platform/kernel/linux-rpi.git
2 years agocrypto: dh - allow for passing NULL to the ffdheXYZ(dh)s' ->set_secret()
Nicolai Stange [Mon, 21 Feb 2022 12:10:57 +0000 (13:10 +0100)]
crypto: dh - allow for passing NULL to the ffdheXYZ(dh)s' ->set_secret()

Ephemeral key generation can be requested from any of the ffdheXYZ(dh)
variants' common ->set_secret() by passing it an (encoded) struct dh
with the key parameter being unset, i.e. with ->key_size == 0. As the
whole purpose of the ffdheXYZ(dh) templates is to fill in the group
parameters as appropriate, they expect ->p and ->g to be unset in any
input struct dh as well. This means that a user would have to encode an
all-zeroes struct dh instance via crypto_dh_encode_key() when requesting
ephemeral key generation from a ffdheXYZ(dh) instance, which is kind of
pointless.

Make dh_safe_prime_set_secret() to decode a struct dh from the supplied
buffer only if the latter is non-NULL and initialize it with all zeroes
otherwise.

That is, it is now possible to call

  crypto_kpp_set_secret(tfm, NULL, 0);

on any ffdheXYZ(dh) tfm for requesting ephemeral key generation.

Signed-off-by: Nicolai Stange <nstange@suse.de>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2 years agocrypto: testmgr - add keygen tests for ffdheXYZ(dh) templates
Nicolai Stange [Mon, 21 Feb 2022 12:10:56 +0000 (13:10 +0100)]
crypto: testmgr - add keygen tests for ffdheXYZ(dh) templates

Now that the ffdheXYZ(dh) templates support ephemeral key generation, add
->keygen = 1 TVs for each of them to the testmgr.c.

In order to facilitate string merging by the compiler, set party B's secret
and public keys to the ones specified for party A in the respective
existing known answer test. With GCC 7.5 on x86_64, this leads to an
increase of testmgr.o size by less than half a kB.

Signed-off-by: Nicolai Stange <nstange@suse.de>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2 years agocrypto: dh - implement private key generation primitive for ffdheXYZ(dh)
Nicolai Stange [Mon, 21 Feb 2022 12:10:55 +0000 (13:10 +0100)]
crypto: dh - implement private key generation primitive for ffdheXYZ(dh)

The support for NVME in-band authentication currently in the works ([1])
needs to generate ephemeral DH keys for use with the RFC 7919 safe-prime
FFDHE groups.

In analogy to ECDH and its ecc_gen_privkey(), implement a
dh_safe_prime_gen_privkey() and invoke it from the ffdheXYZ(dh) templates'
common ->set_secret(), i.e. dh_safe_prime_set_secret(), in case the input
->key_size is zero.

As the RFC 7919 FFDHE groups are classified as approved safe-prime groups
by SP800-56Arev3, it's worthwhile to make the new
dh_safe_prime_gen_privkey() to follow the approach specified in
SP800-56Arev3, sec. 5.6.1.1.3 ("Key-Pair Generation Using Extra Random
Bits") in order to achieve conformance.

SP800-56Arev3 specifies a lower as well as an upper bound on the generated
key's length:
- it must be >= two times the maximum supported security strength of
  the group in question and
- it must be <= the length of the domain parameter Q.

For any safe-prime group Q = (P - 1)/2 by definition and the individual
maximum supported security strengths as specified by SP800-56Arev3 have
been made available as part of the FFDHE dh_safe_prime definitions
introduced with a previous patch. Make dh_safe_prime_gen_privkey() pick
twice the maximum supported strength rounded up to the next power of two
for the output key size. This choice respects both, the lower and upper
bounds given by SP800-90Arev3 for any of the approved safe-prime groups and
is also in line with the NVME base spec 2.0, which requires the key size to
be >= 256bits.

[1] https://lore.kernel.org/r/20211202152358.60116-1-hare@suse.de

Signed-off-by: Nicolai Stange <nstange@suse.de>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2 years agocrypto: testmgr - add known answer tests for ffdheXYZ(dh) templates
Nicolai Stange [Mon, 21 Feb 2022 12:10:54 +0000 (13:10 +0100)]
crypto: testmgr - add known answer tests for ffdheXYZ(dh) templates

Add known answer tests for the ffdhe2048(dh), ffdhe3072(dh), ffdhe4096(dh),
ffdhe6144(dh) and ffdhe8192(dh) templates introduced with the previous
patch to the testmgr. All TVs have been generated with OpenSSL.

Signed-off-by: Nicolai Stange <nstange@suse.de>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2 years agocrypto: dh - implement ffdheXYZ(dh) templates
Nicolai Stange [Mon, 21 Feb 2022 12:10:53 +0000 (13:10 +0100)]
crypto: dh - implement ffdheXYZ(dh) templates

Current work on NVME in-band authentication support ([1]) needs to invoke
DH with the FFDHE safe-prime group parameters specified in RFC 7919.

Introduce a new CRYPTO_DH_RFC7919_GROUPS Kconfig option. If enabled, make
dh_generic register a couple of ffdheXYZ(dh) templates, one for each group:
ffdhe2048(dh), ffdhe3072(dh), ffdhe4096(dh), ffdhe6144(dh) and
ffdhe8192(dh). Their respective ->set_secret() expects a (serialized)
struct dh, just like the underlying "dh" implementation does, but with the
P and G values unset so that the safe-prime constants for the given group
can be filled in by the wrapping template.

Internally, a struct dh_safe_prime instance is being defined for each of
the ffdheXYZ(dh) templates as appropriate. In order to prepare for future
key generation, fill in the maximum security strength values as specified
by SP800-56Arev3 on the go, even though they're not needed at this point
yet.

Implement the respective ffdheXYZ(dh) crypto_template's ->create() by
simply forwarding any calls to the __dh_safe_prime_create() helper
introduced with the previous commit, passing the associated dh_safe_prime
in addition to the received ->create() arguments.

[1] https://lore.kernel.org/r/20211202152358.60116-1-hare@suse.de

Signed-off-by: Nicolai Stange <nstange@suse.de>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2 years agocrypto: dh - introduce common code for built-in safe-prime group support
Nicolai Stange [Mon, 21 Feb 2022 12:10:52 +0000 (13:10 +0100)]
crypto: dh - introduce common code for built-in safe-prime group support

Recent work on NVME in-band authentication support ([1]) needs to invoke
the "dh" KPP with the FFDHE safe-prime group parameters as specified in
RFC 7919 and generate ephemeral keys suitable for the respective group. By
coincidence, the requirements from NIST SP800-56Arev3,
sec. 5.5.2 ("Assurance of Domain-Parameter Validity") basically boil down
to disallowing any group parameters not among the approved safe-prime
groups specified in either RFC 7919 or RFC 3526 in FIPS mode. Furthermore,
SP800-56Arev3 specifies the respective security strength for each of the
approved safe-prime groups, which has a direct impact on the minimum key
lengths.

In this light, it's desirable to introduce built-in support for the
RFC 7919 safe-prime groups to the kernel's DH implementation, provide a
SP800-56Arev3 conforming key generation primitive for those and render
non-approved group parameters unusable in FIPS mode on the way.

As suggested ([2]) in the course of discussion to previous iterations of
this patchset, the built-in support for ffdhe groups would be best made
available in the form of templates wrapping the existing "dh"
implementation, one for each group specified by RFC 7919: ffdhe2048(dh),
ffdhe3072(dh), ffdhe4096(dh), ffdhe6144(dh) and ffdhe8192(dh). As these
templates differ only in the safe-prime constants they'd configure the
inner "dh" transforms with, they can share almost all of their
"dh"-wrapping template implementation code.

Introduce this common code to dh_generic. The actual dump of the RFC 7919
safe-prime constants will be deferred to the next patch in order to
facilitate review. The ephemeral key generation primitive mentioned above
likewise deserves a patch on its own, as does the mechanism by which
unapproved groups are rendered unusable in FIPS mode.

Define a struct dh_safe_prime container for specifying the individual
templates' associated safe-prime group constants. All ffdheXYZ(dh) template
instances will store a pointer to such a dh_safe_prime in their context
areas each. Implement the common __dh_safe_prime_create() template
instantiation helper. The intention is that the individual ffdheXYZ(dh)
crypto_templates' ->create() implementations will simply forward any calls
to __dh_safe_prime_create(), passing a suitable dh_safe_prime in addition
to the received ->create() arguments. __dh_safe_prime_create() would then
create and register a kpp_instance as appropriate, storing the given
dh_safe_prime pointer alongside a crypto_kpp_spawn for the inner "dh"
kpp_alg in the context area.

As the ffdheXYZ(dh) kpp_instances are supposed to act as proxies to the
inner "dh" kpp_alg, make each of their associated crypto_kpp transforms to
in turn own an inner "dh" transform, a pointer to which gets stored in the
context area. Setup and teardown are getting handled from the outer
->init_tfm() and ->exit_tfm() respectively.

In order to achieve the overall goal and let the ffdheXYZ(dh) kpp_instances
configure the inner "dh" transforms with the respective group parameters,
make their common ->set_secret(), the new dh_safe_prime_set_secret(), fill
in the P and G values before forwarding the call to the inner "dh"'s
->set_secret(). Note that the outer ->set_secret() can obtain the P value
associated with the given ffdheXYZ(dh) kpp_instance by means of the
dh_safe_prime referenced from the latter's context. The value of G OTOH
always equals constant 2 for the safe-prime groups.

Finally, make the remaining two kpp_alg primitives both operating on
kpp_requests, i.e. ->generate_public_key() and ->compute_shared_secret(),
to merely forward any request to the inner "dh" implementation. However, a
kpp_request instance received from the outside cannot get simply passed
on as-is, because its associated transform (crypto_kpp_reqtfm()) will have
been set to the outer ffdheXYZ(dh) one. In order to handle this, reserve
some space in the outer ffdheXYZ(dh) kpp_requests' context areas for in
turn storing an inner kpp_request suitable for "dh" each. Make the outer
->generate_public_key() and ->compute_shared_secret() respectively to setup
this inner kpp_request by means of the new dh_safe_prime_prepare_dh_req()
helper before handing it over to the "dh" implementation for further
processing. dh_safe_prime_prepare_dh_req() basically copies the outer
kpp_request received from the outside over to the inner one, but installs
the inner transform and its own ->complete() proxy callback therein. This
completion callback, the new dh_safe_prime_complete_req(), doesn't do
anything beyond completing the outer request. Note that there exist some
examples in crypto/, which would simply install the completion handler
from the outer request at the inner one in similar setups, e.g. seqiv.
However, this would mean that the user-provided completion handler won't
get called with the address of the outer kpp_request initially submitted
and the handler might not be prepared for this. Users could certainly work
around this by setting the callback ->data properly, but IMO it's cleaner
this way. Furthermore, it might make sense to extend
dh_safe_prime_complete_req() in the future and move e.g. those
post-computation FIPS checks from the generic "dh" implementation to the
ffdheXYZ(dh) templates.

[1] https://lore.kernel.org/r/20211202152358.60116-1-hare@suse.de
[2] https://lore.kernel.org/r/20211217055227.GA20698@gondor.apana.org.au

Signed-off-by: Nicolai Stange <nstange@suse.de>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2 years agocrypto: dh - split out deserialization code from crypto_dh_decode()
Nicolai Stange [Mon, 21 Feb 2022 12:10:51 +0000 (13:10 +0100)]
crypto: dh - split out deserialization code from crypto_dh_decode()

A subsequent commit will introduce "dh" wrapping templates of the form
"ffdhe2048(dh)", "ffdhe3072(dh)" and so on in order to provide built-in
support for the well-known safe-prime ffdhe group parameters specified in
RFC 7919.

Those templates' ->set_secret() will wrap the inner "dh" implementation's
->set_secret() and set the ->p and ->g group parameters as appropriate on
the way inwards. More specifically,
- A ffdheXYZ(dh) user would call crypto_dh_encode() on a struct dh instance
  having ->p == ->g == NULL as well as ->p_size == ->g_size == 0 and pass
  the resulting buffer to the outer ->set_secret().
- This outer ->set_secret() would then decode the struct dh via
  crypto_dh_decode_key(), set ->p, ->g, ->p_size as well as ->g_size as
  appropriate for the group in question and encode the struct dh again
  before passing it further down to the inner "dh"'s ->set_secret().

The problem is that crypto_dh_decode_key() implements some basic checks
which would reject parameter sets with ->p_size == 0 and thus, the ffdheXYZ
templates' ->set_secret() cannot use it as-is for decoding the passed
buffer. As the inner "dh"'s ->set_secret() will eventually conduct said
checks on the final parameter set anyway, the outer ->set_secret() really
only needs the decoding functionality.

Split out the pure struct dh decoding part from crypto_dh_decode_key() into
the new __crypto_dh_decode_key().

__crypto_dh_decode_key() gets defined in crypto/dh_helper.c, but will have
to get called from crypto/dh.c and thus, its declaration must be somehow
made available to the latter. Strictly speaking, __crypto_dh_decode_key()
is internal to the dh_generic module, yet it would be a bit over the top
to introduce a new header like e.g. include/crypto/internal/dh.h
containing just a single prototype. Add the __crypto_dh_decode_key()
declaration to include/crypto/dh.h instead.

Provide a proper kernel-doc annotation, even though
__crypto_dh_decode_key() is purposedly not on the function list specified
in Documentation/crypto/api-kpp.rst.

Signed-off-by: Nicolai Stange <nstange@suse.de>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2 years agocrypto: dh - constify struct dh's pointer members
Nicolai Stange [Mon, 21 Feb 2022 12:10:50 +0000 (13:10 +0100)]
crypto: dh - constify struct dh's pointer members

struct dh contains several pointer members corresponding to DH parameters:
->key, ->p and ->g. A subsequent commit will introduce "dh" wrapping
templates of the form "ffdhe2048(dh)", "ffdhe3072(dh)" and so on in order
to provide built-in support for the well-known safe-prime ffdhe group
parameters specified in RFC 7919. These templates will need to set the
group parameter related members of the (serialized) struct dh instance
passed to the inner "dh" kpp_alg instance, i.e. ->p and ->g, to some
constant, static storage arrays.

Turn the struct dh pointer members' types into "pointer to const" in
preparation for this.

Signed-off-by: Nicolai Stange <nstange@suse.de>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2 years agocrypto: dh - remove struct dh's ->q member
Nicolai Stange [Mon, 21 Feb 2022 12:10:49 +0000 (13:10 +0100)]
crypto: dh - remove struct dh's ->q member

The only current user of the DH KPP algorithm, the
keyctl(KEYCTL_DH_COMPUTE) syscall, doesn't set the domain parameter ->q
in struct dh. Remove it and any associated (de)serialization code in
crypto_dh_encode_key() and crypto_dh_decode_key. Adjust the encoded
->secret values in testmgr's DH test vectors accordingly.

Note that the dh-generic implementation would have initialized its
struct dh_ctx's ->q from the decoded struct dh's ->q, if present. If this
struct dh_ctx's ->q would ever have been non-NULL, it would have enabled a
full key validation as specified in NIST SP800-56A in dh_is_pubkey_valid().
However, as outlined above, ->q is always NULL in practice and the full key
validation code is effectively dead. A later patch will make
dh_is_pubkey_valid() to calculate Q from P on the fly, if possible, so
don't remove struct dh_ctx's ->q now, but leave it there until that has
happened.

Signed-off-by: Nicolai Stange <nstange@suse.de>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2 years agocrypto: kpp - provide support for KPP spawns
Nicolai Stange [Mon, 21 Feb 2022 12:10:48 +0000 (13:10 +0100)]
crypto: kpp - provide support for KPP spawns

The upcoming support for the RFC 7919 ffdhe group parameters will be
made available in the form of templates like "ffdhe2048(dh)",
"ffdhe3072(dh)" and so on. Template instantiations thereof would wrap the
inner "dh" kpp_alg and also provide kpp_alg services to the outside again.

The primitves needed for providing kpp_alg services from template instances
have been introduced with the previous patch. Continue this work now and
implement everything needed for enabling template instances to make use
of inner KPP algorithms like "dh".

More specifically, define a struct crypto_kpp_spawn in close analogy to
crypto_skcipher_spawn, crypto_shash_spawn and alike. Implement a
crypto_grab_kpp() and crypto_drop_kpp() pair for binding such a spawn to
some inner kpp_alg and for releasing it respectively. Template
implementations can instantiate transforms from the underlying kpp_alg by
means of the new crypto_spawn_kpp(). Finally, provide the
crypto_spawn_kpp_alg() helper for accessing a spawn's underlying kpp_alg
during template instantiation.

Annotate everything with proper kernel-doc comments, even though
include/crypto/internal/kpp.h is not considered for the generated docs.

Signed-off-by: Nicolai Stange <nstange@suse.de>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2 years agocrypto: kpp - provide support for KPP template instances
Nicolai Stange [Mon, 21 Feb 2022 12:10:47 +0000 (13:10 +0100)]
crypto: kpp - provide support for KPP template instances

The upcoming support for the RFC 7919 ffdhe group parameters will be
made available in the form of templates like "ffdhe2048(dh)",
"ffdhe3072(dh)" and so on. Template instantiations thereof would wrap the
inner "dh" kpp_alg and also provide kpp_alg services to the outside again.
Furthermore, it might be perhaps be desirable to provide KDF templates in
the future, which would similarly wrap an inner kpp_alg and present
themselves to the outside as another kpp_alg, transforming the shared
secret on its way out.

Introduce the bits needed for supporting KPP template instances. Everything
related to inner kpp_alg spawns potentially being held by such template
instances will be deferred to a subsequent patch in order to facilitate
review.

Define struct struct kpp_instance in close analogy to the already existing
skcipher_instance, shash_instance and alike, but wrapping a struct kpp_alg.
Implement the new kpp_register_instance() template instance registration
primitive. Provide some helper functions for
- going back and forth between a generic struct crypto_instance and the new
  struct kpp_instance,
- obtaining the instantiating kpp_instance from a crypto_kpp transform and
- for accessing a given kpp_instance's implementation specific context
  data.

Annotate everything with proper kernel-doc comments, even though
include/crypto/internal/kpp.h is not considered for the generated docs.

Signed-off-by: Nicolai Stange <nstange@suse.de>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2 years agocrypto: engine - check if BH is disabled during completion
Corentin Labbe [Mon, 21 Feb 2022 12:08:33 +0000 (12:08 +0000)]
crypto: engine - check if BH is disabled during completion

When doing iperf over ipsec with crypto hardware sun8i-ce, I hit some
spinlock recursion bug.

This is due to completion function called with enabled BH.

Add check a to detect this.

Fixes: 735d37b5424b ("crypto: engine - Introduce the block request crypto engine framework")
Signed-off-by: Corentin Labbe <clabbe@baylibre.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2 years agocrypto: gemini - call finalize with bh disabled
Corentin Labbe [Mon, 21 Feb 2022 12:08:32 +0000 (12:08 +0000)]
crypto: gemini - call finalize with bh disabled

Doing ipsec produces a spinlock recursion warning.
This is due to not disabling BH during crypto completion function.

Fixes: 46c5338db7bd45b2 ("crypto: sl3516 - Add sl3516 crypto engine")
Signed-off-by: Corentin Labbe <clabbe@baylibre.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2 years agocrypto: amlogic - call finalize with bh disabled
Corentin Labbe [Mon, 21 Feb 2022 12:08:31 +0000 (12:08 +0000)]
crypto: amlogic - call finalize with bh disabled

Doing ipsec produces a spinlock recursion warning.
This is due to not disabling BH during crypto completion function.

Fixes: 48fe583fe541 ("crypto: amlogic - Add crypto accelerator for amlogic GXL")
Signed-off-by: Corentin Labbe <clabbe@baylibre.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2 years agocrypto: sun8i-ce - call finalize with bh disabled
Corentin Labbe [Mon, 21 Feb 2022 12:08:30 +0000 (12:08 +0000)]
crypto: sun8i-ce - call finalize with bh disabled

Doing ipsec produces a spinlock recursion warning.
This is due to not disabling BH during crypto completion function.

Fixes: 06f751b61329 ("crypto: allwinner - Add sun8i-ce Crypto Engine")
Signed-off-by: Corentin Labbe <clabbe@baylibre.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2 years agocrypto: sun8i-ss - call finalize with bh disabled
Corentin Labbe [Mon, 21 Feb 2022 12:08:29 +0000 (12:08 +0000)]
crypto: sun8i-ss - call finalize with bh disabled

Doing ipsec produces a spinlock recursion warning.
This is due to not disabling BH during crypto completion function.

Fixes: f08fcced6d00 ("crypto: allwinner - Add sun8i-ss cryptographic offloader")
Signed-off-by: Corentin Labbe <clabbe@baylibre.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2 years agocrypto: cavium/zip - register algorithm only if hardware is present
Corentin Labbe [Mon, 21 Feb 2022 11:52:34 +0000 (11:52 +0000)]
crypto: cavium/zip - register algorithm only if hardware is present

On my renesas salavator-X, I saw some cavium driver failing crypto
self-tests.
But salvator does not have such hardware.
This is due to cavium/zip driver registering algorithms even if hardware
is not present.
The solution is to move algorithm registration in the probe function.
This permits also to simplify module init/exit by using
module_pci_driver.

Signed-off-by: Corentin Labbe <clabbe@baylibre.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2 years agohwrng: atmel - remove extra line
Claudiu Beznea [Mon, 21 Feb 2022 07:59:28 +0000 (09:59 +0200)]
hwrng: atmel - remove extra line

Remove extra line.

Signed-off-by: Claudiu Beznea <claudiu.beznea@microchip.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2 years agohwrng: atmel - add runtime pm support
Claudiu Beznea [Mon, 21 Feb 2022 07:59:27 +0000 (09:59 +0200)]
hwrng: atmel - add runtime pm support

Add runtime PM support.

Signed-off-by: Claudiu Beznea <claudiu.beznea@microchip.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2 years agohwrng: atmel - use __maybe_unused and pm_ptr() for pm ops
Claudiu Beznea [Mon, 21 Feb 2022 07:59:26 +0000 (09:59 +0200)]
hwrng: atmel - use __maybe_unused and pm_ptr() for pm ops

Use __maybe_unused and pm_ptr() for pm ops.

Signed-off-by: Claudiu Beznea <claudiu.beznea@microchip.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2 years agohwrng: atmel - move set of TRNG_HALFR in atmel_trng_init()
Claudiu Beznea [Mon, 21 Feb 2022 07:59:25 +0000 (09:59 +0200)]
hwrng: atmel - move set of TRNG_HALFR in atmel_trng_init()

Move set of TRNG_HALFR in atmel_trng_init() as this function is
also called on resume path. In case of SAMA7G5 where backup and
self-refresh PM mode is available most of the SoC parts are
powered of (including TRNG) when entering suspend. In that case
on resuming path TRNG_HALFR should also be re-configured.

Signed-off-by: Claudiu Beznea <claudiu.beznea@microchip.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2 years agohwrng: atmel - rename enable/disable functions to init/cleanup
Claudiu Beznea [Mon, 21 Feb 2022 07:59:24 +0000 (09:59 +0200)]
hwrng: atmel - rename enable/disable functions to init/cleanup

s/atmel_trng_disable/atmel_trng_cleanup/g and
s/atmel_trng_enable/atmel_trng_init/g to cope with
struct hwrng::{init, cleanup} members.

Signed-off-by: Claudiu Beznea <claudiu.beznea@microchip.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2 years agohwrng: atmel - disable trng on failure path
Claudiu Beznea [Mon, 21 Feb 2022 07:59:23 +0000 (09:59 +0200)]
hwrng: atmel - disable trng on failure path

Call atmel_trng_disable() on failure path of probe.

Fixes: a1fa98d8116f ("hwrng: atmel - disable TRNG during suspend")
Signed-off-by: Claudiu Beznea <claudiu.beznea@microchip.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2 years agohwrng: atmel - add wait for ready support on read
Claudiu Beznea [Mon, 21 Feb 2022 07:59:22 +0000 (09:59 +0200)]
hwrng: atmel - add wait for ready support on read

Add wait for ready support on read.

Signed-off-by: Claudiu Beznea <claudiu.beznea@microchip.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2 years agocrypto: hisilicon/sec - fix the aead software fallback for engine
Kai Ye [Sat, 19 Feb 2022 08:08:08 +0000 (16:08 +0800)]
crypto: hisilicon/sec - fix the aead software fallback for engine

Due to the subreq pointer misuse the private context memory. The aead
soft crypto occasionally casues the OS panic as setting the 64K page.
Here is fix it.

Fixes: 6c46a3297bea ("crypto: hisilicon/sec - add fallback tfm...")
Signed-off-by: Kai Ye <yekai13@huawei.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2 years agocrypto: ccree - don't attempt 0 len DMA mappings
Gilad Ben-Yossef [Thu, 17 Feb 2022 19:27:26 +0000 (21:27 +0200)]
crypto: ccree - don't attempt 0 len DMA mappings

Refuse to try mapping zero bytes as this may cause a fault
on some configurations / platforms and it seems the prev.
attempt is not enough and we need to be more explicit.

Signed-off-by: Gilad Ben-Yossef <gilad@benyossef.com>
Reported-by: Corentin Labbe <clabbe.montjoie@gmail.com>
Fixes: ce0fc6db38de ("crypto: ccree - protect against empty or NULL
scatterlists")
Tested-by: Corentin Labbe <clabbe.montjoie@gmail.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2 years agocrypto: x86/des3 - Remove unused inline function des3_ede_enc_blk_3way()
YueHaibing [Wed, 16 Feb 2022 11:45:21 +0000 (19:45 +0800)]
crypto: x86/des3 - Remove unused inline function des3_ede_enc_blk_3way()

This is unused after commit 768db5fee3bb ("crypto: x86/des - drop CTR mode implementation")

Signed-off-by: YueHaibing <yuehaibing@huawei.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2 years agocrypto: x86/blowfish - Remove unused inline functions
YueHaibing [Wed, 16 Feb 2022 11:44:06 +0000 (19:44 +0800)]
crypto: x86/blowfish - Remove unused inline functions

This is unused after commit c0a64926c53e ("crypto: x86/blowfish - drop CTR mode implementation")

Signed-off-by: YueHaibing <yuehaibing@huawei.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2 years agocrypto: octeontx2 - add synchronization between mailbox accesses
Harman Kalra [Fri, 4 Feb 2022 12:46:01 +0000 (18:16 +0530)]
crypto: octeontx2 - add synchronization between mailbox accesses

Since there are two workqueues implemented in CPTPF driver - one
for handling mailbox requests from VFs and another for handling FLR.
In both cases PF driver will forward the request to AF driver by
writing to mailbox memory. A race condition may arise if two
simultaneous requests are written to mailbox memory. Introducing
locking mechanism to maintain synchronization between multiple
mailbox accesses.

Signed-off-by: Harman Kalra <hkalra@marvell.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2 years agohwrng: core - introduce rng_quality sysfs attribute
Dominik Brodowski [Sun, 13 Feb 2022 20:46:31 +0000 (21:46 +0100)]
hwrng: core - introduce rng_quality sysfs attribute

The rng_quality sysfs attribute returns the quality setting for the
currently active hw_random device, in entropy bits per 1024 bits of
input. Storing a value between 0 and 1024 to this file updates this
estimate accordingly.

Based on the updates to the quality setting, the rngd kernel thread
may be stopped (if no hw_random device is trusted to return entropy),
may be started (if the quality setting is increased from zero), or
may use a different hw_random source (if that has higher quality
output).

Cc: Herbert Xu <herbert@gondor.apana.org.au>
Cc: Jason A. Donenfeld <Jason@zx2c4.com>
Signed-off-by: Dominik Brodowski <linux@dominikbrodowski.net>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2 years agohwrng: core - use per-rng quality value instead of global setting
Dominik Brodowski [Sun, 13 Feb 2022 20:46:30 +0000 (21:46 +0100)]
hwrng: core - use per-rng quality value instead of global setting

The current_quality variable exposed as a module parameter is
fundamentally broken: If it is set at boot time, it is overwritten once
the first hw rng device is loaded; if it is set at runtime, it is
without effect if the hw rng device had its quality value set to 0 (and
no default_quality was set); and if a new rng is selected, it gets
overwritten. Therefore, mark it as obsolete, and replace it by the
per-rng quality setting.

Cc: Herbert Xu <herbert@gondor.apana.org.au>
Cc: Jason A. Donenfeld <Jason@zx2c4.com>
Signed-off-by: Dominik Brodowski <linux@dominikbrodowski.net>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2 years agohwrng: core - start and stop in-kernel rngd in separate function
Dominik Brodowski [Sun, 13 Feb 2022 20:46:29 +0000 (21:46 +0100)]
hwrng: core - start and stop in-kernel rngd in separate function

Extract the start/stop logic for the in-kernel rngd thread to
a separate function.

Cc: Herbert Xu <herbert@gondor.apana.org.au>
Cc: Jason A. Donenfeld <Jason@zx2c4.com>
Signed-off-by: Dominik Brodowski <linux@dominikbrodowski.net>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2 years agohwrng: core - do not bother to order list of devices by quality
Dominik Brodowski [Sun, 13 Feb 2022 20:46:28 +0000 (21:46 +0100)]
hwrng: core - do not bother to order list of devices by quality

There is no real reason why this list needs to be kept ordered by
the driver-provided quality value -- a value which is set only by
a handful of hw_random devices anyway.

Cc: Herbert Xu <herbert@gondor.apana.org.au>
Cc: Jason A. Donenfeld <Jason@zx2c4.com>
Signed-off-by: Dominik Brodowski <linux@dominikbrodowski.net>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2 years agocrypto: rockchip - ECB does not need IV
Corentin Labbe [Fri, 11 Feb 2022 12:16:17 +0000 (12:16 +0000)]
crypto: rockchip - ECB does not need IV

When loading rockchip crypto module, testmgr complains that ivsize of ecb-des3-ede-rk
is not the same than generic implementation.
In fact ECB does not use an IV.

Fixes: ce0183cb6464b ("crypto: rockchip - switch to skcipher API")
Signed-off-by: Corentin Labbe <clabbe@baylibre.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2 years agocrypto: hisilicon/sec - not need to enable sm4 extra mode at HW V3
Kai Ye [Fri, 11 Feb 2022 09:08:18 +0000 (17:08 +0800)]
crypto: hisilicon/sec - not need to enable sm4 extra mode at HW V3

It is not need to enable sm4 extra mode in at HW V3. Here is fix it.

Signed-off-by: Kai Ye <yekai13@huawei.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2 years agocrypto: hisilicon/sec - add the register configuration for HW V3
Kai Ye [Fri, 11 Feb 2022 09:07:58 +0000 (17:07 +0800)]
crypto: hisilicon/sec - add the register configuration for HW V3

Added the register configuration of the SVA mode for HW V3.

Signed-off-by: Kai Ye <yekai13@huawei.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2 years agocrypto: ux500 - use GFP_KERNEL
Julia Lawall [Thu, 10 Feb 2022 20:42:18 +0000 (21:42 +0100)]
crypto: ux500 - use GFP_KERNEL

Platform_driver probe functions aren't called with locks held
and thus don't need GFP_ATOMIC. Use GFP_KERNEL instead.

Problem found with Coccinelle.

Signed-off-by: Julia Lawall <Julia.Lawall@inria.fr>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2 years agocrypto: nx - Constify static attribute_group structs
Rikard Falkeborn [Thu, 10 Feb 2022 20:28:05 +0000 (21:28 +0100)]
crypto: nx - Constify static attribute_group structs

The only usage of these is to pass their address to
sysfs_{create,remove}_group(), which takes pointers to const struct
attribute_group. Make them const to allow the compiler to put them in
read-only memory.

Signed-off-by: Rikard Falkeborn <rikard.falkeborn@gmail.com>
Reviewed-by: Daniel Axtens <dja@axtens.net>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2 years agocrypto: omap-sham - Constify static attribute_group
Rikard Falkeborn [Thu, 10 Feb 2022 20:28:04 +0000 (21:28 +0100)]
crypto: omap-sham - Constify static attribute_group

The only usage of omap_sham_attr_group is to pass its address to
sysfs_{create,remove}_group(), which takes pointers to const struct
attribute_group. Make it const to allow the compiler to put it in
read-only memory.

Signed-off-by: Rikard Falkeborn <rikard.falkeborn@gmail.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2 years agocrypto: omap-aes - Constify static attribute_group
Rikard Falkeborn [Thu, 10 Feb 2022 20:28:03 +0000 (21:28 +0100)]
crypto: omap-aes - Constify static attribute_group

The only usage of omap_aes_attr_group is to pass its address to
sysfs_{create,remove}_group(), which takes pointers to const struct
attribute_group. Make it const to allow the compiler to put it in
read-only memory.

Signed-off-by: Rikard Falkeborn <rikard.falkeborn@gmail.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2 years agocrypto: qat - enable power management for QAT GEN4
Wojciech Ziemba [Thu, 10 Feb 2022 13:38:27 +0000 (13:38 +0000)]
crypto: qat - enable power management for QAT GEN4

Add support for HW QAT Power Management (PM) feature.
This feature is enabled at init time (1) by sending an admin message to
the firmware, targeting the admin AE, that sets the idle time before
the device changes state and (2) by unmasking the PM source of interrupt
in ERRMSK2.

The interrupt handler is extended to handle a PM interrupt which
is triggered by HW when a PM transition occurs. In this case, the
driver responds acknowledging the transaction using the HOST_MSG
mailbox.

Signed-off-by: Wojciech Ziemba <wojciech.ziemba@intel.com>
Co-developed-by: Marcinx Malinowski <marcinx.malinowski@intel.com>
Signed-off-by: Marcinx Malinowski <marcinx.malinowski@intel.com>
Reviewed-by: Giovanni Cabiddu <giovanni.cabiddu@intel.com>
Reviewed-by: Marco Chiappero <marco.chiappero@intel.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2 years agocrypto: qat - move and rename GEN4 error register definitions
Wojciech Ziemba [Thu, 10 Feb 2022 13:38:26 +0000 (13:38 +0000)]
crypto: qat - move and rename GEN4 error register definitions

Move error source related CSRs from 4xxx to the wider GEN4 header file.

Signed-off-by: Wojciech Ziemba <wojciech.ziemba@intel.com>
Reviewed-by: Giovanni Cabiddu <giovanni.cabiddu@intel.com>
Reviewed-by: Marco Chiappero <marco.chiappero@intel.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2 years agocrypto: qat - add misc workqueue
Wojciech Ziemba [Thu, 10 Feb 2022 13:38:25 +0000 (13:38 +0000)]
crypto: qat - add misc workqueue

In an effort to reduce the amount of workqueues, scattered across
the QAT driver, introduce the misc workqueue. This queue will be used
to handle bottom halves, Power Management and more in the future.

The function adf_misc_wq_queue_work() has been added to simplify
the enqueuing of jobs.

Signed-off-by: Wojciech Ziemba <wojciech.ziemba@intel.com>
Reviewed-by: Giovanni Cabiddu <giovanni.cabiddu@intel.com>
Reviewed-by: Marco Chiappero <marco.chiappero@intel.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2 years agocrypto: lrw - Add dependency on ecb
Herbert Xu [Thu, 10 Feb 2022 02:31:13 +0000 (13:31 +1100)]
crypto: lrw - Add dependency on ecb

The lrw template relies on ecb to work.  So we need to declare
a Kconfig dependency as well as a module softdep on it.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2 years agocrypto: xts - Add softdep on ecb
Herbert Xu [Thu, 10 Feb 2022 02:09:40 +0000 (13:09 +1100)]
crypto: xts - Add softdep on ecb

The xts module needs ecb to be present as it's meant to work
on top of ecb.  This patch adds a softdep so ecb can be included
automatically into the initramfs.

Reported-by: rftc <rftc@gmx.de>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2 years agocrypto: qat - don't cast parameter in bit operations
Andy Shevchenko [Wed, 9 Feb 2022 18:00:47 +0000 (20:00 +0200)]
crypto: qat - don't cast parameter in bit operations

While in this particular case it would not be a (critical) issue,
the pattern itself is bad and error prone in case the location
of the parameter is changed.

Don't cast parameter to unsigned long pointer in the bit operations.
Instead copy to a local variable on stack of a proper type and use.

Fixes: b4b7e67c917f ("crypto: qat - Intel(R) QAT ucode part of fw loader")
Signed-off-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Reviewed-by: Giovanni Cabiddu <giovanni.cabiddu@intel.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2 years agocrypto: cavium/nitrox - fix typo on crypto
Corentin Labbe [Wed, 9 Feb 2022 10:21:58 +0000 (10:21 +0000)]
crypto: cavium/nitrox - fix typo on crypto

crypto had a typo, fix it.

Signed-off-by: Corentin Labbe <clabbe@baylibre.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2 years agocrypto: arm/xor - make vectorized C code Clang-friendly
Ard Biesheuvel [Sat, 5 Feb 2022 15:23:46 +0000 (16:23 +0100)]
crypto: arm/xor - make vectorized C code Clang-friendly

The ARM version of the accelerated XOR routines are simply the 8-way C
routines passed through the auto-vectorizer with SIMD codegen enabled.
This used to require GCC version 4.6 at least, but given that 5.1 is now
the baseline, this check is no longer necessary, and actually
misidentifies Clang as GCC < 4.6 as Clang defines the GCC major/minor as
well, but makes no attempt at doing this in a way that conveys feature
parity with a certain version of GCC (which would not be a great idea in
the first place).

So let's drop the version check, and make the auto-vectorize pragma
(which is based on a GCC-specific command line option) GCC-only. Since
Clang performs SIMD auto-vectorization by default at -O2, no pragma is
necessary here.

Tested-by: Nathan Chancellor <nathan@kernel.org>
Signed-off-by: Ard Biesheuvel <ardb@kernel.org>
Reviewed-by: Nick Desaulniers <ndesaulniers@google.com>
Link: https://github.com/ClangBuiltLinux/linux/issues/496
Link: https://github.com/ClangBuiltLinux/linux/issues/503
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2 years agolib/xor: make xor prototypes more friendly to compiler vectorization
Ard Biesheuvel [Sat, 5 Feb 2022 15:23:45 +0000 (16:23 +0100)]
lib/xor: make xor prototypes more friendly to compiler vectorization

Modern compilers are perfectly capable of extracting parallelism from
the XOR routines, provided that the prototypes reflect the nature of the
input accurately, in particular, the fact that the input vectors are
expected not to overlap. This is not documented explicitly, but is
implied by the interchangeability of the various C routines, some of
which use temporary variables while others don't: this means that these
routines only behave identically for non-overlapping inputs.

So let's decorate these input vectors with the __restrict modifier,
which informs the compiler that there is no overlap. While at it, make
the input-only vectors pointer-to-const as well.

Tested-by: Nathan Chancellor <nathan@kernel.org>
Signed-off-by: Ard Biesheuvel <ardb@kernel.org>
Reviewed-by: Nick Desaulniers <ndesaulniers@google.com>
Link: https://github.com/ClangBuiltLinux/linux/issues/563
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2 years agocrypto: atmel-tdes - Add support for the TDES IP available on sama7g5 SoC
Tudor Ambarus [Fri, 4 Feb 2022 13:35:22 +0000 (15:35 +0200)]
crypto: atmel-tdes - Add support for the TDES IP available on sama7g5 SoC

Add support for the TDES IP found on sama7g5.

Signed-off-by: Tudor Ambarus <tudor.ambarus@microchip.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2 years agocrypto: hmac - disallow keys < 112 bits in FIPS mode
Stephan Müller [Tue, 1 Feb 2022 08:41:32 +0000 (09:41 +0100)]
crypto: hmac - disallow keys < 112 bits in FIPS mode

FIPS 140 requires a minimum security strength of 112 bits. This implies
that the HMAC key must not be smaller than 112 in FIPS mode.

This restriction implies that the test vectors for HMAC that have a key
that is smaller than 112 bits must be disabled when FIPS support is
compiled.

Signed-off-by: Stephan Mueller <smueller@chronox.de>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2 years agocrypto: hmac - add fips_skip support
Stephan Müller [Tue, 1 Feb 2022 08:40:58 +0000 (09:40 +0100)]
crypto: hmac - add fips_skip support

By adding the support for the flag fips_skip, hash / HMAC test vectors
may be marked to be not applicable in FIPS mode. Such vectors are
silently skipped in FIPS mode.

Signed-off-by: Stephan Mueller <smueller@chronox.de>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2 years agocrypto: sl3516 - remove redundant initializations of pointers in_sg and out_sg
Colin Ian King [Sun, 30 Jan 2022 23:05:15 +0000 (23:05 +0000)]
crypto: sl3516 - remove redundant initializations of pointers in_sg and out_sg

Pointers in_sg and out_sg are being initialized with values that are
never read, they are being re-assigned the same values later on. The
initializations are redundant, remove them in preference to the later
assignments that are closer to when the pointers are being used.

Signed-off-by: Colin Ian King <colin.i.king@gmail.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2 years agocrypto: marvell/octeontx - remove redundant initialization of variable c_size
Colin Ian King [Sun, 30 Jan 2022 22:53:25 +0000 (22:53 +0000)]
crypto: marvell/octeontx - remove redundant initialization of variable c_size

Variable c_size is being initialized with a value that is never read, it
is being re-assigned with a different value later on. The initialization
is redundant and can be removed.

Signed-off-by: Colin Ian King <colin.i.king@gmail.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2 years agocrypto: octeontx2 - remove CONFIG_DM_CRYPT check
Shijith Thotton [Fri, 28 Jan 2022 13:57:42 +0000 (19:27 +0530)]
crypto: octeontx2 - remove CONFIG_DM_CRYPT check

No issues were found while using the driver with dm-crypt enabled. So
CONFIG_DM_CRYPT check in the driver can be removed.

This also fixes the NULL pointer dereference in driver release if
CONFIG_DM_CRYPT is enabled.

...
Unable to handle kernel NULL pointer dereference at virtual address 0000000000000008
...
Call trace:
 crypto_unregister_alg+0x68/0xfc
 crypto_unregister_skciphers+0x44/0x60
 otx2_cpt_crypto_exit+0x100/0x1a0
 otx2_cptvf_remove+0xf8/0x200
 pci_device_remove+0x3c/0xd4
 __device_release_driver+0x188/0x234
 device_release_driver+0x2c/0x4c
...

Fixes: 6f03f0e8b6c8 ("crypto: octeontx2 - register with linux crypto framework")
Signed-off-by: Shijith Thotton <sthotton@marvell.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2 years agocrypto: tcrypt - remove all multibuffer ahash tests
Tianjia Zhang [Fri, 28 Jan 2022 08:38:35 +0000 (16:38 +0800)]
crypto: tcrypt - remove all multibuffer ahash tests

The multibuffer algorithms was removed already in 2018, so it is
necessary to clear the test code left by tcrypt.

Suggested-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: Tianjia Zhang <tianjia.zhang@linux.alibaba.com>
Acked-by: Ard Biesheuvel <ardb@kernel.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2 years agocrypto: atmel - add support for AES and SHA IPs available on lan966x SoC
Kavyasree Kotagiri [Fri, 28 Jan 2022 07:17:55 +0000 (12:47 +0530)]
crypto: atmel - add support for AES and SHA IPs available on lan966x SoC

This patch adds support for hardware version of AES and SHA IPs
available on lan966x SoC.

Signed-off-by: Kavyasree Kotagiri <kavyasree.kotagiri@microchip.com>
Reviewed-by: Tudor Ambarus <tudor.ambarus@microchip.com>
Tested-by: Tudor Ambarus <tudor.ambarus@microchip.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2 years agohwrng: core - credit entropy for low quality sources of randomness
Dominik Brodowski [Fri, 28 Jan 2022 07:02:54 +0000 (08:02 +0100)]
hwrng: core - credit entropy for low quality sources of randomness

In case the entropy quality is low, there may be less than one bit to
credit in the call to add_hwgenerator_randomness(): The number of bytes
returned by rng_get_data() multiplied by the current quality (in entropy
bits per 1024 bits of input) must be larger than 128 to credit at least
one bit. However, imx-rngc.c sets the quality to 19, but may return less
than 32 bytes; hid_u2fzero.c sets the quality to 1; and users may override
the quality setting manually.

In case there is less than one bit to credit, keep track of it and add
that credit to the next iteration.

Cc: Herbert Xu <herbert@gondor.apana.org.au>
Cc: Jason A. Donenfeld <Jason@zx2c4.com>
Signed-off-by: Dominik Brodowski <linux@dominikbrodowski.net>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2 years agocrypto: arm64/aes-neonbs-xts - use plain NEON for non-power-of-2 input sizes
Ard Biesheuvel [Thu, 27 Jan 2022 11:35:45 +0000 (12:35 +0100)]
crypto: arm64/aes-neonbs-xts - use plain NEON for non-power-of-2 input sizes

Even though the kernel's implementations of AES-XTS were updated to
implement ciphertext stealing and can operate on inputs of any size
larger than or equal to the AES block size, this feature is rarely used
in practice.

In fact, in the kernel, AES-XTS is only used to operate on 4096 or 512
byte blocks, which means that not only the ciphertext stealing is
effectively dead code, the logic in the bit sliced NEON implementation
to deal with fewer than 8 blocks at a time is also never used.

Since the bit-sliced NEON driver already depends on the plain NEON
version, which is slower but can operate on smaller data quantities more
straightforwardly, let's fallback to the plain NEON implementation of
XTS for any residual inputs that are not multiples of 128 bytes. This
allows us to remove a lot of complicated logic that rarely gets
exercised in practice.

Signed-off-by: Ard Biesheuvel <ardb@kernel.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2 years agocrypto: arm64/aes-neonbs-ctr - fallback to plain NEON for final chunk
Ard Biesheuvel [Thu, 27 Jan 2022 11:35:44 +0000 (12:35 +0100)]
crypto: arm64/aes-neonbs-ctr - fallback to plain NEON for final chunk

Instead of processing the entire input with the 8-way bit sliced
algorithm, which is sub-optimal for inputs that are not a multiple of
128 bytes in size, invoke the plain NEON version of CTR for the
remainder of the input after processing the bulk using 128 byte strides.

This allows us to greatly simplify the asm code that implements CTR, and
get rid of all the branches and special code paths. It also gains us a
couple of percent of performance.

Signed-off-by: Ard Biesheuvel <ardb@kernel.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2 years agocrypto: arm/aes-neonbs-ctr - deal with non-multiples of AES block size
Ard Biesheuvel [Thu, 27 Jan 2022 11:35:43 +0000 (12:35 +0100)]
crypto: arm/aes-neonbs-ctr - deal with non-multiples of AES block size

Instead of falling back to C code to deal with the final bit of input
that is not a round multiple of the block size, handle this in the asm
code, permitting us to use overlapping loads and stores for performance,
and implement the 16-byte wide XOR using a single NEON instruction.

Since NEON loads and stores have a natural width of 16 bytes, we need to
handle inputs of less than 16 bytes in a special way, but this rarely
occurs in practice so it does not impact performance. All other input
sizes can be consumed directly by the NEON asm code, although it should
be noted that the core AES transform can still only process 128 bytes (8
AES blocks) at a time.

Signed-off-by: Ard Biesheuvel <ardb@kernel.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2 years agocrypto: arm64/aes-neon-ctr - improve handling of single tail block
Ard Biesheuvel [Thu, 27 Jan 2022 09:52:11 +0000 (10:52 +0100)]
crypto: arm64/aes-neon-ctr - improve handling of single tail block

Instead of falling back to C code to do a memcpy of the output of the
last block, handle this in the asm code directly if possible, which is
the case if the entire input is longer than 16 bytes.

Cc: Nathan Huckleberry <nhuck@google.com>
Cc: Eric Biggers <ebiggers@google.com>
Signed-off-by: Ard Biesheuvel <ardb@kernel.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2 years agocrypto: octeontx2 - increase CPT HW instruction queue length
Srujana Challa [Tue, 25 Jan 2022 18:26:24 +0000 (23:56 +0530)]
crypto: octeontx2 - increase CPT HW instruction queue length

LDWB is getting incorrectly used in HW when
CPT_AF_LF()_PTR_CTL[IQB_LDWB]=1 and CPT instruction queue has less than
320 free entries. So, increase HW instruction queue size by 320 and give
320 entries less for SW/NIX RX as a SW workaround.

Signed-off-by: Srujana Challa <schalla@marvell.com>
Signed-off-by: Shijith Thotton <sthotton@marvell.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2 years agocrypto: octeontx2 - disable DMA black hole on an DMA fault
Srujana Challa [Tue, 25 Jan 2022 18:26:23 +0000 (23:56 +0530)]
crypto: octeontx2 - disable DMA black hole on an DMA fault

When CPT_AF_DIAG[FLT_DIS] = 0 and a CPT engine access to
LLC/DRAM encounters a fault/poison, a rare case may result
in unpredictable data being delivered to a CPT engine.
So, this patch adds code to set FLT_DIS as a workaround.

Signed-off-by: Srujana Challa <schalla@marvell.com>
Signed-off-by: Shijith Thotton <sthotton@marvell.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2 years agocrypto: octeontx2 - CN10K CPT to RNM workaround
Srujana Challa [Tue, 25 Jan 2022 18:26:22 +0000 (23:56 +0530)]
crypto: octeontx2 - CN10K CPT to RNM workaround

When software sets CPT_AF_CTL[RNM_REQ_EN]=1 and RNM in not producing
entropy(i.e., RNM_ENTROPY_STATUS[NORMAL_CNT] < 0x40), the first cycle of
the response may be lost due to a conditional clocking issue. Due to
this, the subsequent random number stream will be corrupted. So, this
patch adds support to ensure RNM_ENTROPY_STATUS[NORMAL_CNT] = 0x40
before writing CPT_AF_CTL[RNM_REQ_EN] = 1, as a workaround.

Signed-off-by: Srujana Challa <schalla@marvell.com>
Signed-off-by: Shijith Thotton <sthotton@marvell.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2 years agohwrng: core - break out of hwrng_fillfn if current rng is not trusted
Dominik Brodowski [Mon, 24 Jan 2022 20:29:50 +0000 (21:29 +0100)]
hwrng: core - break out of hwrng_fillfn if current rng is not trusted

For two reasons, current_quality may become zero within the rngd
kernel thread: (1) The user lowers current_quality to 0 by writing
to the sysfs module parameter file (note that increasing the quality
from zero is without effect at the moment), or (2) there are two or
more hwrng devices registered, and those which provide quality>0 are
unregistered, but one with quality==0 remains.

If current_quality is 0, the randomness is not trusted and cannot help
to increase the entropy count. That will lead to continuous calls to
the hwrngd thread and continuous stirring of the input pool with
untrusted bits.

Cc: Matt Mackall <mpm@selenic.com>
Cc: Herbert Xu <herbert@gondor.apana.org.au>
Cc: Jason A. Donenfeld <Jason@zx2c4.com>
Signed-off-by: Dominik Brodowski <linux@dominikbrodowski.net>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2 years agohwrng: core - only set cur_rng_set_by_user if it is working
Dominik Brodowski [Mon, 24 Jan 2022 20:29:49 +0000 (21:29 +0100)]
hwrng: core - only set cur_rng_set_by_user if it is working

In case the user-specified rng device is not working, it is not used;
therefore cur_rng_set_by_user must not be set to 1.

Cc: Matt Mackall <mpm@selenic.com>
Cc: Herbert Xu <herbert@gondor.apana.org.au>
Cc: Jason A. Donenfeld <Jason@zx2c4.com>
Signed-off-by: Dominik Brodowski <linux@dominikbrodowski.net>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2 years agohwrng: core - use rng_fillbuf in add_early_randomness()
Dominik Brodowski [Mon, 24 Jan 2022 20:29:48 +0000 (21:29 +0100)]
hwrng: core - use rng_fillbuf in add_early_randomness()

Using rng_buffer in add_early_randomness() may race with rng_dev_read().
Use rng_fillbuf instead, as it is otherwise only used within the kernel
by hwrng_fillfn() and therefore never exposed to userspace.

Cc: Matt Mackall <mpm@selenic.com>
Cc: Herbert Xu <herbert@gondor.apana.org.au>
Cc: Jason A. Donenfeld <Jason@zx2c4.com>
Signed-off-by: Dominik Brodowski <linux@dominikbrodowski.net>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2 years agohwrng: core - read() callback must be called for size of 32 or more bytes
Dominik Brodowski [Mon, 24 Jan 2022 20:29:47 +0000 (21:29 +0100)]
hwrng: core - read() callback must be called for size of 32 or more bytes

According to <linux/hw_random.h>, the @max parameter of the ->read
callback "is a multiple of 4 and >= 32 bytes". That promise was not
kept by add_early_randomness(), which only asked for 16 bytes. As
rng_buffer_size() is at least 32, we can simply ask for 32 bytes.

Cc: Matt Mackall <mpm@selenic.com>
Cc: Herbert Xu <herbert@gondor.apana.org.au>
Cc: Jason A. Donenfeld <Jason@zx2c4.com>
Signed-off-by: Dominik Brodowski <linux@dominikbrodowski.net>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2 years agohwrng: core - explicit ordering of initcalls
Dominik Brodowski [Mon, 24 Jan 2022 20:29:46 +0000 (21:29 +0100)]
hwrng: core - explicit ordering of initcalls

hw-random device drivers depend on the hw-random core being
initialized. Make this ordering explicit, also for the case
these drivers are built-in. As the core itself depends on
misc_register() which is set up at subsys_initcall time,
advance the initialization of the core (only) to the
fs_initcall() level.

Cc: Matt Mackall <mpm@selenic.com>
Cc: Herbert Xu <herbert@gondor.apana.org.au>
Cc: Jason A. Donenfeld <Jason@zx2c4.com>
Signed-off-by: Dominik Brodowski <linux@dominikbrodowski.net>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2 years agopadata: replace cpumask_weight with cpumask_empty in padata.c
Yury Norov [Sun, 23 Jan 2022 18:38:52 +0000 (10:38 -0800)]
padata: replace cpumask_weight with cpumask_empty in padata.c

padata_do_parallel() calls cpumask_weight() to check if any bit of a
given cpumask is set. We can do it more efficiently with cpumask_empty()
because cpumask_empty() stops traversing the cpumask as soon as it finds
first set bit, while cpumask_weight() counts all bits unconditionally.

Signed-off-by: Yury Norov <yury.norov@gmail.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2 years agocrypto: mxs-dcp - Fix scatterlist processing
Tomas Paukrt [Sat, 22 Jan 2022 17:07:53 +0000 (18:07 +0100)]
crypto: mxs-dcp - Fix scatterlist processing

This patch fixes a bug in scatterlist processing that may cause incorrect AES block encryption/decryption.

Fixes: 2e6d793e1bf0 ("crypto: mxs-dcp - Use sg_mapping_iter to copy data")
Signed-off-by: Tomas Paukrt <tomaspaukrt@email.cz>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2 years agocrypto: hisilicon/qm - cleanup warning in qm_vf_read_qos
Kai Ye [Sat, 22 Jan 2022 09:30:42 +0000 (17:30 +0800)]
crypto: hisilicon/qm - cleanup warning in qm_vf_read_qos

The kernel test rebot report this warning: Uninitialized variable: ret.
The code flow may return value of ret directly. This value is an
uninitialized variable, here is fix it.

Signed-off-by: Kai Ye <yekai13@huawei.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2 years agocrypto: hisilicon/sec - use the correct print format
Kai Ye [Sat, 22 Jan 2022 08:13:12 +0000 (16:13 +0800)]
crypto: hisilicon/sec - use the correct print format

Use the correct print format. Printing an unsigned int value should
use %u instead of %d.

Signed-off-by: Kai Ye <yekai13@huawei.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2 years agocrypto: hisilicon/sec - fix the CTR mode BD configuration
Kai Ye [Sat, 22 Jan 2022 08:13:11 +0000 (16:13 +0800)]
crypto: hisilicon/sec - fix the CTR mode BD configuration

The CTR counter is 32bit rollover default on the BD.
But the NIST standard is 128bit rollover. it cause the
testing failed, so need to fix the BD configuration.

Signed-off-by: Kai Ye <yekai13@huawei.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2 years agocrypto: hisilicon/sec - fix the max length of AAD for the CCM mode
Kai Ye [Sat, 22 Jan 2022 08:13:10 +0000 (16:13 +0800)]
crypto: hisilicon/sec - fix the max length of AAD for the CCM mode

Fix the maximum length of AAD for the CCM mode due to the hardware limited.

Signed-off-by: Kai Ye <yekai13@huawei.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2 years agocrypto: hisilicon/sec - add some comments for soft fallback
Kai Ye [Sat, 22 Jan 2022 08:13:09 +0000 (16:13 +0800)]
crypto: hisilicon/sec - add some comments for soft fallback

Modify the print of information that might lead to user misunderstanding.
Currently only XTS mode need the fallback tfm when using 192bit key.
Others algs not need soft fallback tfm. So others algs can return
directly.

Signed-off-by: Kai Ye <yekai13@huawei.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2 years agocrypto: hisilicon/sec - fixup icv checking enabled on Kunpeng 930
Kai Ye [Sat, 22 Jan 2022 08:13:08 +0000 (16:13 +0800)]
crypto: hisilicon/sec - fixup icv checking enabled on Kunpeng 930

Fixup icv(integrity check value) checking enabled wrong on
Kunpeng 930

Signed-off-by: Kai Ye <yekai13@huawei.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2 years agocrypto: octeontx2 - select CONFIG_NET_DEVLINK
Shijith Thotton [Fri, 21 Jan 2022 11:12:32 +0000 (16:42 +0530)]
crypto: octeontx2 - select CONFIG_NET_DEVLINK

OcteonTX2 CPT driver will fail to link without devlink support.

aarch64-linux-gnu-ld: otx2_cpt_devlink.o: in function `otx2_cpt_dl_egrp_delete':
otx2_cpt_devlink.c:18: undefined reference to `devlink_priv'
aarch64-linux-gnu-ld: otx2_cpt_devlink.o: in function `otx2_cpt_dl_egrp_create':
otx2_cpt_devlink.c:9: undefined reference to `devlink_priv'
aarch64-linux-gnu-ld: otx2_cpt_devlink.o: in function `otx2_cpt_dl_uc_info':
otx2_cpt_devlink.c:27: undefined reference to `devlink_priv'

Fixes: fed8f4d5f946 ("crypto: octeontx2 - parameters for custom engine groups")

Signed-off-by: Shijith Thotton <sthotton@marvell.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2 years agocrypto: memneq - avoid implicit unaligned accesses
Ard Biesheuvel [Wed, 19 Jan 2022 09:31:09 +0000 (10:31 +0100)]
crypto: memneq - avoid implicit unaligned accesses

The C standard does not support dereferencing pointers that are not
aligned with respect to the pointed-to type, and doing so is technically
undefined behavior, even if the underlying hardware supports it.

This means that conditionally dereferencing such pointers based on
whether CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS=y is not the right thing
to do, and actually results in alignment faults on ARM, which are fixed
up on a slow path. Instead, we should use the unaligned accessors in
such cases: on architectures that don't care about alignment, they will
result in identical codegen whereas, e.g., codegen on ARM will avoid
doubleword loads and stores but use ordinary ones, which are able to
tolerate misalignment.

Link: https://lore.kernel.org/linux-crypto/CAHk-=wiKkdYLY0bv+nXrcJz3NH9mAqPAafX7PpW5EwVtxsEu7Q@mail.gmail.com/
Signed-off-by: Ard Biesheuvel <ardb@kernel.org>
Reviewed-by: Arnd Bergmann <arnd@arndb.de>
Reviewed-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2 years agocrypto: authenc - Fix sleep in atomic context in decrypt_tail
Herbert Xu [Wed, 19 Jan 2022 06:58:40 +0000 (17:58 +1100)]
crypto: authenc - Fix sleep in atomic context in decrypt_tail

The function crypto_authenc_decrypt_tail discards its flags
argument and always relies on the flags from the original request
when starting its sub-request.

This is clearly wrong as it may cause the SLEEPABLE flag to be
set when it shouldn't.

Fixes: 92d95ba91772 ("crypto: authenc - Convert to new AEAD interface")
Reported-by: Corentin Labbe <clabbe.montjoie@gmail.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Tested-by: Corentin Labbe <clabbe.montjoie@gmail.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2 years agocrypto: rsa-pkcs1pad - use clearer variable names
Eric Biggers [Wed, 19 Jan 2022 00:13:06 +0000 (16:13 -0800)]
crypto: rsa-pkcs1pad - use clearer variable names

The new convention for akcipher_alg::verify makes it unclear which
values are the lengths of the signature and digest.  Add local variables
to make it clearer what is going on.

Also rename the digest_size variable in pkcs1pad_sign(), as it is
actually the digest *info* size, not the digest size which is different.

Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2 years agocrypto: rsa-pkcs1pad - fix buffer overread in pkcs1pad_verify_complete()
Eric Biggers [Wed, 19 Jan 2022 00:13:05 +0000 (16:13 -0800)]
crypto: rsa-pkcs1pad - fix buffer overread in pkcs1pad_verify_complete()

Before checking whether the expected digest_info is present, we need to
check that there are enough bytes remaining.

Fixes: a49de377e051 ("crypto: Add hash param to pkcs1pad")
Cc: <stable@vger.kernel.org> # v4.6+
Cc: Tadeusz Struk <tadeusz.struk@linaro.org>
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2 years agocrypto: rsa-pkcs1pad - restore signature length check
Eric Biggers [Wed, 19 Jan 2022 00:13:04 +0000 (16:13 -0800)]
crypto: rsa-pkcs1pad - restore signature length check

RSA PKCS#1 v1.5 signatures are required to be the same length as the RSA
key size.  RFC8017 specifically requires the verifier to check this
(https://datatracker.ietf.org/doc/html/rfc8017#section-8.2.2).

Commit a49de377e051 ("crypto: Add hash param to pkcs1pad") changed the
kernel to allow longer signatures, but didn't explain this part of the
change; it seems to be unrelated to the rest of the commit.

Revert this change, since it doesn't appear to be correct.

We can be pretty sure that no one is relying on overly-long signatures
(which would have to be front-padded with zeroes) being supported, given
that they would have been broken since commit c7381b012872
("crypto: akcipher - new verify API for public key algorithms").

Fixes: a49de377e051 ("crypto: Add hash param to pkcs1pad")
Cc: <stable@vger.kernel.org> # v4.6+
Cc: Tadeusz Struk <tadeusz.struk@linaro.org>
Suggested-by: Vitaly Chikunov <vt@altlinux.org>
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2 years agocrypto: rsa-pkcs1pad - correctly get hash from source scatterlist
Eric Biggers [Wed, 19 Jan 2022 00:13:03 +0000 (16:13 -0800)]
crypto: rsa-pkcs1pad - correctly get hash from source scatterlist

Commit c7381b012872 ("crypto: akcipher - new verify API for public key
algorithms") changed akcipher_alg::verify to take in both the signature
and the actual hash and do the signature verification, rather than just
return the hash expected by the signature as was the case before.  To do
this, it implemented a hack where the signature and hash are
concatenated with each other in one scatterlist.

Obviously, for this to work correctly, akcipher_alg::verify needs to
correctly extract the two items from the scatterlist it is given.
Unfortunately, it doesn't correctly extract the hash in the case where
the signature is longer than the RSA key size, as it assumes that the
signature's length is equal to the RSA key size.  This causes a prefix
of the hash, or even the entire hash, to be taken from the *signature*.

(Note, the case of a signature longer than the RSA key size should not
be allowed in the first place; a separate patch will fix that.)

It is unclear whether the resulting scheme has any useful security
properties.

Fix this by correctly extracting the hash from the scatterlist.

Fixes: c7381b012872 ("crypto: akcipher - new verify API for public key algorithms")
Cc: <stable@vger.kernel.org> # v5.2+
Reviewed-by: Vitaly Chikunov <vt@altlinux.org>
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2 years agocrypto: rsa-pkcs1pad - only allow with rsa
Eric Biggers [Wed, 19 Jan 2022 00:13:02 +0000 (16:13 -0800)]
crypto: rsa-pkcs1pad - only allow with rsa

The pkcs1pad template can be instantiated with an arbitrary akcipher
algorithm, which doesn't make sense; it is specifically an RSA padding
scheme.  Make it check that the underlying algorithm really is RSA.

Fixes: 3d5b1ecdea6f ("crypto: rsa - RSA padding algorithm")
Cc: <stable@vger.kernel.org> # v4.5+
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2 years agocrypto: qat - fix access to PFVF interrupt registers for GEN4
Giovanni Cabiddu [Tue, 18 Jan 2022 10:35:15 +0000 (10:35 +0000)]
crypto: qat - fix access to PFVF interrupt registers for GEN4

The logic that detects, enables and disables pfvf interrupts was
expecting a single CSR per VF. Instead, the source and mask register are
two registers with a bit per VF.
Due to this, the driver is reading and setting reserved CSRs and not
masking the correct source of interrupts.

Fix the access to the source and mask register for QAT GEN4 devices by
removing the outer loop in adf_gen4_get_vf2pf_sources(),
adf_gen4_enable_vf2pf_interrupts() and
adf_gen4_disable_vf2pf_interrupts() and changing the helper macros
ADF_4XXX_VM2PF_SOU and ADF_4XXX_VM2PF_MSK.

Fixes: a9dc0d966605 ("crypto: qat - add PFVF support to the GEN4 host driver")
Signed-off-by: Giovanni Cabiddu <giovanni.cabiddu@intel.com>
Co-developed-by: Siming Wan <siming.wan@intel.com>
Signed-off-by: Siming Wan <siming.wan@intel.com>
Reviewed-by: Xin Zeng <xin.zeng@intel.com>
Reviewed-by: Wojciech Ziemba <wojciech.ziemba@intel.com>
Reviewed-by: Marco Chiappero <marco.chiappero@intel.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2 years agoarm64: lib: accelerate crc32_be
Kevin Bracey [Tue, 18 Jan 2022 10:23:51 +0000 (12:23 +0200)]
arm64: lib: accelerate crc32_be

It makes no sense to leave crc32_be using the generic code while we
only accelerate the little-endian ops.

Even though the big-endian form doesn't fit as smoothly into the arm64,
we can speed it up and avoid hitting the D cache.

Tested on Cortex-A53. Without acceleration:

    crc32: CRC_LE_BITS = 64, CRC_BE BITS = 64
    crc32: self tests passed, processed 225944 bytes in 192240 nsec
    crc32c: CRC_LE_BITS = 64
    crc32c: self tests passed, processed 112972 bytes in 21360 nsec

With acceleration:

    crc32: CRC_LE_BITS = 64, CRC_BE BITS = 64
    crc32: self tests passed, processed 225944 bytes in 53480 nsec
    crc32c: CRC_LE_BITS = 64
    crc32c: self tests passed, processed 112972 bytes in 21480 nsec

Signed-off-by: Kevin Bracey <kevin@bracey.fi>
Tested-by: Ard Biesheuvel <ardb@kernel.org>
Reviewed-by: Ard Biesheuvel <ardb@kernel.org>
Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2 years agolib/crc32test: correct printed bytes count
Kevin Bracey [Tue, 18 Jan 2022 10:23:50 +0000 (12:23 +0200)]
lib/crc32test: correct printed bytes count

crc32c_le self test had a stray multiply by two inherited from
the crc32_le+crc32_be test loop.

Signed-off-by: Kevin Bracey <kevin@bracey.fi>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2 years agolib/crc32: Make crc32_be weak for arch override
Kevin Bracey [Tue, 18 Jan 2022 10:23:49 +0000 (12:23 +0200)]
lib/crc32: Make crc32_be weak for arch override

crc32_le and __crc32c_le can be overridden - extend this to crc32_be.

Signed-off-by: Kevin Bracey <kevin@bracey.fi>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2 years agolib/crc32: remove unneeded casts
Kevin Bracey [Tue, 18 Jan 2022 10:23:48 +0000 (12:23 +0200)]
lib/crc32: remove unneeded casts

Casts were added in commit 8f243af42ade ("sections: fix const sections
for crc32 table") to cope with the tables not being const. They are no
longer required since commit f5e38b9284e1 ("lib: crc32: constify crc32
lookup table").

Signed-off-by: Kevin Bracey <kevin@bracey.fi>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2 years agocrypto: kdf - Select hmac in addition to sha256
Herbert Xu [Tue, 18 Jan 2022 05:31:29 +0000 (16:31 +1100)]
crypto: kdf - Select hmac in addition to sha256

In addition to sha256 we must also enable hmac for the kdf self-test
to work.

Reported-by: kernel test robot <oliver.sang@intel.com>
Fixes: 304b4acee2f0 ("crypto: kdf - select SHA-256 required...")
Fixes: 026a733e6659 ("crypto: kdf - add SP800-108 counter key...")
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2 years agocrypto: sun8i-ss - really disable hash on A80
Corentin Labbe [Sat, 15 Jan 2022 10:07:14 +0000 (10:07 +0000)]
crypto: sun8i-ss - really disable hash on A80

When adding hashes support to sun8i-ss, I have added them only on A83T.
But I forgot that 0 is a valid algorithm ID, so hashes are enabled on A80 but
with an incorrect ID.
Anyway, even with correct IDs, hashes do not work on A80 and I cannot
find why.
So let's disable all of them on A80.

Fixes: d9b45418a917 ("crypto: sun8i-ss - support hash algorithms")
Signed-off-by: Corentin Labbe <clabbe@baylibre.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2 years agocrypto: x86 - Convert to SPDX identifier
Nathan Huckleberry [Sat, 15 Jan 2022 00:17:19 +0000 (18:17 -0600)]
crypto: x86 - Convert to SPDX identifier

Use SPDX-License-Identifier instead of a verbose license text and
update external link.

Cc: James Guilford <james.guilford@intel.com>
Cc: Sean Gulley <sean.m.gulley@intel.com>
Cc: Chandramouli Narayanan <mouli@linux.intel.com>
Signed-off-by: Nathan Huckleberry <nhuck@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2 years agocrypto: testmgr - Move crypto_simd_disabled_for_test out
Herbert Xu [Fri, 14 Jan 2022 06:40:30 +0000 (17:40 +1100)]
crypto: testmgr - Move crypto_simd_disabled_for_test out

As testmgr is part of cryptomgr which was designed to be unloadable
as a module, it shouldn't export any symbols for other crypto
modules to use as that would prevent it from being unloaded.  All
its functionality is meant to be accessed through notifiers.

The symbol crypto_simd_disabled_for_test was added to testmgr
which caused it to be pinned as a module if its users were also
loaded.  This patch moves it out of testmgr and into crypto/algapi.c
so cryptomgr can again be unloaded and replaced on demand.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Reviewed-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2 years agohwrng: cavium - HW_RANDOM_CAVIUM should depend on ARCH_THUNDER
Geert Uytterhoeven [Wed, 12 Jan 2022 14:05:03 +0000 (15:05 +0100)]
hwrng: cavium - HW_RANDOM_CAVIUM should depend on ARCH_THUNDER

The Cavium ThunderX Random Number Generator is only present on Cavium
ThunderX SoCs, and not available as an independent PCIe endpoint.  Hence
add a dependency on ARCH_THUNDER, to prevent asking the user about this
driver when configuring a kernel without Cavium Thunder SoC  support.

Fixes: cc2f1908c6b8f625 ("hwrng: cavium - Add Cavium HWRNG driver for ThunderX SoC.")
Signed-off-by: Geert Uytterhoeven <geert+renesas@glider.be>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2 years agocrypto: ccp - remove redundant ret variable
Minghao Chi [Wed, 12 Jan 2022 08:05:44 +0000 (08:05 +0000)]
crypto: ccp - remove redundant ret variable

Return value from ccp_crypto_enqueue_request() directly instead
of taking this in another redundant variable.

Reported-by: Zeal Robot <zealci@zte.com.cn>
Signed-off-by: Minghao Chi <chi.minghao@zte.com.cn>
Signed-off-by: CGEL ZTE <cgel.zte@gmail.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2 years agocrypto: qat - fix a signedness bug in get_service_enabled()
Dan Carpenter [Tue, 11 Jan 2022 07:18:06 +0000 (10:18 +0300)]
crypto: qat - fix a signedness bug in get_service_enabled()

The "ret" variable needs to be signed or there is an error message which
will not be printed correctly.

Fixes: 0cec19c761e5 ("crypto: qat - add support for compression for 4xxx")
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Acked-by: Giovanni Cabiddu <giovanni.cabiddu@intel.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2 years agocrypto: ccp - Ensure psp_ret is always init'd in __sev_platform_init_locked()
Peter Gonda [Mon, 10 Jan 2022 21:18:37 +0000 (13:18 -0800)]
crypto: ccp - Ensure psp_ret is always init'd in __sev_platform_init_locked()

Initialize psp_ret inside of __sev_platform_init_locked() because there
are many failure paths with PSP initialization that do not set
__sev_do_cmd_locked().

Fixes: e423b9d75e77: ("crypto: ccp - Move SEV_INIT retry for corrupted data")

Signed-off-by: Peter Gonda <pgonda@google.com>
Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
Cc: Tom Lendacky <thomas.lendacky@amd.com>
Cc: Brijesh Singh <brijesh.singh@amd.com>
Cc: Marc Orr <marcorr@google.com>
Cc: Herbert Xu <herbert@gondor.apana.org.au>
Cc: John Allen <john.allen@amd.com>
Cc: "David S. Miller" <davem@davemloft.net>
Cc: linux-crypto@vger.kernel.org
Cc: linux-kernel@vger.kernel.org
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2 years agocrypto: tcrypt - add asynchronous speed test for SM3
Tianjia Zhang [Fri, 7 Jan 2022 12:07:00 +0000 (20:07 +0800)]
crypto: tcrypt - add asynchronous speed test for SM3

tcrypt supports testing of SM3 hash algorithms that use AVX
instruction acceleration.

In order to add the sm3 asynchronous test to the appropriate
position, shift the testcase sequence number of the multi buffer
backward and start from 450.

Signed-off-by: Tianjia Zhang <tianjia.zhang@linux.alibaba.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>