by checking the return value of
skcipher_walk_virt().

This bug was detected by my patches that improve testmgr to fuzz
algorithms against their generic implementation.  When the extra
self-tests were run on a KASAN-enabled kernel, a KASAN use-after-free
splat occured during lrw(aes) testing.

Fixes: c778f96bf347 ("crypto: lrw - Optimize tweak computation")
Cc: <stable@vger.kernel.org> # v4.20+
Cc: Ondrej Mosnacek <omosnace@redhat.com>
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: mxs-dcp - remove set but not used variable 'fini'
YueHaibing [Wed, 10 Apr 2019 02:47:42 +0000 (02:47 +0000)]
crypto: mxs-dcp - remove set but not used variable 'fini'

Fixes gcc '-Wunused-but-set-variable' warning:

drivers/crypto/mxs-dcp.c: In function 'dcp_chan_thread_sha':
drivers/crypto/mxs-dcp.c:707:11: warning:
 variable 'fini' set but not used [-Wunused-but-set-variable]

It's not used since commit d80771c08363 ("crypto: mxs-dcp - Fix wait
logic on chan threads"),so can be removed.

Signed-off-by: YueHaibing <yuehaibing@huawei.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: sahara - Convert IS_ENABLED uses to __is_defined
Joe Perches [Tue, 9 Apr 2019 16:33:13 +0000 (09:33 -0700)]
crypto: sahara - Convert IS_ENABLED uses to __is_defined

IS_ENABLED should be reserved for CONFIG_<FOO> uses so convert
the uses of IS_ENABLED with a #define to __is_defined.

Signed-off-by: Joe Perches <joe@perches.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: caam/jr - Remove extra memory barrier during job ring dequeue
Vakul Garg [Tue, 9 Apr 2019 06:38:08 +0000 (06:38 +0000)]
crypto: caam/jr - Remove extra memory barrier during job ring dequeue

In function caam_jr_dequeue(), a full memory barrier is used before
writing response job ring's register to signal removal of the completed
job. Therefore for writing the register, we do not need another write
memory barrier. Hence it is removed by replacing the call to wr_reg32()
with a newly defined function wr_reg32_relaxed().

Signed-off-by: Vakul Garg <vakul.garg@nxp.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: ccp - Do not free psp_master when PLATFORM_INIT fails
Singh, Brijesh [Mon, 8 Apr 2019 20:42:55 +0000 (20:42 +0000)]
crypto: ccp - Do not free psp_master when PLATFORM_INIT fails

Currently, we free the psp_master if the PLATFORM_INIT fails during the
SEV FW probe. If psp_master is freed then driver does not invoke the PSP
FW. As per SEV FW spec, there are several commands (PLATFORM_RESET,
PLATFORM_STATUS, GET_ID etc) which can be executed in the UNINIT state
We should not free the psp_master when PLATFORM_INIT fails.

Fixes: 200664d5237f ("crypto: ccp: Add SEV support")
Cc: Tom Lendacky <thomas.lendacky@amd.com>
Cc: Herbert Xu <herbert@gondor.apana.org.au>
Cc: Gary Hook <gary.hook@amd.com>
Cc: stable@vger.kernel.org # 4.19.y
Signed-off-by: Brijesh Singh <brijesh.singh@amd.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: stm32/hash - Fix self test issue during export
Lionel Debieve [Mon, 8 Apr 2019 07:41:58 +0000 (09:41 +0200)]
crypto: stm32/hash - Fix self test issue during export

Change the wait condition to check if the hash is busy.
Context can be saved as soon as hash has finishing processing
data. Remove unused lock in the device structure.

Signed-off-by: Lionel Debieve <lionel.debieve@st.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: mxc-scc - Remove broken driver
Herbert Xu [Mon, 1 Apr 2019 12:54:30 +0000 (20:54 +0800)]
crypto: mxc-scc - Remove broken driver

This driver has been completely broken since the very beginning
because it doesn't even have a setkey function.  This means that
nobody has ever used it as it would crash during setkey.

This patch removes this driver.

Fixes: d293b640ebd5 ("crypto: mxc-scc - add basic driver for the...")
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agohwrng: stm32 - set default random quality
Lionel Debieve [Mon, 1 Apr 2019 10:30:46 +0000 (12:30 +0200)]
hwrng: stm32 - set default random quality

Add a default quality to hw_random device to be
automatically set as new default entropy. Setting
random quality will decrease the crng init time by
switching to this hardware random source.

Signed-off-by: Lionel Debieve <lionel.debieve@st.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agohwrng: stm32 - fix unbalanced pm_runtime_enable
Lionel Debieve [Mon, 1 Apr 2019 10:30:45 +0000 (12:30 +0200)]
hwrng: stm32 - fix unbalanced pm_runtime_enable

No remove function implemented yet in the driver.
Without remove function, the pm_runtime implementation
complains when removing and probing again the driver.

Signed-off-by: Lionel Debieve <lionel.debieve@st.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: testmgr - add panic_on_fail module parameter
Eric Biggers [Sun, 31 Mar 2019 20:09:14 +0000 (13:09 -0700)]
crypto: testmgr - add panic_on_fail module parameter

Add a module parameter cryptomgr.panic_on_fail which causes the kernel
to panic if any crypto self-tests fail.

Use cases:

- More easily detect crypto self-test failures by boot testing,
  e.g. on KernelCI.
- Get a bug report if syzkaller manages to use the template system to
  instantiate an algorithm that fails its self-tests.

The command-line option "fips=1" already does this, but it also makes
other changes not wanted for general testing, such as disabling
"unapproved" algorithms.  panic_on_fail just does what it says.

Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: arm64/cbcmac - handle empty messages in same way as template
Eric Biggers [Sun, 31 Mar 2019 20:04:21 +0000 (13:04 -0700)]
crypto: arm64/cbcmac - handle empty messages in same way as template

My patches to make testmgr fuzz algorithms against their generic
implementation detected that the arm64 implementations of "cbcmac(aes)"
handle empty messages differently from the cbcmac template.  Namely, the
arm64 implementations return the encrypted initial value, but the cbcmac
template returns the initial value directly.

This isn't actually a meaningful case because any user of cbcmac needs
to prepend the message length, as CCM does; otherwise it's insecure.
However, we should keep the behavior consistent; at the very least this
makes testing easier.

Do it the easy way, which is to change the arm64 implementations to have
the same behavior as the cbcmac template.

For what it's worth, ghash does things essentially the same way: it
returns its initial value when given an empty message, even though in
practice ghash is never passed an empty message.

Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: cts - don't support empty messages
Eric Biggers [Sun, 31 Mar 2019 20:04:20 +0000 (13:04 -0700)]
crypto: cts - don't support empty messages

My patches to make testmgr fuzz algorithms against their generic
implementation detected that the arm64 implementations of
"cts(cbc(aes))" handle empty messages differently from the cts template.
Namely, the arm64 implementations forbids (with -EINVAL) all messages
shorter than the block size, including the empty message; but the cts
template permits empty messages as a special case.

No user should be CTS-encrypting/decrypting empty messages, but we need
to keep the behavior consistent.  Unfortunately, as noted in the source
of OpenSSL's CTS implementation [1], there's no common specification for
CTS.  This makes it somewhat debatable what the behavior should be.

However, all CTS specifications seem to agree that messages shorter than
the block size are not allowed, and OpenSSL follows this in both CTS
conventions it implements.  It would also simplify the user-visible
semantics to have empty messages no longer be a special case.

Therefore, make the cts template return -EINVAL on *all* messages
shorter than the block size, including the empty message.

[1] https://github.com/openssl/openssl/blob/master/crypto/modes/cts128.c

Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: streebog - fix unaligned memory accesses
Eric Biggers [Sun, 31 Mar 2019 20:04:19 +0000 (13:04 -0700)]
crypto: streebog - fix unaligned memory accesses

Don't cast the data buffer directly to streebog_uint512, as this
violates alignment rules.

Fixes: fe18957e8e87 ("crypto: streebog - add Streebog hash function")
Cc: Vitaly Chikunov <vt@altlinux.org>
Signed-off-by: Eric Biggers <ebiggers@google.com>
Reviewed-by: Vitaly Chikunov <vt@altlinux.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: chacha20poly1305 - set cra_name correctly
Eric Biggers [Sun, 31 Mar 2019 20:04:16 +0000 (13:04 -0700)]
crypto: chacha20poly1305 - set cra_name correctly

If the rfc7539 template is instantiated with specific implementations,
e.g. "rfc7539(chacha20-generic,poly1305-generic)" rather than
"rfc7539(chacha20,poly1305)", then the implementation names end up
included in the instance's cra_name.  This is incorrect because it then
prevents all users from allocating "rfc7539(chacha20,poly1305)", if the
highest priority implementations of chacha20 and poly1305 were selected.
Also, the self-tests aren't run on an instance allocated in this way.

Fix it by setting the instance's cra_name from the underlying
algorithms' actual cra_names, rather than from the requested names.
This matches what other templates do.

Fixes: 71ebc4d1b27d ("crypto: chacha20poly1305 - Add a ChaCha20-Poly1305 AEAD construction, RFC7539")
Cc: <stable@vger.kernel.org> # v4.2+
Cc: Martin Willi <martin@strongswan.org>
Signed-off-by: Eric Biggers <ebiggers@google.com>
Reviewed-by: Martin Willi <martin@strongswan.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: skcipher - don't WARN on unprocessed data after slow walk step
Eric Biggers [Sun, 31 Mar 2019 20:04:15 +0000 (13:04 -0700)]
crypto: skcipher - don't WARN on unprocessed data after slow walk step

skcipher_walk_done() assumes it's a bug if, after the "slow" path is
executed where the next chunk of data is processed via a bounce buffer,
the algorithm says it didn't process all bytes.  Thus it WARNs on this.

However, this can happen legitimately when the message needs to be
evenly divisible into "blocks" but isn't, and the algorithm has a
'walksize' greater than the block size.  For example, ecb-aes-neonbs
sets 'walksize' to 128 bytes and only supports messages evenly divisible
into 16-byte blocks.  If, say, 17 message bytes remain but they straddle
scatterlist elements, the skcipher_walk code will take the "slow" path
and pass the algorithm all 17 bytes in the bounce buffer.  But the
algorithm will only be able to process 16 bytes, triggering the WARN.

Fix this by just removing the WARN_ON().  Returning -EINVAL, as the code
already does, is the right behavior.

This bug was detected by my patches that improve testmgr to fuzz
algorithms against their generic implementation.

Fixes: b286d8b1a690 ("crypto: skcipher - Add skcipher walk interface")
Cc: <stable@vger.kernel.org> # v4.10+
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: x86/crct10dif-pcl - fix use via crypto_shash_digest()
Eric Biggers [Sun, 31 Mar 2019 20:04:13 +0000 (13:04 -0700)]
crypto: x86/crct10dif-pcl - fix use via crypto_shash_digest()

The ->digest() method of crct10dif-pclmul reads the current CRC value
from the shash_desc context.  But this value is uninitialized, causing
crypto_shash_digest() to compute the wrong result.  Fix it.

Probably this wasn't noticed before because lib/crc-t10dif.c only uses
crypto_shash_update(), not crypto_shash_digest().  Likewise,
crypto_shash_digest() is not yet tested by the crypto self-tests because
those only test the ahash API which only uses shash init/update/final.

Fixes: 0b95a7f85718 ("crypto: crct10dif - Glue code to cast accelerated CRCT10DIF assembly as a crypto transform")
Cc: <stable@vger.kernel.org> # v3.11+
Cc: Tim Chen <tim.c.chen@linux.intel.com>
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: crct10dif-generic - fix use via crypto_shash_digest()
Eric Biggers [Sun, 31 Mar 2019 20:04:12 +0000 (13:04 -0700)]
crypto: crct10dif-generic - fix use via crypto_shash_digest()

The ->digest() method of crct10dif-generic reads the current CRC value
from the shash_desc context.  But this value is uninitialized, causing
crypto_shash_digest() to compute the wrong result.  Fix it.

Probably this wasn't noticed before because lib/crc-t10dif.c only uses
crypto_shash_update(), not crypto_shash_digest().  Likewise,
crypto_shash_digest() is not yet tested by the crypto self-tests because
those only test the ahash API which only uses shash init/update/final.

This bug was detected by my patches that improve testmgr to fuzz
algorithms against their generic implementation.

Fixes: 2d31e518a428 ("crypto: crct10dif - Wrap crc_t10dif function all to use crypto transform framework")
Cc: <stable@vger.kernel.org> # v3.11+
Cc: Tim Chen <tim.c.chen@linux.intel.com>
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: nx842 - remove set but not used variables 'dpadding' and 'max_sync_size'
YueHaibing [Sat, 30 Mar 2019 05:54:29 +0000 (13:54 +0800)]
crypto: nx842 - remove set but not used variables 'dpadding' and 'max_sync_size'

Fixes gcc '-Wunused-but-set-variable' warning:

drivers/crypto/nx/nx-842.c: In function 'decompress':
drivers/crypto/nx/nx-842.c:356:25: warning: variable 'dpadding' set but not used [-Wunused-but-set-variable]
drivers/crypto/nx/nx-842-pseries.c: In function 'nx842_pseries_compress':
drivers/crypto/nx/nx-842-pseries.c:299:15: warning: variable 'max_sync_size' set but not used [-Wunused-but-set-variable]
drivers/crypto/nx/nx-842-pseries.c: In function 'nx842_pseries_decompress':
drivers/crypto/nx/nx-842-pseries.c:430:15: warning: variable 'max_sync_size' set but not used [-Wunused-but-set-variable]

They are not used any more and can be removed.

Signed-off-by: YueHaibing <yuehaibing@huawei.com>
Reviewed-by: Mukesh Ojha <mojha@codeaurora.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: mxs-dcp - return errcode in mxs_dcp_aes_enqueue and dcp_sha_update_fx
YueHaibing [Sat, 30 Mar 2019 05:52:21 +0000 (13:52 +0800)]
crypto: mxs-dcp - return errcode in mxs_dcp_aes_enqueue and dcp_sha_update_fx

'err' is set in err path, but it's not returned to callers.
Don't always return -EINPROGRESS, return err.

Signed-off-by: YueHaibing <yuehaibing@huawei.com>
Reviewed-by: Mukesh Ojha <mojha@codeaurora.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: marvell - remove set but not used variable 'index'
YueHaibing [Sat, 30 Mar 2019 05:28:58 +0000 (13:28 +0800)]
crypto: marvell - remove set but not used variable 'index'

Fixes gcc '-Wunused-but-set-variable' warning:

drivers/crypto/marvell/hash.c: In function 'mv_cesa_ahash_pad_req':
drivers/crypto/marvell/hash.c:138:15: warning: variable 'index' set but not used [-Wunused-but-set-variable]

It's never used and can be removed.

Signed-off-by: YueHaibing <yuehaibing@huawei.com>
Reviewed-by: Mukesh Ojha <mojha@codeaurora.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: ccp - Use kmemdup in ccp_copy_and_save_keypart()
YueHaibing [Sat, 30 Mar 2019 01:43:16 +0000 (01:43 +0000)]
crypto: ccp - Use kmemdup in ccp_copy_and_save_keypart()

Use kmemdup rather than duplicating its implementation

Signed-off-by: YueHaibing <yuehaibing@huawei.com>
Acked-by: Gary R Hook <gary.hook@amd.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: aes - Use ___cacheline_aligned for aes data
Andi Kleen [Sat, 30 Mar 2019 00:46:29 +0000 (17:46 -0700)]
crypto: aes - Use ___cacheline_aligned for aes data

cacheline_aligned is a special section. It cannot be const at the same
time because it's not read-only. It doesn't give any MMU protection.

Mark it ____cacheline_aligned to not place it in a special section,
but just align it in .rodata

Cc: herbert@gondor.apana.org.au
Suggested-by: Rasmus Villemoes <linux@rasmusvillemoes.dk>
Signed-off-by: Andi Kleen <ak@linux.intel.com>
Acked-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: scompress - Use per-CPU struct instead multiple variables
Sebastian Andrzej Siewior [Fri, 29 Mar 2019 13:09:56 +0000 (14:09 +0100)]
crypto: scompress - Use per-CPU struct instead multiple variables

Two per-CPU variables are allocated as pointer to per-CPU memory which
then are used as scratch buffers.
We could be smart about this and use instead a per-CPU struct which
contains the pointers already and then we need to allocate just the
scratch buffers.
Add a lock to the struct. By doing so we can avoid the get_cpu()
statement and gain lockdep coverage (if enabled) to ensure that the lock
is always acquired in the right context. On non-preemptible kernels the
lock vanishes.
It is okay to use raw_cpu_ptr() in order to get a pointer to the struct
since it is protected by the spinlock.

The diffstat of this is negative and according to size scompress.o:
   text    data     bss     dec     hex filename
   1847     160      24    2031     7ef dbg_before.o
   1754     232       4    1990     7c6 dbg_after.o
   1799      64      24    1887     75f no_dbg-before.o
   1703      88       4    1795     703 no_dbg-after.o

The overall size increase difference is also negative. The increase in
the data section is only four bytes without lockdep.

Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: scompress - return proper error code for allocation failure
Sebastian Andrzej Siewior [Fri, 29 Mar 2019 13:09:55 +0000 (14:09 +0100)]
crypto: scompress - return proper error code for allocation failure

If scomp_acomp_comp_decomp() fails to allocate memory for the
destination then we never copy back the data we compressed.
It is probably best to return an error code instead 0 in case of
failure.
I haven't found any user that is using acomp_request_set_params()
without the `dst' buffer so there is probably no harm.

Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: ccp - introduce SEV_GET_ID2 command
Singh, Brijesh [Thu, 28 Mar 2019 21:58:52 +0000 (21:58 +0000)]
crypto: ccp - introduce SEV_GET_ID2 command

The current definition and implementation of the SEV_GET_ID command
does not provide the length of the unique ID returned by the firmware.
As per the firmware specification, the firmware may return an ID
length that is not restricted to 64 bytes as assumed by the SEV_GET_ID
command.

Introduce the SEV_GET_ID2 command to overcome with the SEV_GET_ID
limitations. Deprecate the SEV_GET_ID in the favor of SEV_GET_ID2.

At the same time update SEV API web link.

Cc: Janakarajan Natarajan <Janakarajan.Natarajan@amd.com>
Cc: Herbert Xu <herbert@gondor.apana.org.au>
Cc: Gary Hook <gary.hook@amd.com>
Cc: Tom Lendacky <thomas.lendacky@amd.com>
Cc: Nathaniel McCallum <npmccallum@redhat.com>
Signed-off-by: Brijesh Singh <brijesh.singh@amd.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: caam/qi - Change a couple IS_ERR_OR_NULL() checks to IS_ERR()
Dan Carpenter [Thu, 28 Mar 2019 14:36:01 +0000 (17:36 +0300)]
crypto: caam/qi - Change a couple IS_ERR_OR_NULL() checks to IS_ERR()

create_caam_req_fq() doesn't return NULL pointers so there is no need to
check.  The NULL checks are problematic because it's hard to say how a
NULL return should be handled, so removing the checks is a nice cleanup.

Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Reviewed-by: Horia Geantă <horia.geanta@nxp.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: cavium/nitrox - Added rfc4106(gcm(aes)) cipher support
Nagadheeraj Rottela [Thu, 28 Mar 2019 13:15:49 +0000 (13:15 +0000)]
crypto: cavium/nitrox - Added rfc4106(gcm(aes)) cipher support

Added rfc4106(gcm(aes)) cipher.

Signed-off-by: Nagadheeraj Rottela <rnagadheeraj@marvell.com>
Reviewed-by: Srikanth Jampala <jsrikanth@marvell.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: caam - limit AXI pipeline to a depth of 1
Iuliana Prodan [Fri, 22 Mar 2019 13:39:28 +0000 (15:39 +0200)]
crypto: caam - limit AXI pipeline to a depth of 1

Some i.MX6 devices (imx6D, imx6Q, imx6DL, imx6S, imx6DP and imx6DQ) have
an issue wherein AXI bus transactions may not occur in the correct order.
This isn't a problem running single descriptors, but can be if running
multiple concurrent descriptors. Reworking the CAAM driver to throttle
to single requests is impractical, so this patch limits the AXI pipeline
to a depth of one (from a default of 4) to preclude this situation from
occurring.
This patch applies to known affected platforms.

Signed-off-by: Radu Solea <radu.solea@nxp.com>
Signed-off-by: Iuliana Prodan <iuliana.prodan@nxp.com>
Reviewed-by: Horia Geantă <horia.geanta@nxp.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: caam/jr - Remove extra memory barrier during job ring enqueue
Vakul Garg [Fri, 22 Mar 2019 02:00:37 +0000 (02:00 +0000)]
crypto: caam/jr - Remove extra memory barrier during job ring enqueue

In caam_jr_enqueue(), a write barrier is needed to order stores to job
ring slot before declaring addition of new job into input job ring.
The register write is done using wr_reg32() which internally uses
iowrite32() for write operation. The api iowrite32() issues a write
barrier before issuing write operation. Therefore, the wmb() preceding
wr_reg32() can be safely removed.

Signed-off-by: Vakul Garg <vakul.garg@nxp.com>
Reviewed-by: Horia Geanta <horia.geanta@nxp.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: caam/jr - Removed redundant vars from job ring private data
Vakul Garg [Fri, 22 Mar 2019 02:00:35 +0000 (02:00 +0000)]
crypto: caam/jr - Removed redundant vars from job ring private data

For each job ring, the variable 'ringsize' is initialised but never
used. Similarly variables 'inp_ring_write_index' and 'head' always track
the same value and instead of 'inp_ring_write_index', caam_jr_enqueue()
can use 'head' itself. Both these variables have been removed.

Signed-off-by: Vakul Garg <vakul.garg@nxp.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: caam/jr - Remove spinlock for output job ring
Vakul Garg [Fri, 22 Mar 2019 02:00:34 +0000 (02:00 +0000)]
crypto: caam/jr - Remove spinlock for output job ring

For each job ring pair, the output ring is processed exactly by one cpu
at a time under a tasklet context (one per ring). Therefore, there is no
need to protect a job ring's access & its private data structure using a
lock. Hence the lock can be removed.

Signed-off-by: Vakul Garg <vakul.garg@nxp.com>
Reviewed-by: Horia Geanta <horia.geanta@nxp.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: vmx - Make p8_init and p8_exit static
YueHaibing [Thu, 21 Mar 2019 15:11:48 +0000 (23:11 +0800)]
crypto: vmx - Make p8_init and p8_exit static

Fix sparse warnings:

drivers/crypto/vmx/vmx.c:44:12: warning:
 symbol 'p8_init' was not declared. Should it be static?
drivers/crypto/vmx/vmx.c:70:13: warning:
 symbol 'p8_exit' was not declared. Should it be static?

Signed-off-by: YueHaibing <yuehaibing@huawei.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: fips - Grammar s/options/option/, s/to/the/
Geert Uytterhoeven [Wed, 20 Mar 2019 10:41:03 +0000 (11:41 +0100)]
crypto: fips - Grammar s/options/option/, s/to/the/

Fixes: ccb778e1841ce04b ("crypto: api - Add fips_enable flag")
Signed-off-by: Geert Uytterhoeven <geert+renesas@glider.be>
Reviewed-by: Mukesh Ojha <mojha@codeaurora.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: cavium - Make cptvf_device_init static
YueHaibing [Tue, 19 Mar 2019 13:59:32 +0000 (21:59 +0800)]
crypto: cavium - Make cptvf_device_init static

Fix sparse warning:

drivers/crypto/cavium/cpt/cptvf_main.c:644:6:
 warning: symbol 'cptvf_device_init' was not declared. Should it be static?

Signed-off-by: YueHaibing <yuehaibing@huawei.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: bcm - remove unused array tag_to_hash_idx
YueHaibing [Tue, 19 Mar 2019 13:57:49 +0000 (21:57 +0800)]
crypto: bcm - remove unused array tag_to_hash_idx

It's never used since introduction in commit
9d12ba86f818 ("crypto: brcm - Add Broadcom SPU driver")

Signed-off-by: YueHaibing <yuehaibing@huawei.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: zip - Make some functions static
YueHaibing [Tue, 19 Mar 2019 13:55:07 +0000 (21:55 +0800)]
crypto: zip - Make some functions static

Fix following sparse warnings:

drivers/crypto/cavium/zip/zip_crypto.c:72:5: warning: symbol 'zip_ctx_init' was not declared. Should it be static?
drivers/crypto/cavium/zip/zip_crypto.c:110:6: warning: symbol 'zip_ctx_exit' was not declared. Should it be static?
drivers/crypto/cavium/zip/zip_crypto.c:122:5: warning: symbol 'zip_compress' was not declared. Should it be static?
drivers/crypto/cavium/zip/zip_crypto.c:158:5: warning: symbol 'zip_decompress' was not declared. Should it be static?

Signed-off-by: YueHaibing <yuehaibing@huawei.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: ccp - Make ccp_register_rsa_alg static
YueHaibing [Tue, 19 Mar 2019 13:50:35 +0000 (21:50 +0800)]
crypto: ccp - Make ccp_register_rsa_alg static

Fix sparse warning:

drivers/crypto/ccp/ccp-crypto-rsa.c:251:5:
 warning: symbol 'ccp_register_rsa_alg' was not declared. Should it be static?

Signed-off-by: YueHaibing <yuehaibing@huawei.com>
Acked-by: Gary R Hook <gary.hook@amd.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: cavium - Make some functions static
YueHaibing [Tue, 19 Mar 2019 13:43:48 +0000 (21:43 +0800)]
crypto: cavium - Make some functions static

Fix sparse warnings:

drivers/crypto/cavium/cpt/cptvf_reqmanager.c:226:5: warning: symbol 'send_cpt_command' was not declared. Should it be static?
drivers/crypto/cavium/cpt/cptvf_reqmanager.c:273:6: warning: symbol 'do_request_cleanup' was not declared. Should it be static?
drivers/crypto/cavium/cpt/cptvf_reqmanager.c:319:6: warning: symbol 'do_post_process' was not declared. Should it be static?

Signed-off-by: YueHaibing <yuehaibing@huawei.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: cavium - remove unused fucntions
YueHaibing [Tue, 19 Mar 2019 13:37:00 +0000 (21:37 +0800)]
crypto: cavium - remove unused fucntions

cptvf_mbox_send_ack and cptvf_mbox_send_nack are never
used since introdution in commit c694b233295b ("crypto: cavium
- Add the Virtual Function driver for CPT")

Signed-off-by: YueHaibing <yuehaibing@huawei.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: Kconfig - fix typos AEGSI -> AEGIS
Ondrej Mosnacek [Fri, 15 Mar 2019 07:47:25 +0000 (08:47 +0100)]
crypto: Kconfig - fix typos AEGSI -> AEGIS

Spotted while reviewind patches from Eric Biggers.

Signed-off-by: Ondrej Mosnacek <omosnace@redhat.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: salsa20-generic - use crypto_xor_cpy()
Eric Biggers [Fri, 15 Mar 2019 04:21:28 +0000 (21:21 -0700)]
crypto: salsa20-generic - use crypto_xor_cpy()

In salsa20_docrypt(), use crypto_xor_cpy() instead of crypto_xor().
This avoids having to memcpy() the src buffer to the dst buffer.

Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: chacha-generic - use crypto_xor_cpy()
Eric Biggers [Fri, 15 Mar 2019 04:20:36 +0000 (21:20 -0700)]
crypto: chacha-generic - use crypto_xor_cpy()

In chacha_docrypt(), use crypto_xor_cpy() instead of crypto_xor().
This avoids having to memcpy() the src buffer to the dst buffer.

Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: vmx - fix copy-paste error in CTR mode
Daniel Axtens [Fri, 15 Mar 2019 02:09:01 +0000 (13:09 +1100)]
crypto: vmx - fix copy-paste error in CTR mode

The original assembly imported from OpenSSL has two copy-paste
errors in handling CTR mode. When dealing with a 2 or 3 block tail,
the code branches to the CBC decryption exit path, rather than to
the CTR exit path.

This leads to corruption of the IV, which leads to subsequent blocks
being corrupted.

This can be detected with libkcapi test suite, which is available at
https://github.com/smuellerDD/libkcapi

Reported-by: Ondrej Mosnáček <omosnacek@gmail.com>
Fixes: 5c380d623ed3 ("crypto: vmx - Add support for VMS instructions by ASM")
Cc: stable@vger.kernel.org
Signed-off-by: Daniel Axtens <dja@axtens.net>
Tested-by: Michael Ellerman <mpe@ellerman.id.au>
Tested-by: Ondrej Mosnacek <omosnacek@gmail.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: ccree - reduce kernel stack usage with clang
Arnd Bergmann [Thu, 14 Mar 2019 09:09:44 +0000 (10:09 +0100)]
crypto: ccree - reduce kernel stack usage with clang

Building with clang for a 32-bit architecture runs over the stack
frame limit in the setkey function:

drivers/crypto/ccree/cc_cipher.c:318:12: error: stack frame size of 1152 bytes in function 'cc_cipher_setkey' [-Werror,-Wframe-larger-than=]

The problem is that there are two large variables: the temporary
'tmp' array and the SHASH_DESC_ON_STACK() declaration. Moving
the first into the block in which it is used reduces the
total frame size to 768 bytes, which seems more reasonable
and is under the warning limit.

Fixes: 63ee04c8b491 ("crypto: ccree - add skcipher support")
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Acked-By: Gilad Ben-Yossef <gilad@benyossef.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: testmgr - test the !may_use_simd() fallback code
Eric Biggers [Wed, 13 Mar 2019 05:12:52 +0000 (22:12 -0700)]
crypto: testmgr - test the !may_use_simd() fallback code

All crypto API algorithms are supposed to support the case where they
are called in a context where SIMD instructions are unusable, e.g. IRQ
context on some architectures.  However, this isn't tested for by the
self-tests, causing bugs to go undetected.

Now that all algorithms have been converted to use crypto_simd_usable(),
update the self-tests to test the no-SIMD case.  First, a bool
testvec_config::nosimd is added.  When set, the crypto operation is
executed with preemption disabled and with crypto_simd_usable() mocked
out to return false on the current CPU.

A bool test_sg_division::nosimd is also added.  For hash algorithms it's
honored by the corresponding ->update().  By setting just a subset of
these bools, the case where some ->update()s are done in SIMD context
and some are done in no-SIMD context is also tested.

These bools are then randomly set by generate_random_testvec_config().

For now, all no-SIMD testing is limited to the extra crypto self-tests,
because it might be a bit too invasive for the regular self-tests.
But this could be changed later.

This has already found bugs in the arm64 AES-GCM and ChaCha algorithms.
This would have found some past bugs as well.

Signed-off-by: Eric Biggers <ebiggers@google.com>
Acked-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: simd - convert to use crypto_simd_usable()
Eric Biggers [Wed, 13 Mar 2019 05:12:51 +0000 (22:12 -0700)]
crypto: simd - convert to use crypto_simd_usable()

Replace all calls to may_use_simd() in the shared SIMD helpers with
crypto_simd_usable(), in order to allow testing the no-SIMD code paths.

Signed-off-by: Eric Biggers <ebiggers@google.com>
Reviewed-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: arm64 - convert to use crypto_simd_usable()
Eric Biggers [Wed, 13 Mar 2019 05:12:50 +0000 (22:12 -0700)]
crypto: arm64 - convert to use crypto_simd_usable()

Replace all calls to may_use_simd() in the arm64 crypto code with
crypto_simd_usable(), in order to allow testing the no-SIMD code paths.

Signed-off-by: Eric Biggers <ebiggers@google.com>
Reviewed-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: arm - convert to use crypto_simd_usable()
Eric Biggers [Wed, 13 Mar 2019 05:12:49 +0000 (22:12 -0700)]
crypto: arm - convert to use crypto_simd_usable()

Replace all calls to may_use_simd() in the arm crypto code with
crypto_simd_usable(), in order to allow testing the no-SIMD code paths.

Signed-off-by: Eric Biggers <ebiggers@google.com>
Reviewed-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: x86 - convert to use crypto_simd_usable()
Eric Biggers [Wed, 13 Mar 2019 05:12:48 +0000 (22:12 -0700)]
crypto: x86 - convert to use crypto_simd_usable()

Replace all calls to irq_fpu_usable() in the x86 crypto code with
crypto_simd_usable(), in order to allow testing the no-SIMD code paths.

Signed-off-by: Eric Biggers <ebiggers@google.com>
Reviewed-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: simd,testmgr - introduce crypto_simd_usable()
Eric Biggers [Wed, 13 Mar 2019 05:12:47 +0000 (22:12 -0700)]
crypto: simd,testmgr - introduce crypto_simd_usable()

So that the no-SIMD fallback code can be tested by the crypto
self-tests, add a macro crypto_simd_usable() which wraps may_use_simd(),
but also returns false if the crypto self-tests have set a per-CPU bool
to disable SIMD in crypto code on the current CPU.

Signed-off-by: Eric Biggers <ebiggers@google.com>
Reviewed-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: arm64/gcm-aes-ce - fix no-NEON fallback code
Eric Biggers [Wed, 13 Mar 2019 05:12:46 +0000 (22:12 -0700)]
crypto: arm64/gcm-aes-ce - fix no-NEON fallback code

The arm64 gcm-aes-ce algorithm is failing the extra crypto self-tests
following my patches to test the !may_use_simd() code paths, which
previously were untested.  The problem is that in the !may_use_simd()
case, an odd number of AES blocks can be processed within each step of
the skcipher_walk.  However, the skcipher_walk is being done with a
"stride" of 2 blocks and is advanced by an even number of blocks after
each step.  This causes the encryption to produce the wrong ciphertext
and authentication tag, and causes the decryption to incorrectly fail.

Fix it by only processing an even number of blocks per step.

Fixes: c2b24c36e0a3 ("crypto: arm64/aes-gcm-ce - fix scatterwalk API violation")
Fixes: 71e52c278c54 ("crypto: arm64/aes-ce-gcm - operate on two input blocks at a time")
Cc: <stable@vger.kernel.org> # v4.19+
Signed-off-by: Eric Biggers <ebiggers@google.com>
Reviewed-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: chacha-generic - fix use as arm64 no-NEON fallback
Eric Biggers [Wed, 13 Mar 2019 05:12:45 +0000 (22:12 -0700)]
crypto: chacha-generic - fix use as arm64 no-NEON fallback

The arm64 implementations of ChaCha and XChaCha are failing the extra
crypto self-tests following my patches to test the !may_use_simd() code
paths, which previously were untested.  The problem is as follows:

When !may_use_simd(), the arm64 NEON implementations fall back to the
generic implementation, which uses the skcipher_walk API to iterate
through the src/dst scatterlists.  Due to how the skcipher_walk API
works, walk.stride is set from the skcipher_alg actually being used,
which in this case is the arm64 NEON algorithm.  Thus walk.stride is
5*CHACHA_BLOCK_SIZE, not CHACHA_BLOCK_SIZE.

This unnecessarily large stride shouldn't cause an actual problem.
However, the generic implementation computes round_down(nbytes,
walk.stride).  round_down() assumes the round amount is a power of 2,
which 5*CHACHA_BLOCK_SIZE is not, so it gives the wrong result.

This causes the following case in skcipher_walk_done() to be hit,
causing a WARN() and failing the encryption operation:

if (WARN_ON(err)) {
/* unexpected case; didn't process all bytes */
err = -EINVAL;
goto finish;
}

Fix it by rounding down to CHACHA_BLOCK_SIZE instead of walk.stride.

(Or we could replace round_down() with rounddown(), but that would add a
slow division operation every time, which I think we should avoid.)

Fixes: 2fe55987b262 ("crypto: arm64/chacha - use combined SIMD/ALU routine for more speed")
Cc: <stable@vger.kernel.org> # v5.0+
Signed-off-by: Eric Biggers <ebiggers@google.com>
Reviewed-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agohwrng: omap - Set default quality
Rouven Czerwinski [Mon, 11 Mar 2019 10:58:57 +0000 (11:58 +0100)]
hwrng: omap - Set default quality

Newer combinations of the glibc, kernel and openssh can result in long initial
startup times on OMAP devices:

[    6.671425] systemd-rc-once[102]: Creating ED25519 key; this may take some time ...
[  142.652491] systemd-rc-once[102]: Creating ED25519 key; done.

due to the blocking getrandom(2) system call:

[  142.610335] random: crng init done

Set the quality level for the omap hwrng driver allowing the kernel to use the
hwrng as an entropy source at boot.

Signed-off-by: Rouven Czerwinski <r.czerwinski@pengutronix.de>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: testmgr - remove workaround for AEADs that modify aead_request
Eric Biggers [Sun, 10 Mar 2019 19:00:58 +0000 (12:00 -0700)]
crypto: testmgr - remove workaround for AEADs that modify aead_request

Now that all AEAD algorithms (that I have the hardware to test, at
least) have been fixed to not modify the user-provided aead_request,
remove the workaround from testmgr that reset aead_request::tfm after
each AEAD encryption/decryption.

Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: x86/morus1280 - convert to use AEAD SIMD helpers
Eric Biggers [Sun, 10 Mar 2019 19:00:57 +0000 (12:00 -0700)]
crypto: x86/morus1280 - convert to use AEAD SIMD helpers

Convert the x86 implementations of MORUS-1280 to use the AEAD SIMD
helpers, rather than hand-rolling the same functionality.  This
simplifies the code and also fixes the bug where the user-provided
aead_request is modified.

Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: x86/morus640 - convert to use AEAD SIMD helpers
Eric Biggers [Sun, 10 Mar 2019 19:00:56 +0000 (12:00 -0700)]
crypto: x86/morus640 - convert to use AEAD SIMD helpers

Convert the x86 implementation of MORUS-640 to use the AEAD SIMD
helpers, rather than hand-rolling the same functionality.  This
simplifies the code and also fixes the bug where the user-provided
aead_request is modified.

Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: x86/aegis256 - convert to use AEAD SIMD helpers
Eric Biggers [Sun, 10 Mar 2019 19:00:55 +0000 (12:00 -0700)]
crypto: x86/aegis256 - convert to use AEAD SIMD helpers

Convert the x86 implementation of AEGIS-256 to use the AEAD SIMD
helpers, rather than hand-rolling the same functionality.  This
simplifies the code and also fixes the bug where the user-provided
aead_request is modified.

Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: x86/aegis128l - convert to use AEAD SIMD helpers
Eric Biggers [Sun, 10 Mar 2019 19:00:54 +0000 (12:00 -0700)]
crypto: x86/aegis128l - convert to use AEAD SIMD helpers

Convert the x86 implementation of AEGIS-128L to use the AEAD SIMD
helpers, rather than hand-rolling the same functionality.  This
simplifies the code and also fixes the bug where the user-provided
aead_request is modified.

Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: x86/aegis128 - convert to use AEAD SIMD helpers
Eric Biggers [Sun, 10 Mar 2019 19:00:53 +0000 (12:00 -0700)]
crypto: x86/aegis128 - convert to use AEAD SIMD helpers

Convert the x86 implementation of AEGIS-128 to use the AEAD SIMD
helpers, rather than hand-rolling the same functionality.  This
simplifies the code and also fixes the bug where the user-provided
aead_request is modified.

Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: x86/aesni - convert to use AEAD SIMD helpers
Eric Biggers [Sun, 10 Mar 2019 19:00:52 +0000 (12:00 -0700)]
crypto: x86/aesni - convert to use AEAD SIMD helpers

Convert the AES-NI implementations of "gcm(aes)" and "rfc4106(gcm(aes))"
to use the AEAD SIMD helpers, rather than hand-rolling the same
functionality.  This simplifies the code and also fixes the bug where
the user-provided aead_request is modified.

Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: x86/aesni - convert to use skcipher SIMD bulk registration
Eric Biggers [Sun, 10 Mar 2019 19:00:51 +0000 (12:00 -0700)]
crypto: x86/aesni - convert to use skcipher SIMD bulk registration

Convert the AES-NI glue code to use simd_register_skciphers_compat() to
create SIMD wrappers for all the internal skcipher algorithms at once,
rather than wrapping each one individually.  This simplifies the code.

Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: simd - support wrapping AEAD algorithms
Eric Biggers [Sun, 10 Mar 2019 19:00:50 +0000 (12:00 -0700)]
crypto: simd - support wrapping AEAD algorithms

Update the crypto_simd module to support wrapping AEAD algorithms.
Previously it only supported skciphers.  The code for each is similar.

I'll be converting the x86 implementations of AES-GCM, AEGIS, and MORUS
to use this.  Currently they each independently implement the same
functionality.  This will not only simplify the code, but it will also
fix the bug detected by the improved self-tests: the user-provided
aead_request is modified.  This is because these algorithms currently
reuse the original request, whereas the crypto_simd helpers build a new
request in the original request's context.

Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: caam/jr - optimize job ring enqueue and dequeue operations
Vakul Garg [Tue, 5 Mar 2019 18:03:14 +0000 (18:03 +0000)]
crypto: caam/jr - optimize job ring enqueue and dequeue operations

Instead of reading job ring's occupancy registers for every req/rsp
enqueued/dequeued respectively, we read these registers once and store
them in memory. After completing a job enqueue/dequeue, we decrement
these values. When these values become zero, we refresh the snapshot of
job ring's occupancy registers. This eliminates need of expensive device
register read operations for every job enqueued and dequeued and hence
makes caam_jr_enqueue() and caam_jr_dequeue() faster. The performance of
kernel ipsec improved by about 6% on ls1028 (for frame size 408 bytes).

Signed-off-by: Vakul Garg <vakul.garg@nxp.com>
Reviewed-by: Horia Geantă <horia.geanta@nxp.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agoLinux 5.1-rc1
Linus Torvalds [Sun, 17 Mar 2019 21:22:26 +0000 (14:22 -0700)]
Linux 5.1-rc1