platform/kernel/linux-rpi.git
4 years agocrypto: {arm,arm64,mips}/poly1305 - remove redundant non-reduction from emit
Jason A. Donenfeld [Mon, 6 Jan 2020 03:40:49 +0000 (22:40 -0500)]
crypto: {arm,arm64,mips}/poly1305 - remove redundant non-reduction from emit

This appears to be some kind of copy and paste error, and is actually
dead code.

Pre: f = 0 ⇒ (f >> 32) = 0
    f = (f >> 32) + le32_to_cpu(digest[0]);
Post: 0 ≤ f < 2³²
    put_unaligned_le32(f, dst);

Pre: 0 ≤ f < 2³² ⇒ (f >> 32) = 0
    f = (f >> 32) + le32_to_cpu(digest[1]);
Post: 0 ≤ f < 2³²
    put_unaligned_le32(f, dst + 4);

Pre: 0 ≤ f < 2³² ⇒ (f >> 32) = 0
    f = (f >> 32) + le32_to_cpu(digest[2]);
Post: 0 ≤ f < 2³²
    put_unaligned_le32(f, dst + 8);

Pre: 0 ≤ f < 2³² ⇒ (f >> 32) = 0
    f = (f >> 32) + le32_to_cpu(digest[3]);
Post: 0 ≤ f < 2³²
    put_unaligned_le32(f, dst + 12);

Therefore this sequence is redundant. And Andy's code appears to handle
misalignment acceptably.

Signed-off-by: Jason A. Donenfeld <Jason@zx2c4.com>
Tested-by: Ard Biesheuvel <ardb@kernel.org>
Reviewed-by: Ard Biesheuvel <ardb@kernel.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
4 years agocrypto: x86/poly1305 - wire up faster implementations for kernel
Jason A. Donenfeld [Mon, 6 Jan 2020 03:40:48 +0000 (22:40 -0500)]
crypto: x86/poly1305 - wire up faster implementations for kernel

These x86_64 vectorized implementations support AVX, AVX-2, and AVX512F.
The AVX-512F implementation is disabled on Skylake, due to throttling,
but it is quite fast on >= Cannonlake.

On the left is cycle counts on a Core i7 6700HQ using the AVX-2
codepath, comparing this implementation ("new") to the implementation in
the current crypto api ("old"). On the right are benchmarks on a Xeon
Gold 5120 using the AVX-512 codepath. The new implementation is faster
on all benchmarks.

        AVX-2                  AVX-512
      ---------              -----------

    size    old     new      size   old     new
    ----    ----    ----     ----   ----    ----
    0       70      68       0      74      70
    16      92      90       16     96      92
    32      134     104      32     136     106
    48      172     120      48     184     124
    64      218     136      64     218     138
    80      254     158      80     260     160
    96      298     174      96     300     176
    112     342     192      112    342     194
    128     388     212      128    384     212
    144     428     228      144    420     226
    160     466     246      160    464     248
    176     510     264      176    504     264
    192     550     282      192    544     282
    208     594     302      208    582     300
    224     628     316      224    624     318
    240     676     334      240    662     338
    256     716     354      256    708     358
    272     764     374      272    748     372
    288     802     352      288    788     358
    304     420     366      304    422     370
    320     428     360      320    432     364
    336     484     378      336    486     380
    352     426     384      352    434     390
    368     478     400      368    480     408
    384     488     394      384    490     398
    400     542     408      400    542     412
    416     486     416      416    492     426
    432     534     430      432    538     436
    448     544     422      448    546     432
    464     600     438      464    600     448
    480     540     448      480    548     456
    496     594     464      496    594     476
    512     602     456      512    606     470
    528     656     476      528    656     480
    544     600     480      544    606     498
    560     650     494      560    652     512
    576     664     490      576    662     508
    592     714     508      592    716     522
    608     656     514      608    664     538
    624     708     532      624    710     552
    640     716     524      640    720     516
    656     770     536      656    772     526
    672     716     548      672    722     544
    688     770     562      688    768     556
    704     774     552      704    778     556
    720     826     568      720    832     568
    736     768     574      736    780     584
    752     822     592      752    826     600
    768     830     584      768    836     560
    784     884     602      784    888     572
    800     828     610      800    838     588
    816     884     628      816    884     604
    832     888     618      832    894     598
    848     942     632      848    946     612
    864     884     644      864    896     628
    880     936     660      880    942     644
    896     948     652      896    952     608
    912     1000    664      912    1004    616
    928     942     676      928    954     634
    944     994     690      944    1000    646
    960     1002    680      960    1008    646
    976     1054    694      976    1062    658
    992     1002    706      992    1012    674
    1008    1052    720      1008   1058    690

This commit wires in the prior implementation from Andy, and makes the
following changes to be suitable for kernel land.

  - Some cosmetic and structural changes, like renaming labels to
    .Lname, constants, and other Linux conventions, as well as making
    the code easy for us to maintain moving forward.

  - CPU feature checking is done in C by the glue code.

  - We avoid jumping into the middle of functions, to appease objtool,
    and instead parameterize shared code.

  - We maintain frame pointers so that stack traces make sense.

  - We remove the dependency on the perl xlate code, which transforms
    the output into things that assemblers we don't care about use.

Importantly, none of our changes affect the arithmetic or core code, but
just involve the differing environment of kernel space.

Signed-off-by: Jason A. Donenfeld <Jason@zx2c4.com>
Signed-off-by: Samuel Neves <sneves@dei.uc.pt>
Co-developed-by: Samuel Neves <sneves@dei.uc.pt>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
4 years agocrypto: x86/poly1305 - import unmodified cryptogams implementation
Jason A. Donenfeld [Mon, 6 Jan 2020 03:40:47 +0000 (22:40 -0500)]
crypto: x86/poly1305 - import unmodified cryptogams implementation

These x86_64 vectorized implementations come from Andy Polyakov's
CRYPTOGAMS implementation, and are included here in raw form without
modification, so that subsequent commits that fix these up for the
kernel can see how it has changed.

Signed-off-by: Jason A. Donenfeld <Jason@zx2c4.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
4 years agocrypto: poly1305 - add new 32 and 64-bit generic versions
Jason A. Donenfeld [Mon, 6 Jan 2020 03:40:46 +0000 (22:40 -0500)]
crypto: poly1305 - add new 32 and 64-bit generic versions

These two C implementations from Zinc -- a 32x32 one and a 64x64 one,
depending on the platform -- come from Andrew Moon's public domain
poly1305-donna portable code, modified for usage in the kernel. The
precomputation in the 32-bit version and the use of 64x64 multiplies in
the 64-bit version make these perform better than the code it replaces.
Moon's code is also very widespread and has received many eyeballs of
scrutiny.

There's a bit of interference between the x86 implementation, which
relies on internal details of the old scalar implementation. In the next
commit, the x86 implementation will be replaced with a faster one that
doesn't rely on this, so none of this matters much. But for now, to keep
this passing the tests, we inline the bits of the old implementation
that the x86 implementation relied on. Also, since we now support a
slightly larger key space, via the union, some offsets had to be fixed
up.

Nonce calculation was folded in with the emit function, to take
advantage of 64x64 arithmetic. However, Adiantum appeared to rely on no
nonce handling in emit, so this path was conditionalized. We also
introduced a new struct, poly1305_core_key, to represent the precise
amount of space that particular implementation uses.

Testing with kbench9000, depending on the CPU, the update function for
the 32x32 version has been improved by 4%-7%, and for the 64x64 by
19%-30%. The 32x32 gains are small, but I think there's great value in
having a parallel implementation to the 64x64 one so that the two can be
compared side-by-side as nice stand-alone units.

Signed-off-by: Jason A. Donenfeld <Jason@zx2c4.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
4 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/herbert/crypto-2.6
Herbert Xu [Thu, 16 Jan 2020 07:17:08 +0000 (15:17 +0800)]
Merge git://git./linux/kernel/git/herbert/crypto-2.6

Merge crypto tree to pick up hisilicon patch.

4 years agocrypto: exynos-rng - Rename Exynos to lowercase
Krzysztof Kozlowski [Sat, 4 Jan 2020 15:20:59 +0000 (16:20 +0100)]
crypto: exynos-rng - Rename Exynos to lowercase

Fix up inconsistent usage of upper and lowercase letters in "Exynos"
name.

"EXYNOS" is not an abbreviation but a regular trademarked name.
Therefore it should be written with lowercase letters starting with
capital letter.

The lowercase "Exynos" name is promoted by its manufacturer Samsung
Electronics Co., Ltd., in advertisement materials and on website.

Signed-off-by: Krzysztof Kozlowski <krzk@kernel.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
4 years agocrypto: chelsio - Resetting crypto counters during the driver unregister
Ayush Sawal [Fri, 3 Jan 2020 04:56:51 +0000 (10:26 +0530)]
crypto: chelsio - Resetting crypto counters during the driver unregister

Signed-off-by: Ayush Sawal <ayush.sawal@chelsio.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
4 years agocrypto: algapi - enforce that all instances have a ->free() method
Eric Biggers [Fri, 3 Jan 2020 04:04:40 +0000 (20:04 -0800)]
crypto: algapi - enforce that all instances have a ->free() method

All instances need to have a ->free() method, but people could forget to
set it and then not notice if the instance is never unregistered.  To
help detect this bug earlier, don't allow an instance without a ->free()
method to be registered, and complain loudly if someone tries to do it.

Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
4 years agocrypto: algapi - remove crypto_template::{alloc,free}()
Eric Biggers [Fri, 3 Jan 2020 04:04:39 +0000 (20:04 -0800)]
crypto: algapi - remove crypto_template::{alloc,free}()

Now that all templates provide a ->create() method which creates an
instance, installs a strongly-typed ->free() method directly to it, and
registers it, the older ->alloc() and ->free() methods in
'struct crypto_template' are no longer used.  Remove them.

Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
4 years agocrypto: shash - convert shash_free_instance() to new style
Eric Biggers [Fri, 3 Jan 2020 04:04:38 +0000 (20:04 -0800)]
crypto: shash - convert shash_free_instance() to new style

Convert shash_free_instance() and its users to the new way of freeing
instances, where a ->free() method is installed to the instance struct
itself.  This replaces the weakly-typed method crypto_template::free().

This will allow removing support for the old way of freeing instances.

Also give shash_free_instance() a more descriptive name to reflect that
it's only for instances with a single spawn, not for any instance.

Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
4 years agocrypto: cryptd - convert to new way of freeing instances
Eric Biggers [Fri, 3 Jan 2020 04:04:37 +0000 (20:04 -0800)]
crypto: cryptd - convert to new way of freeing instances

Convert the "cryptd" template to the new way of freeing instances, where
a ->free() method is installed to the instance struct itself.  This
replaces the weakly-typed method crypto_template::free().

This will allow removing support for the old way of freeing instances.

Note that the 'default' case in cryptd_free() was already unreachable.
So, we aren't missing anything by keeping only the ahash and aead parts.

Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
4 years agocrypto: geniv - convert to new way of freeing instances
Eric Biggers [Fri, 3 Jan 2020 04:04:36 +0000 (20:04 -0800)]
crypto: geniv - convert to new way of freeing instances

Convert the "seqiv" template to the new way of freeing instances where a
->free() method is installed to the instance struct itself.  Also remove
the unused implementation of the old way of freeing instances from the
"echainiv" template, since it's already using the new way too.

In doing this, also simplify the code by making the helper function
aead_geniv_alloc() install the ->free() method, instead of making seqiv
and echainiv do this themselves.  This is analogous to how
skcipher_alloc_instance_simple() works.

This will allow removing support for the old way of freeing instances.

Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
4 years agocrypto: hash - add support for new way of freeing instances
Eric Biggers [Fri, 3 Jan 2020 04:04:35 +0000 (20:04 -0800)]
crypto: hash - add support for new way of freeing instances

Add support to shash and ahash for the new way of freeing instances
(already used for skcipher, aead, and akcipher) where a ->free() method
is installed to the instance struct itself.  These methods are more
strongly-typed than crypto_template::free(), which they replace.

This will allow removing support for the old way of freeing instances.

Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
4 years agocrypto: algapi - fold crypto_init_spawn() into crypto_grab_spawn()
Eric Biggers [Fri, 3 Jan 2020 03:59:08 +0000 (19:59 -0800)]
crypto: algapi - fold crypto_init_spawn() into crypto_grab_spawn()

Now that crypto_init_spawn() is only called by crypto_grab_spawn(),
simplify things by moving its functionality into crypto_grab_spawn().

In the process of doing this, also be more consistent about when the
spawn and instance are updated, and remove the crypto_spawn::dropref
flag since now it's always set.

Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
4 years agocrypto: ahash - unexport crypto_ahash_type
Eric Biggers [Fri, 3 Jan 2020 03:59:07 +0000 (19:59 -0800)]
crypto: ahash - unexport crypto_ahash_type

Now that all the templates that need ahash spawns have been converted to
use crypto_grab_ahash() rather than look up the algorithm directly,
crypto_ahash_type is no longer used outside of ahash.c.  Make it static.

Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
4 years agocrypto: algapi - remove obsoleted instance creation helpers
Eric Biggers [Fri, 3 Jan 2020 03:59:06 +0000 (19:59 -0800)]
crypto: algapi - remove obsoleted instance creation helpers

Remove lots of helper functions that were previously used for
instantiating crypto templates, but are now unused:

- crypto_get_attr_alg() and similar functions looked up an inner
  algorithm directly from a template parameter.  These were replaced
  with getting the algorithm's name, then calling crypto_grab_*().

- crypto_init_spawn2() and similar functions initialized a spawn, given
  an algorithm.  Similarly, these were replaced with crypto_grab_*().

- crypto_alloc_instance() and similar functions allocated an instance
  with a single spawn, given the inner algorithm.  These aren't useful
  anymore since crypto_grab_*() need the instance allocated first.

Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
4 years agocrypto: cipher - make crypto_spawn_cipher() take a crypto_cipher_spawn
Eric Biggers [Fri, 3 Jan 2020 03:59:05 +0000 (19:59 -0800)]
crypto: cipher - make crypto_spawn_cipher() take a crypto_cipher_spawn

Now that all users of single-block cipher spawns have been converted to
use 'struct crypto_cipher_spawn' rather than the less specifically typed
'struct crypto_spawn', make crypto_spawn_cipher() take a pointer to a
'struct crypto_cipher_spawn' rather than a 'struct crypto_spawn'.

Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
4 years agocrypto: xcbc - use crypto_grab_cipher() and simplify error paths
Eric Biggers [Fri, 3 Jan 2020 03:59:04 +0000 (19:59 -0800)]
crypto: xcbc - use crypto_grab_cipher() and simplify error paths

Make the xcbc template use the new function crypto_grab_cipher() to
initialize its cipher spawn.

This is needed to make all spawns be initialized in a consistent way.

This required making xcbc_create() allocate the instance directly rather
than use shash_alloc_instance().

Also simplify the error handling by taking advantage of crypto_drop_*()
now accepting (as a no-op) spawns that haven't been initialized yet, and
by taking advantage of crypto_grab_*() now handling ERR_PTR() names.

Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
4 years agocrypto: vmac - use crypto_grab_cipher() and simplify error paths
Eric Biggers [Fri, 3 Jan 2020 03:59:03 +0000 (19:59 -0800)]
crypto: vmac - use crypto_grab_cipher() and simplify error paths

Make the vmac64 template use the new function crypto_grab_cipher() to
initialize its cipher spawn.

This is needed to make all spawns be initialized in a consistent way.

This required making vmac_create() allocate the instance directly rather
than use shash_alloc_instance().

Also simplify the error handling by taking advantage of crypto_drop_*()
now accepting (as a no-op) spawns that haven't been initialized yet, and
by taking advantage of crypto_grab_*() now handling ERR_PTR() names.

Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
4 years agocrypto: cmac - use crypto_grab_cipher() and simplify error paths
Eric Biggers [Fri, 3 Jan 2020 03:59:02 +0000 (19:59 -0800)]
crypto: cmac - use crypto_grab_cipher() and simplify error paths

Make the cmac template use the new function crypto_grab_cipher() to
initialize its cipher spawn.

This is needed to make all spawns be initialized in a consistent way.

This required making cmac_create() allocate the instance directly rather
than use shash_alloc_instance().

Also simplify the error handling by taking advantage of crypto_drop_*()
now accepting (as a no-op) spawns that haven't been initialized yet, and
by taking advantage of crypto_grab_*() now handling ERR_PTR() names.

Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
4 years agocrypto: cbcmac - use crypto_grab_cipher() and simplify error paths
Eric Biggers [Fri, 3 Jan 2020 03:59:01 +0000 (19:59 -0800)]
crypto: cbcmac - use crypto_grab_cipher() and simplify error paths

Make the cbcmac template use the new function crypto_grab_cipher() to
initialize its cipher spawn.

This is needed to make all spawns be initialized in a consistent way.

This required making cbcmac_create() allocate the instance directly
rather than use shash_alloc_instance().

Also simplify the error handling by taking advantage of crypto_drop_*()
now accepting (as a no-op) spawns that haven't been initialized yet, and
by taking advantage of crypto_grab_*() now handling ERR_PTR() names.

Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
4 years agocrypto: skcipher - use crypto_grab_cipher() and simplify error paths
Eric Biggers [Fri, 3 Jan 2020 03:59:00 +0000 (19:59 -0800)]
crypto: skcipher - use crypto_grab_cipher() and simplify error paths

Make skcipher_alloc_instance_simple() use the new function
crypto_grab_cipher() to initialize its cipher spawn.

This is needed to make all spawns be initialized in a consistent way.

Also simplify the error handling by taking advantage of crypto_drop_*()
now accepting (as a no-op) spawns that haven't been initialized yet, and
by taking advantage of crypto_grab_*() now handling ERR_PTR() names.

Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
4 years agocrypto: chacha20poly1305 - use crypto_grab_ahash() and simplify error paths
Eric Biggers [Fri, 3 Jan 2020 03:58:59 +0000 (19:58 -0800)]
crypto: chacha20poly1305 - use crypto_grab_ahash() and simplify error paths

Make the rfc7539 and rfc7539esp templates use the new function
crypto_grab_ahash() to initialize their ahash spawn.

This is needed to make all spawns be initialized in a consistent way.

Also simplify the error handling by taking advantage of crypto_drop_*()
now accepting (as a no-op) spawns that haven't been initialized yet, and
by taking advantage of crypto_grab_*() now handling ERR_PTR() names.

Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
4 years agocrypto: ccm - use crypto_grab_ahash() and simplify error paths
Eric Biggers [Fri, 3 Jan 2020 03:58:58 +0000 (19:58 -0800)]
crypto: ccm - use crypto_grab_ahash() and simplify error paths

Make the ccm and ccm_base templates use the new function
crypto_grab_ahash() to initialize their ahash spawn.

This is needed to make all spawns be initialized in a consistent way.

Also simplify the error handling by taking advantage of crypto_drop_*()
now accepting (as a no-op) spawns that haven't been initialized yet.

Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
4 years agocrypto: gcm - use crypto_grab_ahash() and simplify error paths
Eric Biggers [Fri, 3 Jan 2020 03:58:57 +0000 (19:58 -0800)]
crypto: gcm - use crypto_grab_ahash() and simplify error paths

Make the gcm and gcm_base templates use the new function
crypto_grab_ahash() to initialize their ahash spawn.

This is needed to make all spawns be initialized in a consistent way.

Also simplify the error handling by taking advantage of crypto_drop_*()
now accepting (as a no-op) spawns that haven't been initialized yet.

Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
4 years agocrypto: authencesn - use crypto_grab_ahash() and simplify error paths
Eric Biggers [Fri, 3 Jan 2020 03:58:56 +0000 (19:58 -0800)]
crypto: authencesn - use crypto_grab_ahash() and simplify error paths

Make the authencesn template use the new function crypto_grab_ahash() to
initialize its ahash spawn.

This is needed to make all spawns be initialized in a consistent way.

Also simplify the error handling by taking advantage of crypto_drop_*()
now accepting (as a no-op) spawns that haven't been initialized yet, and
by taking advantage of crypto_grab_*() now handling ERR_PTR() names.

Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
4 years agocrypto: authenc - use crypto_grab_ahash() and simplify error paths
Eric Biggers [Fri, 3 Jan 2020 03:58:55 +0000 (19:58 -0800)]
crypto: authenc - use crypto_grab_ahash() and simplify error paths

Make the authenc template use the new function crypto_grab_ahash() to
initialize its ahash spawn.

This is needed to make all spawns be initialized in a consistent way.

Also simplify the error handling by taking advantage of crypto_drop_*()
now accepting (as a no-op) spawns that haven't been initialized yet, and
by taking advantage of crypto_grab_*() now handling ERR_PTR() names.

Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
4 years agocrypto: hmac - use crypto_grab_shash() and simplify error paths
Eric Biggers [Fri, 3 Jan 2020 03:58:54 +0000 (19:58 -0800)]
crypto: hmac - use crypto_grab_shash() and simplify error paths

Make the hmac template use the new function crypto_grab_shash() to
initialize its shash spawn.

This is needed to make all spawns be initialized in a consistent way.

This required making hmac_create() allocate the instance directly rather
than use shash_alloc_instance().

Also simplify the error handling by taking advantage of crypto_drop_*()
now accepting (as a no-op) spawns that haven't been initialized yet, and
by taking advantage of crypto_grab_*() now handling ERR_PTR() names.

Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
4 years agocrypto: cryptd - use crypto_grab_shash() and simplify error paths
Eric Biggers [Fri, 3 Jan 2020 03:58:53 +0000 (19:58 -0800)]
crypto: cryptd - use crypto_grab_shash() and simplify error paths

Make the cryptd template (in the hash case) use the new function
crypto_grab_shash() to initialize its shash spawn.

This is needed to make all spawns be initialized in a consistent way.

This required making cryptd_create_hash() allocate the instance directly
rather than use cryptd_alloc_instance().

Also simplify the error handling by taking advantage of crypto_drop_*()
now accepting (as a no-op) spawns that haven't been initialized yet, and
by taking advantage of crypto_grab_*() now handling ERR_PTR() names.

Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
4 years agocrypto: adiantum - use crypto_grab_{cipher,shash} and simplify error paths
Eric Biggers [Fri, 3 Jan 2020 03:58:52 +0000 (19:58 -0800)]
crypto: adiantum - use crypto_grab_{cipher,shash} and simplify error paths

Make the adiantum template use the new functions crypto_grab_cipher()
and crypto_grab_shash() to initialize its cipher and shash spawns.

This is needed to make all spawns be initialized in a consistent way.

Also simplify the error handling by taking advantage of crypto_drop_*()
now accepting (as a no-op) spawns that haven't been initialized yet, and
by taking advantage of crypto_grab_*() now handling ERR_PTR() names.

Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
4 years agocrypto: cipher - introduce crypto_cipher_spawn and crypto_grab_cipher()
Eric Biggers [Fri, 3 Jan 2020 03:58:51 +0000 (19:58 -0800)]
crypto: cipher - introduce crypto_cipher_spawn and crypto_grab_cipher()

Currently, "cipher" (single-block cipher) spawns are usually initialized
by using crypto_get_attr_alg() to look up the algorithm, then calling
crypto_init_spawn().  In one case, crypto_grab_spawn() is used directly.

The former way is different from how skcipher, aead, and akcipher spawns
are initialized (they use crypto_grab_*()), and for no good reason.
This difference introduces unnecessary complexity.

The crypto_grab_*() functions used to have some problems, like not
holding a reference to the algorithm and requiring the caller to
initialize spawn->base.inst.  But those problems are fixed now.

Also, the cipher spawns are not strongly typed; e.g., the API requires
that the user manually specify the flags CRYPTO_ALG_TYPE_CIPHER and
CRYPTO_ALG_TYPE_MASK.  Though the "cipher" algorithm type itself isn't
yet strongly typed, we can start by making the spawns strongly typed.

So, let's introduce a new 'struct crypto_cipher_spawn', and functions
crypto_grab_cipher() and crypto_drop_cipher() to grab and drop them.

Later patches will convert all cipher spawns to use these, then make
crypto_spawn_cipher() take 'struct crypto_cipher_spawn' as well, instead
of a bare 'struct crypto_spawn' as it currently does.

Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
4 years agocrypto: ahash - introduce crypto_grab_ahash()
Eric Biggers [Fri, 3 Jan 2020 03:58:50 +0000 (19:58 -0800)]
crypto: ahash - introduce crypto_grab_ahash()

Currently, ahash spawns are initialized by using ahash_attr_alg() or
crypto_find_alg() to look up the ahash algorithm, then calling
crypto_init_ahash_spawn().

This is different from how skcipher, aead, and akcipher spawns are
initialized (they use crypto_grab_*()), and for no good reason.  This
difference introduces unnecessary complexity.

The crypto_grab_*() functions used to have some problems, like not
holding a reference to the algorithm and requiring the caller to
initialize spawn->base.inst.  But those problems are fixed now.

So, let's introduce crypto_grab_ahash() so that we can convert all
templates to the same way of initializing their spawns.

Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
4 years agocrypto: shash - introduce crypto_grab_shash()
Eric Biggers [Fri, 3 Jan 2020 03:58:49 +0000 (19:58 -0800)]
crypto: shash - introduce crypto_grab_shash()

Currently, shash spawns are initialized by using shash_attr_alg() or
crypto_alg_mod_lookup() to look up the shash algorithm, then calling
crypto_init_shash_spawn().

This is different from how skcipher, aead, and akcipher spawns are
initialized (they use crypto_grab_*()), and for no good reason.  This
difference introduces unnecessary complexity.

The crypto_grab_*() functions used to have some problems, like not
holding a reference to the algorithm and requiring the caller to
initialize spawn->base.inst.  But those problems are fixed now.

So, let's introduce crypto_grab_shash() so that we can convert all
templates to the same way of initializing their spawns.

Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
4 years agocrypto: algapi - pass instance to crypto_grab_spawn()
Eric Biggers [Fri, 3 Jan 2020 03:58:48 +0000 (19:58 -0800)]
crypto: algapi - pass instance to crypto_grab_spawn()

Currently, crypto_spawn::inst is first used temporarily to pass the
instance to crypto_grab_spawn().  Then crypto_init_spawn() overwrites it
with crypto_spawn::next, which shares the same union.  Finally,
crypto_spawn::inst is set again when the instance is registered.

Make this less convoluted by just passing the instance as an argument to
crypto_grab_spawn() instead.

Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
4 years agocrypto: akcipher - pass instance to crypto_grab_akcipher()
Eric Biggers [Fri, 3 Jan 2020 03:58:47 +0000 (19:58 -0800)]
crypto: akcipher - pass instance to crypto_grab_akcipher()

Initializing a crypto_akcipher_spawn currently requires:

1. Set spawn->base.inst to point to the instance.
2. Call crypto_grab_akcipher().

But there's no reason for these steps to be separate, and in fact this
unneeded complication has caused at least one bug, the one fixed by
commit 6db43410179b ("crypto: adiantum - initialize crypto_spawn::inst")

So just make crypto_grab_akcipher() take the instance as an argument.

To keep the function call from getting too unwieldy due to this extra
argument, also introduce a 'mask' variable into pkcs1pad_create().

Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
4 years agocrypto: aead - pass instance to crypto_grab_aead()
Eric Biggers [Fri, 3 Jan 2020 03:58:46 +0000 (19:58 -0800)]
crypto: aead - pass instance to crypto_grab_aead()

Initializing a crypto_aead_spawn currently requires:

1. Set spawn->base.inst to point to the instance.
2. Call crypto_grab_aead().

But there's no reason for these steps to be separate, and in fact this
unneeded complication has caused at least one bug, the one fixed by
commit 6db43410179b ("crypto: adiantum - initialize crypto_spawn::inst")

So just make crypto_grab_aead() take the instance as an argument.

To keep the function calls from getting too unwieldy due to this extra
argument, also introduce a 'mask' variable into the affected places
which weren't already using one.

Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
4 years agocrypto: skcipher - pass instance to crypto_grab_skcipher()
Eric Biggers [Fri, 3 Jan 2020 03:58:45 +0000 (19:58 -0800)]
crypto: skcipher - pass instance to crypto_grab_skcipher()

Initializing a crypto_skcipher_spawn currently requires:

1. Set spawn->base.inst to point to the instance.
2. Call crypto_grab_skcipher().

But there's no reason for these steps to be separate, and in fact this
unneeded complication has caused at least one bug, the one fixed by
commit 6db43410179b ("crypto: adiantum - initialize crypto_spawn::inst")

So just make crypto_grab_skcipher() take the instance as an argument.

To keep the function calls from getting too unwieldy due to this extra
argument, also introduce a 'mask' variable into the affected places
which weren't already using one.

Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
4 years agocrypto: ahash - make struct ahash_instance be the full size
Eric Biggers [Fri, 3 Jan 2020 03:58:44 +0000 (19:58 -0800)]
crypto: ahash - make struct ahash_instance be the full size

Define struct ahash_instance in a way analogous to struct
skcipher_instance, struct aead_instance, and struct akcipher_instance,
where the struct is defined to include both the algorithm structure at
the beginning and the additional crypto_instance fields at the end.

This is needed to allow allocating ahash instances directly using
kzalloc(sizeof(*inst) + sizeof(*ictx), ...) in the same way as skcipher,
aead, and akcipher instances.  In turn, that's needed to make spawns be
initialized in a consistent way everywhere.

Also take advantage of the addition of the base instance to struct
ahash_instance by simplifying the ahash_crypto_instance() and
ahash_instance() functions.

Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
4 years agocrypto: shash - make struct shash_instance be the full size
Eric Biggers [Fri, 3 Jan 2020 03:58:43 +0000 (19:58 -0800)]
crypto: shash - make struct shash_instance be the full size

Define struct shash_instance in a way analogous to struct
skcipher_instance, struct aead_instance, and struct akcipher_instance,
where the struct is defined to include both the algorithm structure at
the beginning and the additional crypto_instance fields at the end.

This is needed to allow allocating shash instances directly using
kzalloc(sizeof(*inst) + sizeof(*ictx), ...) in the same way as skcipher,
aead, and akcipher instances.  In turn, that's needed to make spawns be
initialized in a consistent way everywhere.

Also take advantage of the addition of the base instance to struct
shash_instance by simplifying the shash_crypto_instance() and
shash_instance() functions.

Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
4 years agocrypto: algapi - make crypto_grab_spawn() handle an ERR_PTR() name
Eric Biggers [Fri, 3 Jan 2020 03:58:42 +0000 (19:58 -0800)]
crypto: algapi - make crypto_grab_spawn() handle an ERR_PTR() name

To allow further simplifying template ->create() functions, make
crypto_grab_spawn() handle an ERR_PTR() name by passing back the error.

For most templates, this will allow the result of crypto_attr_alg_name()
to be passed directly to crypto_grab_*(), rather than first having to
assign it to a variable [where it can then potentially be misused, as it
was in the rfc7539 template prior to commit 5e27f38f1f3f ("crypto:
chacha20poly1305 - set cra_name correctly")] and check it for error.

Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
4 years agocrypto: algapi - make crypto_drop_spawn() a no-op on uninitialized spawns
Eric Biggers [Fri, 3 Jan 2020 03:58:41 +0000 (19:58 -0800)]
crypto: algapi - make crypto_drop_spawn() a no-op on uninitialized spawns

Make crypto_drop_spawn() do nothing when the spawn hasn't been
initialized with an algorithm yet.  This will allow simplifying error
handling in all the template ->create() functions, since on error they
will be able to just call their usual "free instance" function, rather
than having to handle dropping just the spawns that have been
initialized so far.

This does assume the spawn starts out zero-filled, but that's always the
case since instances are allocated with kzalloc().  And some other code
already assumes this anyway.

Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
4 years agocrypto: ccp - Update MAINTAINERS for CCP driver
Gary R Hook [Thu, 2 Jan 2020 19:57:03 +0000 (13:57 -0600)]
crypto: ccp - Update MAINTAINERS for CCP driver

Remove Gary R Hook as CCP maintainer.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
4 years agocrypto: crypto4xx - use GFP_KERNEL for big allocations
Christian Lamparter [Wed, 1 Jan 2020 22:27:02 +0000 (23:27 +0100)]
crypto: crypto4xx - use GFP_KERNEL for big allocations

The driver should use GFP_KERNEL for the bigger allocation
during the driver's crypto4xx_probe() and not GFP_ATOMIC in
my opinion.

Signed-off-by: Christian Lamparter <chunkeey@gmail.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
4 years agocrypto: crypto4xx - reduce memory fragmentation
Christian Lamparter [Wed, 1 Jan 2020 22:27:01 +0000 (23:27 +0100)]
crypto: crypto4xx - reduce memory fragmentation

With recent kernels (>5.2), the driver fails to probe, as the
allocation of the driver's scatter buffer fails with -ENOMEM.

This happens in crypto4xx_build_sdr(). Where the driver tries
to get 512KiB (=PPC4XX_SD_BUFFER_SIZE * PPC4XX_NUM_SD) of
continuous memory. This big chunk is by design, since the driver
uses this circumstance in the crypto4xx_copy_pkt_to_dst() to
its advantage:
"all scatter-buffers are all neatly organized in one big
continuous ringbuffer; So scatterwalk_map_and_copy() can be
instructed to copy a range of buffers in one go."

The PowerPC arch does not have support for DMA_CMA. Hence,
this patch reorganizes the order in which the memory
allocations are done. Since the driver itself is responsible
for some of the issues.

Signed-off-by: Christian Lamparter <chunkeey@gmail.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
4 years agocrypto: remove propagation of CRYPTO_TFM_RES_* flags
Eric Biggers [Tue, 31 Dec 2019 03:19:38 +0000 (21:19 -0600)]
crypto: remove propagation of CRYPTO_TFM_RES_* flags

The CRYPTO_TFM_RES_* flags were apparently meant as a way to make the
->setkey() functions provide more information about errors.  But these
flags weren't actually being used or tested, and in many cases they
weren't being set correctly anyway.  So they've now been removed.

Also, if someone ever actually needs to start better distinguishing
->setkey() errors (which is somewhat unlikely, as this has been unneeded
for a long time), we'd be much better off just defining different return
values, like -EINVAL if the key is invalid for the algorithm vs.
-EKEYREJECTED if the key was rejected by a policy like "no weak keys".
That would be much simpler, less error-prone, and easier to test.

So just remove CRYPTO_TFM_RES_MASK and all the unneeded logic that
propagates these flags around.

Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
4 years agocrypto: remove CRYPTO_TFM_RES_WEAK_KEY
Eric Biggers [Tue, 31 Dec 2019 03:19:37 +0000 (21:19 -0600)]
crypto: remove CRYPTO_TFM_RES_WEAK_KEY

The CRYPTO_TFM_RES_WEAK_KEY flag was apparently meant as a way to make
the ->setkey() functions provide more information about errors.

However, no one actually checks for this flag, which makes it pointless.
There are also no tests that verify that all algorithms actually set (or
don't set) it correctly.

This is also the last remaining CRYPTO_TFM_RES_* flag, which means that
it's the only thing still needing all the boilerplate code which
propagates these flags around from child => parent tfms.

And if someone ever needs to distinguish this error in the future (which
is somewhat unlikely, as it's been unneeded for a long time), it would
be much better to just define a new return value like -EKEYREJECTED.
That would be much simpler, less error-prone, and easier to test.

So just remove this flag.

Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
4 years agocrypto: remove CRYPTO_TFM_RES_BAD_KEY_LEN
Eric Biggers [Tue, 31 Dec 2019 03:19:36 +0000 (21:19 -0600)]
crypto: remove CRYPTO_TFM_RES_BAD_KEY_LEN

The CRYPTO_TFM_RES_BAD_KEY_LEN flag was apparently meant as a way to
make the ->setkey() functions provide more information about errors.

However, no one actually checks for this flag, which makes it pointless.

Also, many algorithms fail to set this flag when given a bad length key.
Reviewing just the generic implementations, this is the case for
aes-fixed-time, cbcmac, echainiv, nhpoly1305, pcrypt, rfc3686, rfc4309,
rfc7539, rfc7539esp, salsa20, seqiv, and xcbc.  But there are probably
many more in arch/*/crypto/ and drivers/crypto/.

Some algorithms can even set this flag when the key is the correct
length.  For example, authenc and authencesn set it when the key payload
is malformed in any way (not just a bad length), the atmel-sha and ccree
drivers can set it if a memory allocation fails, and the chelsio driver
sets it for bad auth tag lengths, not just bad key lengths.

So even if someone actually wanted to start checking this flag (which
seems unlikely, since it's been unused for a long time), there would be
a lot of work needed to get it working correctly.  But it would probably
be much better to go back to the drawing board and just define different
return values, like -EINVAL if the key is invalid for the algorithm vs.
-EKEYREJECTED if the key was rejected by a policy like "no weak keys".
That would be much simpler, less error-prone, and easier to test.

So just remove this flag.

Signed-off-by: Eric Biggers <ebiggers@google.com>
Reviewed-by: Horia Geantă <horia.geanta@nxp.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
4 years agocrypto: remove CRYPTO_TFM_RES_BAD_BLOCK_LEN
Eric Biggers [Tue, 31 Dec 2019 03:19:35 +0000 (21:19 -0600)]
crypto: remove CRYPTO_TFM_RES_BAD_BLOCK_LEN

The flag CRYPTO_TFM_RES_BAD_BLOCK_LEN is never checked for, and it's
only set by one driver.  And even that single driver's use is wrong
because the driver is setting the flag from ->encrypt() and ->decrypt()
with no locking, which is unsafe because ->encrypt() and ->decrypt() can
be executed by many threads in parallel on the same tfm.

Just remove this flag.

Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
4 years agocrypto: remove unused tfm result flags
Eric Biggers [Tue, 31 Dec 2019 03:19:34 +0000 (21:19 -0600)]
crypto: remove unused tfm result flags

The tfm result flags CRYPTO_TFM_RES_BAD_KEY_SCHED and
CRYPTO_TFM_RES_BAD_FLAGS are never used, so remove them.

Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
4 years agocrypto: atmel-sha - fix error handling when setting hmac key
Eric Biggers [Tue, 31 Dec 2019 03:19:33 +0000 (21:19 -0600)]
crypto: atmel-sha - fix error handling when setting hmac key

HMAC keys can be of any length, and atmel_sha_hmac_key_set() can only
fail due to -ENOMEM.  But atmel_sha_hmac_setkey() incorrectly treated
any error as a "bad key length" error.  Fix it to correctly propagate
the -ENOMEM error code and not set any tfm result flags.

Fixes: 81d8750b2b59 ("crypto: atmel-sha - add support to hmac(shaX)")
Cc: Nicolas Ferre <nicolas.ferre@microchip.com>
Cc: Alexandre Belloni <alexandre.belloni@bootlin.com>
Cc: Ludovic Desroches <ludovic.desroches@microchip.com>
Signed-off-by: Eric Biggers <ebiggers@google.com>
Reviewed-by: Tudor Ambarus <tudor.ambarus@microchip.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
4 years agocrypto: artpec6 - return correct error code for failed setkey()
Eric Biggers [Tue, 31 Dec 2019 03:19:32 +0000 (21:19 -0600)]
crypto: artpec6 - return correct error code for failed setkey()

->setkey() is supposed to retun -EINVAL for invalid key lengths, not -1.

Fixes: a21eb94fc4d3 ("crypto: axis - add ARTPEC-6/7 crypto accelerator driver")
Cc: Jesper Nilsson <jesper.nilsson@axis.com>
Cc: Lars Persson <lars.persson@axis.com>
Signed-off-by: Eric Biggers <ebiggers@google.com>
Acked-by: Lars Persson <lars.persson@axis.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
4 years agocrypto: chelsio - fix writing tfm flags to wrong place
Eric Biggers [Tue, 31 Dec 2019 03:19:31 +0000 (21:19 -0600)]
crypto: chelsio - fix writing tfm flags to wrong place

The chelsio crypto driver is casting 'struct crypto_aead' directly to
'struct crypto_tfm', which is incorrect because the crypto_tfm isn't the
first field of 'struct crypto_aead'.  Consequently, the calls to
crypto_tfm_set_flags() are modifying some other field in the struct.

Also, the driver is setting CRYPTO_TFM_RES_BAD_KEY_LEN in
->setauthsize(), not just in ->setkey().  This is incorrect since this
flag is for bad key lengths, not for bad authentication tag lengths.

Fix these bugs by removing the broken crypto_tfm_set_flags() calls from
->setauthsize() and by fixing them in ->setkey().

Fixes: 324429d74127 ("chcr: Support for Chelsio's Crypto Hardware")
Cc: <stable@vger.kernel.org> # v4.9+
Cc: Atul Gupta <atul.gupta@chelsio.com>
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
4 years agocrypto: skcipher - remove skcipher_walk_aead()
Eric Biggers [Mon, 30 Dec 2019 19:41:15 +0000 (13:41 -0600)]
crypto: skcipher - remove skcipher_walk_aead()

skcipher_walk_aead() is unused and is identical to
skcipher_walk_aead_encrypt(), so remove it.

Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
4 years agocrypto: hisilicon/sec2 - Use atomics instead of __sync
Arnd Bergmann [Tue, 7 Jan 2020 20:08:58 +0000 (21:08 +0100)]
crypto: hisilicon/sec2 - Use atomics instead of __sync

The use of __sync functions for atomic memory access is not
supported in the kernel, and can result in a link error depending
on configuration:

ERROR: "__tsan_atomic32_compare_exchange_strong" [drivers/crypto/hisilicon/sec2/hisi_sec2.ko] undefined!
ERROR: "__tsan_atomic64_fetch_add" [drivers/crypto/hisilicon/sec2/hisi_sec2.ko] undefined!

Use the kernel's own atomic interfaces instead. This way the
debugfs interface actually reads the counter atomically.

Fixes: 416d82204df4 ("crypto: hisilicon - add HiSilicon SEC V2 driver")
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
4 years agoDocumentation: tee: add AMD-TEE driver details
Rijo Thomas [Fri, 27 Dec 2019 05:24:03 +0000 (10:54 +0530)]
Documentation: tee: add AMD-TEE driver details

Update tee.txt with AMD-TEE driver details. The driver is written
to communicate with AMD's TEE.

Acked-by: Jens Wiklander <jens.wiklander@linaro.org>
Co-developed-by: Devaraj Rangasamy <Devaraj.Rangasamy@amd.com>
Signed-off-by: Devaraj Rangasamy <Devaraj.Rangasamy@amd.com>
Signed-off-by: Rijo Thomas <Rijo-john.Thomas@amd.com>
Reviewed-by: Gary R Hook <gary.hook@amd.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
4 years agotee: amdtee: check TEE status during driver initialization
Rijo Thomas [Fri, 27 Dec 2019 05:24:02 +0000 (10:54 +0530)]
tee: amdtee: check TEE status during driver initialization

The AMD-TEE driver should check if TEE is available before
registering itself with TEE subsystem. This ensures that
there is a TEE which the driver can talk to before proceeding
with tee device node allocation.

Cc: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Cc: Tom Lendacky <thomas.lendacky@amd.com>
Acked-by: Jens Wiklander <jens.wiklander@linaro.org>
Co-developed-by: Devaraj Rangasamy <Devaraj.Rangasamy@amd.com>
Signed-off-by: Devaraj Rangasamy <Devaraj.Rangasamy@amd.com>
Signed-off-by: Rijo Thomas <Rijo-john.Thomas@amd.com>
Reviewed-by: Gary R Hook <gary.hook@amd.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
4 years agotee: add AMD-TEE driver
Rijo Thomas [Fri, 27 Dec 2019 05:24:01 +0000 (10:54 +0530)]
tee: add AMD-TEE driver

Adds AMD-TEE driver.
* targets AMD APUs which has AMD Secure Processor with software-based
  Trusted Execution Environment (TEE) support
* registers with TEE subsystem
* defines tee_driver_ops function callbacks
* kernel allocated memory is used as shared memory between normal
  world and secure world.
* acts as REE (Rich Execution Environment) communication agent, which
  uses the services of AMD Secure Processor driver to submit commands
  for processing in TEE environment

Cc: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Cc: Tom Lendacky <thomas.lendacky@amd.com>
Acked-by: Jens Wiklander <jens.wiklander@linaro.org>
Co-developed-by: Devaraj Rangasamy <Devaraj.Rangasamy@amd.com>
Signed-off-by: Devaraj Rangasamy <Devaraj.Rangasamy@amd.com>
Signed-off-by: Rijo Thomas <Rijo-john.Thomas@amd.com>
Reviewed-by: Gary R Hook <gary.hook@amd.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
4 years agotee: allow compilation of tee subsystem for AMD CPUs
Rijo Thomas [Fri, 27 Dec 2019 05:24:00 +0000 (10:54 +0530)]
tee: allow compilation of tee subsystem for AMD CPUs

Allow compilation of tee subsystem for AMD's CPUs which have a dedicated
AMD Secure Processor for Trusted Execution Environment (TEE).

Acked-by: Jens Wiklander <jens.wiklander@linaro.org>
Co-developed-by: Devaraj Rangasamy <Devaraj.Rangasamy@amd.com>
Signed-off-by: Devaraj Rangasamy <Devaraj.Rangasamy@amd.com>
Signed-off-by: Rijo Thomas <Rijo-john.Thomas@amd.com>
Reviewed-by: Gary R Hook <gary.hook@amd.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
4 years agocrypto: qce - allow building only hashes/ciphers
Eneas U de Queiroz [Fri, 20 Dec 2019 19:02:18 +0000 (16:02 -0300)]
crypto: qce - allow building only hashes/ciphers

Allow the user to choose whether to build support for all algorithms
(default), hashes-only, or skciphers-only.

The QCE engine does not appear to scale as well as the CPU to handle
multiple crypto requests.  While the ipq40xx chips have 4-core CPUs, the
QCE handles only 2 requests in parallel.

Ipsec throughput seems to improve when disabling either family of
algorithms, sharing the load with the CPU.  Enabling skciphers-only
appears to work best.

Signed-off-by: Eneas U de Queiroz <cotequeiroz@gmail.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
4 years agocrypto: qce - initialize fallback only for AES
Eneas U de Queiroz [Fri, 20 Dec 2019 19:02:17 +0000 (16:02 -0300)]
crypto: qce - initialize fallback only for AES

Adjust cra_flags to add CRYPTO_NEED_FALLBACK only for AES ciphers, where
AES-192 is not handled by the qce hardware, and don't allocate & free
the fallback skcipher for other algorithms.

Signed-off-by: Eneas U de Queiroz <cotequeiroz@gmail.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
4 years agocrypto: qce - update the skcipher IV
Eneas U de Queiroz [Fri, 20 Dec 2019 19:02:16 +0000 (16:02 -0300)]
crypto: qce - update the skcipher IV

Update the IV after the completion of each cipher operation.

Signed-off-by: Eneas U de Queiroz <cotequeiroz@gmail.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
4 years agocrypto: qce - save a sg table slot for result buf
Eneas U de Queiroz [Fri, 20 Dec 2019 19:02:15 +0000 (16:02 -0300)]
crypto: qce - save a sg table slot for result buf

When ctr-aes-qce is used for gcm-mode, an extra sg entry for the
authentication tag is present, causing trouble when the qce driver
prepares the dst-results sg table for dma.

It computes the number of entries needed with sg_nents_for_len, leaving
out the tag entry.  Then it creates a sg table with that number plus
one, used to store a result buffer.

When copying the sg table, there's no limit to the number of entries
copied, so the extra slot is filled with the authentication tag sg.
When the driver tries to add the result sg, the list is full, and it
returns EINVAL.

By limiting the number of sg entries copied to the dest table, the slot
for the result buffer is guaranteed to be unused.

Signed-off-by: Eneas U de Queiroz <cotequeiroz@gmail.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
4 years agocrypto: qce - fix xts-aes-qce key sizes
Eneas U de Queiroz [Fri, 20 Dec 2019 19:02:14 +0000 (16:02 -0300)]
crypto: qce - fix xts-aes-qce key sizes

XTS-mode uses two keys, so the keysizes should be doubled in
skcipher_def, and halved when checking if it is AES-128/192/256.

Signed-off-by: Eneas U de Queiroz <cotequeiroz@gmail.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
4 years agocrypto: qce - fix ctr-aes-qce block, chunk sizes
Eneas U de Queiroz [Fri, 20 Dec 2019 19:02:13 +0000 (16:02 -0300)]
crypto: qce - fix ctr-aes-qce block, chunk sizes

Set blocksize of ctr-aes-qce to 1, so it can operate as a stream cipher,
adding the definition for chucksize instead, where the underlying block
size belongs.

Signed-off-by: Eneas U de Queiroz <cotequeiroz@gmail.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
4 years agocrypto: skcipher - Add skcipher_ialg_simple helper
Herbert Xu [Fri, 20 Dec 2019 05:29:40 +0000 (13:29 +0800)]
crypto: skcipher - Add skcipher_ialg_simple helper

This patch introduces the skcipher_ialg_simple helper which fetches
the crypto_alg structure from a simple skcipher instance's spawn.

This allows us to remove the third argument from the function
skcipher_alloc_instance_simple.

In doing so the reference count to the algorithm is now maintained
by the Crypto API and the caller no longer needs to drop the alg
refcount.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
4 years agocrypto: chtls - Fixed memory leak
Vinay Kumar Yadav [Thu, 19 Dec 2019 10:51:48 +0000 (16:21 +0530)]
crypto: chtls - Fixed memory leak

Freed work request skbs when connection terminates.
enqueue_wr()/ dequeue_wr() is shared between softirq
and application contexts, should be protected by socket
lock. Moved dequeue_wr() to appropriate file.

Signed-off-by: Vinay Kumar Yadav <vinay.yadav@chelsio.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
4 years agocrypto: chtls - Add support for AES256-GCM based ciphers
Vinay Kumar Yadav [Thu, 19 Dec 2019 10:51:47 +0000 (16:21 +0530)]
crypto: chtls - Add support for AES256-GCM based ciphers

Added support to set 256 bit key to the hardware from
setsockopt for AES256-GCM based ciphers.

Signed-off-by: Vinay Kumar Yadav <vinay.yadav@chelsio.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
4 years agocrypto: caam - Add support for i.MX8M Mini
Adam Ford [Wed, 18 Dec 2019 13:06:14 +0000 (07:06 -0600)]
crypto: caam - Add support for i.MX8M Mini

The i.MX8M Mini uses the same crypto engine as the i.MX8MQ, but
the driver is restricting the check to just the i.MX8MQ.

This patch expands the check for either i.MX8MQ or i.MX8MM.

Signed-off-by: Adam Ford <aford173@gmail.com>
Tested-by: Iuliana Prodan <iuliana.prodan@nxp.com>
Reviewed-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>
4 years agocrypto: api - Retain alg refcount in crypto_grab_spawn
Herbert Xu [Wed, 18 Dec 2019 07:53:01 +0000 (15:53 +0800)]
crypto: api - Retain alg refcount in crypto_grab_spawn

This patch changes crypto_grab_spawn to retain the reference count
on the algorithm.  This is because the caller needs to access the
algorithm parameters and without the reference count the algorithm
can be freed at any time.

The reference count will be subsequently dropped by the crypto API
once the instance has been registered.  The helper crypto_drop_spawn
will also conditionally drop the reference count depending on whether
it has been registered.

Note that the code is actually added to crypto_init_spawn.  However,
unless the caller activates this by setting spawn->dropref beforehand
then nothing happens.  The only caller that sets dropref is currently
crypto_grab_spawn.

Once all legacy users of crypto_init_spawn disappear, then we can
kill the dropref flag.

Internally each instance will maintain a list of its spawns prior
to registration.  This memory used by this list is shared with
other fields that are only used after registration.  In order for
this to work a new flag spawn->registered is added to indicate
whether spawn->inst can be used.

Fixes: d6ef2f198d4c ("crypto: api - Add crypto_grab_spawn primitive")
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
4 years agocrypto: sun4i-ss - make unexported sun4i_ss_pm_ops static
Ben Dooks (Codethink) [Tue, 17 Dec 2019 11:30:24 +0000 (11:30 +0000)]
crypto: sun4i-ss - make unexported sun4i_ss_pm_ops static

The sun4i_ss_pm_ops is not referenced outside the driver
except via a pointer, so make it static to avoid the following
warning:

drivers/crypto/allwinner/sun4i-ss/sun4i-ss-core.c:276:25: warning: symbol 'sun4i_ss_pm_ops' was not declared. Should it be static?

Signed-off-by: Ben Dooks (Codethink) <ben.dooks@codethink.co.uk>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
4 years agocrypto: stm32/hash - Use dma_request_chan() instead dma_request_slave_channel()
Peter Ujfalusi [Tue, 17 Dec 2019 07:35:06 +0000 (09:35 +0200)]
crypto: stm32/hash - Use dma_request_chan() instead dma_request_slave_channel()

dma_request_slave_channel() is a wrapper on top of dma_request_chan()
eating up the error code.

By using dma_request_chan() directly the driver can support deferred
probing against DMA.

Signed-off-by: Peter Ujfalusi <peter.ujfalusi@ti.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
4 years agocrypto: img-hash - Use dma_request_chan instead dma_request_slave_channel
Peter Ujfalusi [Tue, 17 Dec 2019 07:33:05 +0000 (09:33 +0200)]
crypto: img-hash - Use dma_request_chan instead dma_request_slave_channel

dma_request_slave_channel() is a wrapper on top of dma_request_chan()
eating up the error code.

By using dma_request_chan() directly the driver can support deferred
probing against DMA.

Signed-off-by: Peter Ujfalusi <peter.ujfalusi@ti.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
4 years agocrypto: lib/curve25519 - re-add selftests
Jason A. Donenfeld [Mon, 16 Dec 2019 18:53:26 +0000 (19:53 +0100)]
crypto: lib/curve25519 - re-add selftests

Somehow these were dropped when Zinc was being integrated, which is
problematic, because testing the library interface for Curve25519 is
important.. This commit simply adds them back and wires them in in the
same way that the blake2s selftests are wired in.

Signed-off-by: Jason A. Donenfeld <Jason@zx2c4.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
4 years agocrypto: api - remove unneeded semicolon
Chen Zhou [Mon, 16 Dec 2019 10:58:48 +0000 (18:58 +0800)]
crypto: api - remove unneeded semicolon

Fixes coccicheck warning:

./include/linux/crypto.h:573:2-3: Unneeded semicolon

Signed-off-by: Chen Zhou <chenzhou10@huawei.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
4 years agocrypto: allwinner - remove unneeded semicolon
Chen Zhou [Mon, 16 Dec 2019 10:57:04 +0000 (18:57 +0800)]
crypto: allwinner - remove unneeded semicolon

Fixes coccicheck warning:

./drivers/crypto/allwinner/sun8i-ce/sun8i-ce-core.c:558:52-53: Unneeded semicolon

Signed-off-by: Chen Zhou <chenzhou10@huawei.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: algapi - make unregistration functions return void
Eric Biggers [Sun, 15 Dec 2019 23:51:19 +0000 (15:51 -0800)]
crypto: algapi - make unregistration functions return void

Some of the algorithm unregistration functions return -ENOENT when asked
to unregister a non-registered algorithm, while others always return 0
or always return void.  But no users check the return value, except for
two of the bulk unregistration functions which print a message on error
but still always return 0 to their caller, and crypto_del_alg() which
calls crypto_unregister_instance() which always returns 0.

Since unregistering a non-registered algorithm is always a kernel bug
but there isn't anything callers should do to handle this situation at
runtime, let's simplify things by making all the unregistration
functions return void, and moving the error message into
crypto_unregister_alg() and upgrading it to a WARN().

Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: arm64 - Use modern annotations for assembly functions
Mark Brown [Fri, 13 Dec 2019 15:49:10 +0000 (15:49 +0000)]
crypto: arm64 - Use modern annotations for assembly functions

In an effort to clarify and simplify the annotation of assembly functions
in the kernel new macros have been introduced. These replace ENTRY and
ENDPROC and also add a new annotation for static functions which previously
had no ENTRY equivalent. Update the annotations in the crypto code to the
new macros.

There are a small number of files imported from OpenSSL where the assembly
is generated using perl programs, these are not currently annotated at all
and have not been modified.

Signed-off-by: Mark Brown <broonie@kernel.org>
Acked-by: Ard Biesheuvel <ardb@kernel.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: atmel-aes - Fix CTR counter overflow when multiple fragments
Tudor Ambarus [Fri, 13 Dec 2019 14:45:44 +0000 (14:45 +0000)]
crypto: atmel-aes - Fix CTR counter overflow when multiple fragments

The CTR transfer works in fragments of data of maximum 1 MByte because
of the 16 bit CTR counter embedded in the IP. Fix the CTR counter
overflow handling for messages larger than 1 MByte.

Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
Fixes: 781a08d9740a ("crypto: atmel-aes - Fix counter overflow in CTR mode")
Signed-off-by: Tudor Ambarus <tudor.ambarus@microchip.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: chelsio - calculating tx_channel_id as per the max number of channels
Ayush Sawal [Fri, 13 Dec 2019 11:38:52 +0000 (17:08 +0530)]
crypto: chelsio - calculating tx_channel_id as per the max number of channels

chcr driver was not using the number of channels from lld and
assuming that there are always two channels available. With following
patch chcr will use number of channel as passed by cxgb4.

Signed-off-by: Ayush Sawal <ayush.sawal@chelsio.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: atmel-{aes,tdes} - Update the IV only when the op succeeds
Tudor Ambarus [Fri, 13 Dec 2019 09:54:56 +0000 (09:54 +0000)]
crypto: atmel-{aes,tdes} - Update the IV only when the op succeeds

Do not update the IV in case of errors.

Signed-off-by: Tudor Ambarus <tudor.ambarus@microchip.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: atmel-{sha,tdes} - Print warn message even when deferring
Tudor Ambarus [Fri, 13 Dec 2019 09:54:54 +0000 (09:54 +0000)]
crypto: atmel-{sha,tdes} - Print warn message even when deferring

Even when deferring, we would like to know what caused it.
Update dev_warn to dev_err because if the DMA init fails,
the probe is stopped.

Signed-off-by: Tudor Ambarus <tudor.ambarus@microchip.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: atmel-{aes,sha,tdes} - Stop passing unused argument in _dma_init()
Tudor Ambarus [Fri, 13 Dec 2019 09:54:49 +0000 (09:54 +0000)]
crypto: atmel-{aes,sha,tdes} - Stop passing unused argument in _dma_init()

pdata is not used.

Signed-off-by: Tudor Ambarus <tudor.ambarus@microchip.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: atmel-{aes,sha,tdes} - Drop duplicate init of dma_slave_config.direction
Tudor Ambarus [Fri, 13 Dec 2019 09:54:46 +0000 (09:54 +0000)]
crypto: atmel-{aes,sha,tdes} - Drop duplicate init of dma_slave_config.direction

The 'direction' member of the dma_slave_config will be going away
as it duplicates the direction given in the prepare call.

Signed-off-by: Tudor Ambarus <tudor.ambarus@microchip.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: atmel-{aes,sha} - Fix incorrect use of dmaengine_terminate_all()
Tudor Ambarus [Fri, 13 Dec 2019 09:54:42 +0000 (09:54 +0000)]
crypto: atmel-{aes,sha} - Fix incorrect use of dmaengine_terminate_all()

device_terminate_all() is used to abort all the pending and
ongoing transfers on the channel, it should be used just in the
error path.

Also, dmaengine_terminate_all() is deprecated and one should use
dmaengine_terminate_async() or dmaengine_terminate_sync(). The method
is not used in atomic context, use dmaengine_terminate_sync().

A secondary aspect of this patch is that it luckily avoids a deadlock
between atmel_aes and at_hdmac.c. While in tasklet with the lock held,
the dma controller invokes the client callback (dmaengine_terminate_all),
which tries to get the same lock. The at_hdmac fix would be to drop the
lock before invoking the client callback, a fix on at_hdmac will follow.

Signed-off-by: Tudor Ambarus <tudor.ambarus@microchip.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: amlogic - add unspecified HAS_IOMEM dependency
Brendan Higgins [Wed, 11 Dec 2019 19:27:40 +0000 (11:27 -0800)]
crypto: amlogic - add unspecified HAS_IOMEM dependency

Currently CONFIG_CRYPTO_DEV_AMLOGIC_GXL=y implicitly depends on
CONFIG_HAS_IOMEM=y; consequently, on architectures without IOMEM we get
the following build error:

ld: drivers/crypto/amlogic/amlogic-gxl-core.o: in function `meson_crypto_probe':
drivers/crypto/amlogic/amlogic-gxl-core.c:240: undefined reference to `devm_platform_ioremap_resource'

Fix the build error by adding the unspecified dependency.

Reported-by: Brendan Higgins <brendanhiggins@google.com>
Signed-off-by: Brendan Higgins <brendanhiggins@google.com>
Acked-by: Corentin Labbe <clabbe@baylibre.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: inside-secure - add unspecified HAS_IOMEM dependency
Brendan Higgins [Wed, 11 Dec 2019 19:27:39 +0000 (11:27 -0800)]
crypto: inside-secure - add unspecified HAS_IOMEM dependency

Currently CONFIG_CRYPTO_DEV_SAFEXCEL=y implicitly depends on
CONFIG_HAS_IOMEM=y; consequently, on architectures without IOMEM we get
the following build error:

ld: drivers/crypto/inside-secure/safexcel.o: in function `safexcel_probe':
drivers/crypto/inside-secure/safexcel.c:1692: undefined reference to `devm_platform_ioremap_resource'

Fix the build error by adding the unspecified dependency.

Reported-by: Brendan Higgins <brendanhiggins@google.com>
Signed-off-by: Brendan Higgins <brendanhiggins@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: inside-secure - Fix hang case on EIP97 with basic DES/3DES ops
Pascal van Leeuwen [Wed, 11 Dec 2019 16:32:37 +0000 (17:32 +0100)]
crypto: inside-secure - Fix hang case on EIP97 with basic DES/3DES ops

This patch fixes another hang case on the EIP97 caused by sending
invalidation tokens to the hardware when doing basic (3)DES ECB/CBC
operations. Invalidation tokens are an EIP197 feature and needed nor
supported by the EIP97. So they should not be sent for that device.

Signed-off-by: Pascal van Leeuwen <pvanleeuwen@rambus.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: inside-secure - Fix hang case on EIP97 with zero length input data
Pascal van Leeuwen [Wed, 11 Dec 2019 16:32:36 +0000 (17:32 +0100)]
crypto: inside-secure - Fix hang case on EIP97 with zero length input data

The EIP97 hardware cannot handle zero length input data and will (usually)
hang when presented with this anyway. This patch converts any zero length
input to a 1 byte dummy input to prevent this hanging.

Signed-off-by: Pascal van Leeuwen <pvanleeuwen@rambus.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: inside-secure - Fix Unable to fit even 1 command desc error w/ EIP97
Pascal van Leeuwen [Wed, 11 Dec 2019 16:32:35 +0000 (17:32 +0100)]
crypto: inside-secure - Fix Unable to fit even 1 command desc error w/ EIP97

Due to the additions of support for modes like AES-CCM and AES-GCM, which
require large command tokens, the size of the descriptor has grown such that
it now does not fit into the descriptor cache of a standard EIP97 anymore.
This means that the driver no longer works on the Marvell Armada 3700LP chip
(as used on e.g. Espressobin) that it has always supported.
Additionally, performance on EIP197's like Marvell A8K may also degrade
due to being able to fit less descriptors in the on-chip cache.
Putting these tokens into the descriptor was really a hack and not how the
design was supposed to be used - resource allocation did not account for it.

So what this patch does, is move the command token out of the descriptor.
To avoid having to allocate buffers on the fly for these command tokens,
they are stuffed in a "shadow ring", which is a circular buffer of fixed
size blocks that runs in lock-step with the descriptor ring. i.e. there is
one token block per descriptor. The descriptor ring itself is then pre-
populated with the pointers to these token blocks so these do not need to
be filled in when building the descriptors later.

Signed-off-by: Pascal van Leeuwen <pvanleeuwen@rambus.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agopadata: update documentation file path in MAINTAINERS
Daniel Jordan [Wed, 11 Dec 2019 16:21:20 +0000 (11:21 -0500)]
padata: update documentation file path in MAINTAINERS

It's changed since the recent RST conversion.

Fixes: bfcdcef8c8e3 ("padata: update documentation")
Signed-off-by: Daniel Jordan <daniel.m.jordan@oracle.com>
Cc: Herbert Xu <herbert@gondor.apana.org.au>
Cc: Steffen Klassert <steffen.klassert@secunet.com>
Cc: linux-crypto@vger.kernel.org
Cc: linux-kernel@vger.kernel.org
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: api - fix unexpectedly getting generic implementation
Herbert Xu [Wed, 11 Dec 2019 02:50:11 +0000 (10:50 +0800)]
crypto: api - fix unexpectedly getting generic implementation

When CONFIG_CRYPTO_MANAGER_EXTRA_TESTS=y, the first lookup of an
algorithm that needs to be instantiated using a template will always get
the generic implementation, even when an accelerated one is available.

This happens because the extra self-tests for the accelerated
implementation allocate the generic implementation for comparison
purposes, and then crypto_alg_tested() for the generic implementation
"fulfills" the original request (i.e. sets crypto_larval::adult).

This patch fixes this by only fulfilling the original request if
we are currently the best outstanding larval as judged by the
priority.  If we're not the best then we will ask all waiters on
that larval request to retry the lookup.

Note that this patch introduces a behaviour change when the module
providing the new algorithm is unregistered during the process.
Previously we would have failed with ENOENT, after the patch we
will instead redo the lookup.

Fixes: 9a8a6b3f0950 ("crypto: testmgr - fuzz hashes against...")
Fixes: d435e10e67be ("crypto: testmgr - fuzz skciphers against...")
Fixes: 40153b10d91c ("crypto: testmgr - fuzz AEADs against...")
Reported-by: Eric Biggers <ebiggers@google.com>
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>
5 years agocrypto: caam/qi2 - remove double buffering for ahash
Andrei Botila [Mon, 9 Dec 2019 16:59:56 +0000 (18:59 +0200)]
crypto: caam/qi2 - remove double buffering for ahash

Previously double buffering was used for storing previous and next
"less-than-block-size" bytes. Double buffering can be removed by moving
the copy of next "less-than-block-size" bytes after current request is
executed by HW.

Signed-off-by: Andrei Botila <andrei.botila@nxp.com>
Reviewed-by: Horia Geantă <horia.geanta@nxp.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: caam - remove double buffering for ahash
Andrei Botila [Mon, 9 Dec 2019 16:59:55 +0000 (18:59 +0200)]
crypto: caam - remove double buffering for ahash

Previously double buffering was used for storing previous and next
"less-than-block-size" bytes. Double buffering can be removed by moving
the copy of next "less-than-block-size" bytes after current request is
executed by HW.

Signed-off-by: Andrei Botila <andrei.botila@nxp.com>
Reviewed-by: Horia Geantă <horia.geanta@nxp.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: picoxcell - adjust the position of tasklet_init and fix missed tasklet_kill
Chuhong Yuan [Mon, 9 Dec 2019 16:21:44 +0000 (00:21 +0800)]
crypto: picoxcell - adjust the position of tasklet_init and fix missed tasklet_kill

Since tasklet is needed to be initialized before registering IRQ
handler, adjust the position of tasklet_init to fix the wrong order.

Besides, to fix the missed tasklet_kill, this patch adds a helper
function and uses devm_add_action to kill the tasklet automatically.

Fixes: ce92136843cb ("crypto: picoxcell - add support for the picoxcell crypto engines")
Signed-off-by: Chuhong Yuan <hslester96@gmail.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: hisilicon - still no need to check return value of debugfs_create functions
Greg Kroah-Hartman [Mon, 9 Dec 2019 15:21:51 +0000 (16:21 +0100)]
crypto: hisilicon - still no need to check return value of debugfs_create functions

Just like in 4a97bfc79619 ("crypto: hisilicon - no need to check return
value of debugfs_create functions"), there still is no need to ever
check the return value.  The function can work or not, but the code
logic should never do something different based on this.

Cc: Zhou Wang <wangzhou1@hisilicon.com>
Cc: Herbert Xu <herbert@gondor.apana.org.au>
Cc: "David S. Miller" <davem@davemloft.net>
Cc: linux-crypto@vger.kernel.org
Cc: linux-kernel@vger.kernel.org
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: ccp - provide in-kernel API to submit TEE commands
Rijo Thomas [Wed, 4 Dec 2019 06:19:03 +0000 (11:49 +0530)]
crypto: ccp - provide in-kernel API to submit TEE commands

Extend the functionality of AMD Secure Processor (SP) driver by
providing an in-kernel API to submit commands to TEE ring buffer for
processing by Trusted OS running on AMD Secure Processor.

Following TEE commands are supported by Trusted OS:

* TEE_CMD_ID_LOAD_TA : Load Trusted Application (TA) binary into
  TEE environment
* TEE_CMD_ID_UNLOAD_TA : Unload TA binary from TEE environment
* TEE_CMD_ID_OPEN_SESSION : Open session with loaded TA
* TEE_CMD_ID_CLOSE_SESSION : Close session with loaded TA
* TEE_CMD_ID_INVOKE_CMD : Invoke a command with loaded TA
* TEE_CMD_ID_MAP_SHARED_MEM : Map shared memory
* TEE_CMD_ID_UNMAP_SHARED_MEM : Unmap shared memory

Linux AMD-TEE driver will use this API to submit command buffers
for processing in Trusted Execution Environment. The AMD-TEE driver
shall be introduced in a separate patch.

Cc: Jens Wiklander <jens.wiklander@linaro.org>
Cc: Tom Lendacky <thomas.lendacky@amd.com>
Cc: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Co-developed-by: Devaraj Rangasamy <Devaraj.Rangasamy@amd.com>
Signed-off-by: Devaraj Rangasamy <Devaraj.Rangasamy@amd.com>
Signed-off-by: Rijo Thomas <Rijo-john.Thomas@amd.com>
Acked-by: Gary R Hook <gary.hook@amd.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: ccp - add TEE support for Raven Ridge
Rijo Thomas [Wed, 4 Dec 2019 06:19:02 +0000 (11:49 +0530)]
crypto: ccp - add TEE support for Raven Ridge

Adds a PCI device entry for Raven Ridge. Raven Ridge is an APU with a
dedicated AMD Secure Processor having Trusted Execution Environment (TEE)
support. The TEE provides a secure environment for running Trusted
Applications (TAs) which implement security-sensitive parts of a feature.

This patch configures AMD Secure Processor's TEE interface by initializing
a ring buffer (shared memory between Rich OS and Trusted OS) which can hold
multiple command buffer entries. The TEE interface is facilitated by a set
of CPU to PSP mailbox registers.

The next patch will address how commands are submitted to the ring buffer.

Cc: Jens Wiklander <jens.wiklander@linaro.org>
Cc: Tom Lendacky <thomas.lendacky@amd.com>
Cc: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Co-developed-by: Devaraj Rangasamy <Devaraj.Rangasamy@amd.com>
Signed-off-by: Devaraj Rangasamy <Devaraj.Rangasamy@amd.com>
Signed-off-by: Rijo Thomas <Rijo-john.Thomas@amd.com>
Acked-by: Gary R Hook <gary.hook@amd.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: ccp - check whether PSP supports SEV or TEE before initialization
Rijo Thomas [Wed, 4 Dec 2019 06:19:01 +0000 (11:49 +0530)]
crypto: ccp - check whether PSP supports SEV or TEE before initialization

Read PSP feature register to check for TEE (Trusted Execution Environment)
support.

If neither SEV nor TEE is supported by PSP, then skip PSP initialization.

Cc: Tom Lendacky <thomas.lendacky@amd.com>
Cc: Jens Wiklander <jens.wiklander@linaro.org>
Cc: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Co-developed-by: Devaraj Rangasamy <Devaraj.Rangasamy@amd.com>
Signed-off-by: Devaraj Rangasamy <Devaraj.Rangasamy@amd.com>
Signed-off-by: Rijo Thomas <Rijo-john.Thomas@amd.com>
Acked-by: Gary R Hook <gary.hook@amd.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: ccp - move SEV vdata to a dedicated data structure
Rijo Thomas [Wed, 4 Dec 2019 06:19:00 +0000 (11:49 +0530)]
crypto: ccp - move SEV vdata to a dedicated data structure

PSP can support both SEV and TEE interface. Therefore, move
SEV specific registers to a dedicated data structure.
TEE interface specific registers will be added in a later
patch.

Cc: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Cc: Tom Lendacky <thomas.lendacky@amd.com>
Cc: Jens Wiklander <jens.wiklander@linaro.org>
Co-developed-by: Devaraj Rangasamy <Devaraj.Rangasamy@amd.com>
Signed-off-by: Devaraj Rangasamy <Devaraj.Rangasamy@amd.com>
Signed-off-by: Rijo Thomas <Rijo-john.Thomas@amd.com>
Acked-by: Gary R Hook <gary.hook@amd.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: ccp - create a generic psp-dev file
Rijo Thomas [Wed, 4 Dec 2019 06:18:59 +0000 (11:48 +0530)]
crypto: ccp - create a generic psp-dev file

The PSP (Platform Security Processor) provides support for key management
commands in Secure Encrypted Virtualization (SEV) mode, along with
software-based Trusted Execution Environment (TEE) to enable third-party
Trusted Applications.

Therefore, introduce psp-dev.c and psp-dev.h files, which can invoke
SEV (or TEE) initialization based on platform feature support.

TEE interface support will be introduced in a later patch.

Cc: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Cc: Tom Lendacky <thomas.lendacky@amd.com>
Cc: Jens Wiklander <jens.wiklander@linaro.org>
Co-developed-by: Devaraj Rangasamy <Devaraj.Rangasamy@amd.com>
Signed-off-by: Devaraj Rangasamy <Devaraj.Rangasamy@amd.com>
Signed-off-by: Rijo Thomas <Rijo-john.Thomas@amd.com>
Acked-by: Gary R Hook <gary.hook@amd.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>