platform/kernel/linux-starfive.git
5 years agopadata: use smp_mb in padata_reorder to avoid orphaned padata jobs
Daniel Jordan [Tue, 16 Jul 2019 16:32:53 +0000 (12:32 -0400)]
padata: use smp_mb in padata_reorder to avoid orphaned padata jobs

Testing padata with the tcrypt module on a 5.2 kernel...

    # modprobe tcrypt alg="pcrypt(rfc4106(gcm(aes)))" type=3
    # modprobe tcrypt mode=211 sec=1

...produces this splat:

    INFO: task modprobe:10075 blocked for more than 120 seconds.
          Not tainted 5.2.0-base+ #16
    modprobe        D    0 10075  10064 0x80004080
    Call Trace:
     ? __schedule+0x4dd/0x610
     ? ring_buffer_unlock_commit+0x23/0x100
     schedule+0x6c/0x90
     schedule_timeout+0x3b/0x320
     ? trace_buffer_unlock_commit_regs+0x4f/0x1f0
     wait_for_common+0x160/0x1a0
     ? wake_up_q+0x80/0x80
     { crypto_wait_req }             # entries in braces added by hand
     { do_one_aead_op }
     { test_aead_jiffies }
     test_aead_speed.constprop.17+0x681/0xf30 [tcrypt]
     do_test+0x4053/0x6a2b [tcrypt]
     ? 0xffffffffa00f4000
     tcrypt_mod_init+0x50/0x1000 [tcrypt]
     ...

The second modprobe command never finishes because in padata_reorder,
CPU0's load of reorder_objects is executed before the unlocking store in
spin_unlock_bh(pd->lock), causing CPU0 to miss CPU1's increment:

CPU0                                 CPU1

padata_reorder                       padata_do_serial
  LOAD reorder_objects  // 0
                                       INC reorder_objects  // 1
                                       padata_reorder
                                         TRYLOCK pd->lock   // failed
  UNLOCK pd->lock

CPU0 deletes the timer before returning from padata_reorder and since no
other job is submitted to padata, modprobe waits indefinitely.

Add a pair of full barriers to guarantee proper ordering:

CPU0                                 CPU1

padata_reorder                       padata_do_serial
  UNLOCK pd->lock
  smp_mb()
  LOAD reorder_objects
                                       INC reorder_objects
                                       smp_mb__after_atomic()
                                       padata_reorder
                                         TRYLOCK pd->lock

smp_mb__after_atomic is needed so the read part of the trylock operation
comes after the INC, as Andrea points out.   Thanks also to Andrea for
help with writing a litmus test.

Fixes: 16295bec6398 ("padata: Generic parallelization/serialization interface")
Signed-off-by: Daniel Jordan <daniel.m.jordan@oracle.com>
Cc: <stable@vger.kernel.org>
Cc: Andrea Parri <andrea.parri@amarulasolutions.com>
Cc: Boqun Feng <boqun.feng@gmail.com>
Cc: Herbert Xu <herbert@gondor.apana.org.au>
Cc: Paul E. McKenney <paulmck@linux.ibm.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Steffen Klassert <steffen.klassert@secunet.com>
Cc: linux-arch@vger.kernel.org
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: ccp - Fix SEV_VERSION_GREATER_OR_EQUAL
David Rientjes [Fri, 12 Jul 2019 20:41:58 +0000 (13:41 -0700)]
crypto: ccp - Fix SEV_VERSION_GREATER_OR_EQUAL

SEV_VERSION_GREATER_OR_EQUAL() will fail if upgrading from 2.2 to 3.1, for
example, because the minor version is not equal to or greater than the
major.

Fix this and move to a static inline function for appropriate type
checking.

Fixes: edd303ff0e9e ("crypto: ccp - Add DOWNLOAD_FIRMWARE SEV command")
Reported-by: Cfir Cohen <cfir@google.com>
Signed-off-by: David Rientjes <rientjes@google.com>
Acked-by: Tom Lendacky <thomas.lendacky@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/gcm - use const time tag comparison.
Cfir Cohen [Tue, 2 Jul 2019 17:32:56 +0000 (10:32 -0700)]
crypto: ccp/gcm - use const time tag comparison.

Avoid leaking GCM tag through timing side channel.

Fixes: 36cf515b9bbe ("crypto: ccp - Enable support for AES GCM on v5 CCPs")
Cc: <stable@vger.kernel.org> # v4.12+
Signed-off-by: Cfir Cohen <cfir@google.com>
Acked-by: Gary R Hook <ghook@amd.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: ccp - memset structure fields to zero before reuse
Hook, Gary [Wed, 10 Jul 2019 00:09:22 +0000 (00:09 +0000)]
crypto: ccp - memset structure fields to zero before reuse

The AES GCM function reuses an 'op' data structure, which members
contain values that must be cleared for each (re)use.

This fix resolves a crypto self-test failure:
alg: aead: gcm-aes-ccp encryption test failed (wrong result) on test vector 2, cfg="two even aligned splits"

Fixes: 36cf515b9bbe ("crypto: ccp - Enable support for AES GCM on v5 CCPs")
Cc: <stable@vger.kernel.org>
Signed-off-by: Gary R Hook <gary.hook@amd.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: crypto4xx - fix a potential double free in ppc4xx_trng_probe
Wen Yang [Mon, 8 Jul 2019 06:19:03 +0000 (14:19 +0800)]
crypto: crypto4xx - fix a potential double free in ppc4xx_trng_probe

There is a possible double free issue in ppc4xx_trng_probe():

85: dev->trng_base = of_iomap(trng, 0);
86: of_node_put(trng);          ---> released here
87: if (!dev->trng_base)
88: goto err_out;
...
110: ierr_out:
111: of_node_put(trng);  ---> double released here
...

This issue was detected by using the Coccinelle software.
We fix it by removing the unnecessary of_node_put().

Fixes: 5343e674f32f ("crypto4xx: integrate ppc4xx-rng into crypto4xx")
Signed-off-by: Wen Yang <wen.yang99@zte.com.cn>
Cc: <stable@vger.kernel.org>
Cc: "David S. Miller" <davem@davemloft.net>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: Allison Randal <allison@lohutok.net>
Cc: Armijn Hemel <armijn@tjaldur.nl>
Cc: Julia Lawall <Julia.Lawall@lip6.fr>
Cc: linux-crypto@vger.kernel.org
Cc: linux-kernel@vger.kernel.org
Acked-by: Julia Lawall <julia.lawall@lip6.fr>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: stm32/hash - Fix incorrect printk modifier for size_t
Herbert Xu [Wed, 3 Jul 2019 14:26:53 +0000 (22:26 +0800)]
crypto: stm32/hash - Fix incorrect printk modifier for size_t

This patch fixes a warning when compiling stm32 because %d is being
used on a size_t argument instead of %zd.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: stm32/hash - remove interruptible condition for dma
Lionel Debieve [Fri, 28 Jun 2019 11:26:55 +0000 (13:26 +0200)]
crypto: stm32/hash - remove interruptible condition for dma

When DMA is used, waiting for completion must not be
interruptible as it can generate an error that is not handle
by the driver. There is no need to put the completion
interruptible in this driver.

Signed-off-by: Lionel Debieve <lionel.debieve@st.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: stm32/hash - Fix hmac issue more than 256 bytes
Lionel Debieve [Fri, 28 Jun 2019 11:26:54 +0000 (13:26 +0200)]
crypto: stm32/hash - Fix hmac issue more than 256 bytes

Correct condition for the second hmac loop. Key must be only
set in the first loop. Initial condition was wrong,
HMAC_KEY flag was not properly checked.

Signed-off-by: Lionel Debieve <lionel.debieve@st.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: stm32/crc32 - rename driver file
Lionel Debieve [Fri, 28 Jun 2019 11:26:41 +0000 (13:26 +0200)]
crypto: stm32/crc32 - rename driver file

Use the same naming convention for all stm32 crypto
drivers.

Signed-off-by: Lionel Debieve <lionel.debieve@st.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: amcc - remove memset after dma_alloc_coherent
Fuqian Huang [Fri, 28 Jun 2019 02:46:31 +0000 (10:46 +0800)]
crypto: amcc - remove memset after dma_alloc_coherent

In commit af7ddd8a627c
("Merge tag 'dma-mapping-4.21' of git://git.infradead.org/users/hch/dma-mapping"),
dma_alloc_coherent has already zeroed the memory.
So memset is not needed.

Signed-off-by: Fuqian Huang <huangfq.daxian@gmail.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: ccp - Switch to SPDX license identifiers
Hook, Gary [Thu, 27 Jun 2019 16:23:29 +0000 (16:23 +0000)]
crypto: ccp - Switch to SPDX license identifiers

Add an SPDX identifier and remove any specific statements.

Signed-off-by: Gary R Hook <gary.hook@amd.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: ccp - Validate the the error value used to index error messages
Hook, Gary [Thu, 27 Jun 2019 16:16:23 +0000 (16:16 +0000)]
crypto: ccp - Validate the the error value used to index error messages

The error code read from the queue status register is only 6 bits wide,
but we need to verify its value is within range before indexing the error
messages.

Fixes: 81422badb3907 ("crypto: ccp - Make syslog errors human-readable")
Cc: <stable@vger.kernel.org>
Reported-by: Cfir Cohen <cfir@google.com>
Signed-off-by: Gary R Hook <gary.hook@amd.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: doc - Fix formatting of new crypto engine content
Hook, Gary [Tue, 25 Jun 2019 23:43:50 +0000 (23:43 +0000)]
crypto: doc - Fix formatting of new crypto engine content

Tidy up the formatting/grammar in crypto_engine.rst. Use bulleted lists
where appropriate.

Signed-off-by: Gary R Hook <gary.hook@amd.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: doc - Add parameter documentation
Hook, Gary [Tue, 25 Jun 2019 23:43:43 +0000 (23:43 +0000)]
crypto: doc - Add parameter documentation

Fill in missing parameter descriptions for the compression algorithm,
then pick them up to document for the compression_alg structure.

Signed-off-by: Gary R Hook <gary.hook@amd.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: arm64/aes-ce - implement 5 way interleave for ECB, CBC and CTR
Ard Biesheuvel [Mon, 24 Jun 2019 17:38:31 +0000 (19:38 +0200)]
crypto: arm64/aes-ce - implement 5 way interleave for ECB, CBC and CTR

This implements 5-way interleaving for ECB, CBC decryption and CTR,
resulting in a speedup of ~11% on Marvell ThunderX2, which has a
very deep pipeline and therefore a high issue latency for NEON
instructions operating on the same registers.

Note that XTS is left alone: implementing 5-way interleave there
would either involve spilling of the calculated tweaks to the
stack, or recalculating them after the encryption operation, and
doing either of those would most likely penalize low end cores.

For ECB, this is not a concern at all, given that we have plenty
of spare registers. For CTR and CBC decryption, we take advantage
of the fact that v16 is not used by the CE version of the code
(which is the only one targeted by the optimization), and so we
can reshuffle the code a bit and avoid having to spill to memory
(with the exception of one extra reload in the CBC routine)

Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: arm64/aes-ce - add 5 way interleave routines
Ard Biesheuvel [Mon, 24 Jun 2019 17:38:30 +0000 (19:38 +0200)]
crypto: arm64/aes-ce - add 5 way interleave routines

In preparation of tweaking the accelerated AES chaining mode routines
to be able to use a 5-way stride, implement the core routines to
support processing 5 blocks of input at a time. While at it, drop
the 2 way versions, which have been unused for a while now.

Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: talitos - drop icv_ool
Christophe Leroy [Mon, 24 Jun 2019 07:20:17 +0000 (07:20 +0000)]
crypto: talitos - drop icv_ool

icv_ool is not used anymore, drop it.

Fixes: e345177ded17 ("crypto: talitos - fix AEAD processing.")
Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: talitos - fix hash on SEC1.
Christophe Leroy [Mon, 24 Jun 2019 07:20:16 +0000 (07:20 +0000)]
crypto: talitos - fix hash on SEC1.

On SEC1, hash provides wrong result when performing hashing in several
steps with input data SG list has more than one element. This was
detected with CONFIG_CRYPTO_MANAGER_EXTRA_TESTS:

[   44.185947] alg: hash: md5-talitos test failed (wrong result) on test vector 6, cfg="random: may_sleep use_finup src_divs=[<reimport>25.88%@+8063, <flush>24.19%@+9588, 28.63%@+16333, <reimport>4.60%@+6756, 16.70%@+16281] dst_divs=[71.61%@alignmask+16361, 14.36%@+7756, 14.3%@+"
[   44.325122] alg: hash: sha1-talitos test failed (wrong result) on test vector 3, cfg="random: inplace use_final src_divs=[<flush,nosimd>16.56%@+16378, <reimport>52.0%@+16329, 21.42%@alignmask+16380, 10.2%@alignmask+16380] iv_offset=39"
[   44.493500] alg: hash: sha224-talitos test failed (wrong result) on test vector 4, cfg="random: use_final nosimd src_divs=[<reimport>52.27%@+7401, <reimport>17.34%@+16285, <flush>17.71%@+26, 12.68%@+10644] iv_offset=43"
[   44.673262] alg: hash: sha256-talitos test failed (wrong result) on test vector 4, cfg="random: may_sleep use_finup src_divs=[<reimport>60.6%@+12790, 17.86%@+1329, <reimport>12.64%@alignmask+16300, 8.29%@+15, 0.40%@+13506, <reimport>0.51%@+16322, <reimport>0.24%@+16339] dst_divs"

This is due to two issues:
- We have an overlap between the buffer used for copying the input
data (SEC1 doesn't do scatter/gather) and the chained descriptor.
- Data copy is wrong when the previous hash left less than one
blocksize of data to hash, implying a complement of the previous
block with a few bytes from the new request.

Fix it by:
- Moving the second descriptor after the buffer, as moving the buffer
after the descriptor would make it more complex for other cipher
operations (AEAD, ABLKCIPHER)
- Skip the bytes taken from the new request to complete the previous
one by moving the SG list forward.

Fixes: 37b5e8897eb5 ("crypto: talitos - chain in buffered data for ahash on SEC1")
Cc: stable@vger.kernel.org
Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: talitos - move struct talitos_edesc into talitos.h
Christophe Leroy [Mon, 24 Jun 2019 07:20:15 +0000 (07:20 +0000)]
crypto: talitos - move struct talitos_edesc into talitos.h

Moves struct talitos_edesc into talitos.h so that it can be used
from any place in talitos.c

It will be required for next patch ("crypto: talitos - fix hash
on SEC1")

Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
Cc: stable@vger.kernel.org
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agolib/scatterlist: Fix mapping iterator when sg->offset is greater than PAGE_SIZE
Christophe Leroy [Mon, 24 Jun 2019 07:20:14 +0000 (07:20 +0000)]
lib/scatterlist: Fix mapping iterator when sg->offset is greater than PAGE_SIZE

All mapping iterator logic is based on the assumption that sg->offset
is always lower than PAGE_SIZE.

But there are situations where sg->offset is such that the SG item
is on the second page. In that case sg_copy_to_buffer() fails
properly copying the data into the buffer. One of the reason is
that the data will be outside the kmapped area used to access that
data.

This patch fixes the issue by adjusting the mapping iterator
offset and pgoffset fields such that offset is always lower than
PAGE_SIZE.

Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
Fixes: 4225fc8555a9 ("lib/scatterlist: use page iterator in the mapping iterator")
Cc: stable@vger.kernel.org
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto/NX: Set receive window credits to max number of CRBs in RxFIFO
Haren Myneni [Tue, 18 Jun 2019 19:09:22 +0000 (12:09 -0700)]
crypto/NX: Set receive window credits to max number of CRBs in RxFIFO

System gets checkstop if RxFIFO overruns with more requests than the
maximum possible number of CRBs in FIFO at the same time. The max number
of requests per window is controlled by window credits. So find max
CRBs from FIFO size and set it to receive window credits.

Fixes: b0d6c9bab5e4 ("crypto/nx: Add P9 NX support for 842 compression engine")
CC: stable@vger.kernel.org # v4.14+
Signed-off-by:Haren Myneni <haren@us.ibm.com>

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: asymmetric_keys - select CRYPTO_HASH where needed
Arnd Bergmann [Tue, 18 Jun 2019 12:13:47 +0000 (14:13 +0200)]
crypto: asymmetric_keys - select CRYPTO_HASH where needed

Build testing with some core crypto options disabled revealed
a few modules that are missing CRYPTO_HASH:

crypto/asymmetric_keys/x509_public_key.o: In function `x509_get_sig_params':
x509_public_key.c:(.text+0x4c7): undefined reference to `crypto_alloc_shash'
x509_public_key.c:(.text+0x5e5): undefined reference to `crypto_shash_digest'
crypto/asymmetric_keys/pkcs7_verify.o: In function `pkcs7_digest.isra.0':
pkcs7_verify.c:(.text+0xab): undefined reference to `crypto_alloc_shash'
pkcs7_verify.c:(.text+0x1b2): undefined reference to `crypto_shash_digest'
pkcs7_verify.c:(.text+0x3c1): undefined reference to `crypto_shash_update'
pkcs7_verify.c:(.text+0x411): undefined reference to `crypto_shash_finup'

This normally doesn't show up in randconfig tests because there is
a large number of other options that select CRYPTO_HASH.

Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: serpent - mark __serpent_setkey_sbox noinline
Arnd Bergmann [Tue, 18 Jun 2019 11:19:42 +0000 (13:19 +0200)]
crypto: serpent - mark __serpent_setkey_sbox noinline

The same bug that gcc hit in the past is apparently now showing
up with clang, which decides to inline __serpent_setkey_sbox:

crypto/serpent_generic.c:268:5: error: stack frame size of 2112 bytes in function '__serpent_setkey' [-Werror,-Wframe-larger-than=]

Marking it 'noinline' reduces the stack usage from 2112 bytes to
192 and 96 bytes, respectively, and seems to generate more
useful object code.

Fixes: c871c10e4ea7 ("crypto: serpent - improve __serpent_setkey with UBSAN")
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Reviewed-by: Eric Biggers <ebiggers@kernel.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: testmgr - dynamically allocate crypto_shash
Arnd Bergmann [Tue, 18 Jun 2019 09:21:53 +0000 (11:21 +0200)]
crypto: testmgr - dynamically allocate crypto_shash

The largest stack object in this file is now the shash descriptor.
Since there are many other stack variables, this can push it
over the 1024 byte warning limit, in particular with clang and
KASAN:

crypto/testmgr.c:1693:12: error: stack frame size of 1312 bytes in function '__alg_test_hash' [-Werror,-Wframe-larger-than=]

Make test_hash_vs_generic_impl() do the same thing as the
corresponding eaed and skcipher functions by allocating the
descriptor dynamically. We can still do better than this,
but it brings us well below the 1024 byte limit.

Suggested-by: Eric Biggers <ebiggers@kernel.org>
Fixes: 9a8a6b3f0950 ("crypto: testmgr - fuzz hashes against their generic implementation")
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Reviewed-by: Eric Biggers <ebiggers@kernel.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: testmgr - dynamically allocate testvec_config
Arnd Bergmann [Tue, 18 Jun 2019 09:21:52 +0000 (11:21 +0200)]
crypto: testmgr - dynamically allocate testvec_config

On arm32, we get warnings about high stack usage in some of the functions:

crypto/testmgr.c:2269:12: error: stack frame size of 1032 bytes in function 'alg_test_aead' [-Werror,-Wframe-larger-than=]
static int alg_test_aead(const struct alg_test_desc *desc, const char *driver,
           ^
crypto/testmgr.c:1693:12: error: stack frame size of 1312 bytes in function '__alg_test_hash' [-Werror,-Wframe-larger-than=]
static int __alg_test_hash(const struct hash_testvec *vecs,
           ^

On of the larger objects on the stack here is struct testvec_config, so
change that to dynamic allocation.

Fixes: 40153b10d91c ("crypto: testmgr - fuzz AEADs against their generic implementation")
Fixes: d435e10e67be ("crypto: testmgr - fuzz skciphers against their generic implementation")
Fixes: 9a8a6b3f0950 ("crypto: testmgr - fuzz hashes against their generic implementation")
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Reviewed-by: Eric Biggers <ebiggers@kernel.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: talitos - eliminate unneeded 'done' functions at build time
Christophe Leroy [Mon, 17 Jun 2019 21:14:45 +0000 (21:14 +0000)]
crypto: talitos - eliminate unneeded 'done' functions at build time

When building for SEC1 only, talitos2_done functions are unneeded
and should go away.

For this, use has_ftr_sec1() which will always return true when only
SEC1 support is being built, allowing GCC to drop TALITOS2 functions.

Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
Reviewed-by: Horia Geantă <horia.geanta@nxp.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: sun4i-ss - reduce stack usage
Arnd Bergmann [Mon, 17 Jun 2019 13:25:17 +0000 (15:25 +0200)]
crypto: sun4i-ss - reduce stack usage

After the latest addition, the stack usage of sun4i_ss_cipher_poll
grew beyond the warning limit when KASAN is enabled:

drivers/crypto/sunxi-ss/sun4i-ss-cipher.c:118:12: error: stack frame size of 1152 bytes in function 'sun4i_ss_cipher_poll' [-Werror,-Wframe-larger-than=]
static int sun4i_ss_cipher_poll(struct skcipher_request *areq)

Reduce it in three ways:

- split out the new code into a separate function so its stack
  usage can overlap that of the sun4i_ss_opti_poll() code path
- mark both special cases as noinline_for_stack, which should
  ideally result in a tail call that frees the rest of the
  stack
- move the buf and obuf variables into the code blocks in
  which they are used.

The three separate functions now use 144, 640 and 304 bytes of kernel
stack, respectively.

Fixes: 0ae1f46c55f8 ("crypto: sun4i-ss - fallback when length is not multiple of blocksize")
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Tested-by: Corentin LABBE <clabbe.montjoie@gmail.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: ccree - add HW engine config check
Gilad Ben-Yossef [Mon, 17 Jun 2019 08:46:30 +0000 (11:46 +0300)]
crypto: ccree - add HW engine config check

Add check to verify the stated device tree HW configuration
matches the HW.

Signed-off-by: Gilad Ben-Yossef <gilad@benyossef.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: ccree - prevent isr handling in case driver is suspended
Ofir Drang [Mon, 17 Jun 2019 08:46:29 +0000 (11:46 +0300)]
crypto: ccree - prevent isr handling in case driver is suspended

ccree irq may be shared with other devices, in order to prevent ccree isr
handling while device maybe suspended we added a check to verify that the
device is not suspended.

Signed-off-by: Ofir Drang <ofir.drang@arm.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: ccree - check that cryptocell reset completed
Ofir Drang [Mon, 17 Jun 2019 08:46:28 +0000 (11:46 +0300)]
crypto: ccree - check that cryptocell reset completed

In case of driver probe and pm resume we need to check that the cryptocell
hardware reset cycle is completed. during the reset cycle that Cryptocell
provide read only access to the APB interface which allows to verify
through the CC registers that the reset is completed. Until reset
completion we assume that any write/crypto operation is blocked.

Signed-off-by: Ofir Drang <ofir.drang@arm.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: ccree - Relocate driver irq registration after clk init
ofir.drang@arm.com [Mon, 17 Jun 2019 08:46:27 +0000 (11:46 +0300)]
crypto: ccree - Relocate driver irq registration after clk init

Signed-off-by: Ofir Drang <ofir.drang@arm.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: nx - no need to check return value of debugfs_create functions
Greg Kroah-Hartman [Fri, 14 Jun 2019 14:29:04 +0000 (16:29 +0200)]
crypto: nx - no need to check return value of debugfs_create functions

When calling debugfs functions, there 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.

Also, there is no need to store the individual debugfs file names,
especially as the whole directiry is deleted at once, so remove the
unneeded structure entirely.

Cc: "Breno Leitão" <leitao@debian.org>
Cc: Nayna Jain <nayna@linux.ibm.com>
Cc: Paulo Flabiano Smorigo <pfsmorigo@gmail.com>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Herbert Xu <herbert@gondor.apana.org.au>
Cc: "David S. Miller" <davem@davemloft.net>
Cc: linux-crypto@vger.kernel.org
Cc: linuxppc-dev@lists.ozlabs.org
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agofs: cifs: switch to RC4 library interface
Ard Biesheuvel [Wed, 12 Jun 2019 16:19:59 +0000 (18:19 +0200)]
fs: cifs: switch to RC4 library interface

The CIFS code uses the sync skcipher API to invoke the ecb(arc4) skcipher,
of which only a single generic C code implementation exists. This means
that going through all the trouble of using scatterlists etc buys us
very little, and we're better off just invoking the arc4 library directly.

This also reverts commit 5f4b55699aaf ("CIFS: Fix BUG() in calc_seckey()"),
since it is no longer necessary to allocate sec_key on the heap.

Cc: linux-cifs@vger.kernel.org
Cc: Steve French <sfrench@samba.org>
Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Acked-by: Steve French <stfrench@microsoft.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agoppp: mppe: switch to RC4 library interface
Ard Biesheuvel [Wed, 12 Jun 2019 16:19:58 +0000 (18:19 +0200)]
ppp: mppe: switch to RC4 library interface

The MPPE code uses the sync skcipher to invoke the ecb(arc4) skcipher,
of which only a single generic C code implementation exists. This means
that going through all the trouble of using scatterlists etc buys us
very little, and we're better off just invoking the arc4 library directly.

Note that the SHA1 shash used by this driver has several accelerated
implementations for various architectures, so retaining that part does
make sense.

Cc: linux-ppp@vger.kernel.org
Cc: Paul Mackerras <paulus@samba.org>
Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: arc4 - remove cipher implementation
Ard Biesheuvel [Wed, 12 Jun 2019 16:19:57 +0000 (18:19 +0200)]
crypto: arc4 - remove cipher implementation

There are no remaining users of the cipher implementation, and there
are no meaningful ways in which the arc4 cipher can be combined with
templates other than ECB (and the way we do provide that combination
is highly dubious to begin with).

So let's drop the arc4 cipher altogether, and only keep the ecb(arc4)
skcipher, which is used in various places in the kernel.

Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agonet/lib80211: move TKIP handling to ARC4 library code
Ard Biesheuvel [Wed, 12 Jun 2019 16:19:56 +0000 (18:19 +0200)]
net/lib80211: move TKIP handling to ARC4 library code

The crypto API abstraction is not very useful for invoking ciphers
directly, especially in the case of arc4, which only has a generic
implementation in C. So let's invoke the library code directly.

Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agonet/lib80211: move WEP handling to ARC4 library code
Ard Biesheuvel [Wed, 12 Jun 2019 16:19:55 +0000 (18:19 +0200)]
net/lib80211: move WEP handling to ARC4 library code

The crypto API abstraction is not very useful for invoking ciphers
directly, especially in the case of arc4, which only has a generic
implementation in C. So let's invoke the library code directly.

Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agonet/mac80211: move WEP handling to ARC4 library interface
Ard Biesheuvel [Wed, 12 Jun 2019 16:19:54 +0000 (18:19 +0200)]
net/mac80211: move WEP handling to ARC4 library interface

The WEP code in the mac80211 subsystem currently uses the crypto
API to access the arc4 (RC4) cipher, which is overly complicated,
and doesn't really have an upside in this particular case, since
ciphers are always synchronous and therefore always implemented in
software. Given that we have no accelerated software implementations
either, it is much more straightforward to invoke a generic library
interface directly.

Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: arc4 - refactor arc4 core code into separate library
Ard Biesheuvel [Wed, 12 Jun 2019 16:19:53 +0000 (18:19 +0200)]
crypto: arc4 - refactor arc4 core code into separate library

Refactor the core rc4 handling so we can move most users to a library
interface, permitting us to drop the cipher interface entirely in a
future patch. This is part of an effort to simplify the crypto API
and improve its robustness against incorrect use.

Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: talitos - fix max key size for sha384 and sha512
Christophe Leroy [Wed, 12 Jun 2019 05:49:50 +0000 (05:49 +0000)]
crypto: talitos - fix max key size for sha384 and sha512

Below commit came with a typo in the CONFIG_ symbol, leading
to a permanently reduced max key size regarless of the driver
capabilities.

Reported-by: Horia Geantă <horia.geanta@nxp.com>
Fixes: b8fbdc2bc4e7 ("crypto: talitos - reduce max key size for SEC1")
Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
Reviewed-by: Horia Geantă <horia.geanta@nxp.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: vmx - Document CTR mode counter width quirks
Daniel Axtens [Tue, 11 Jun 2019 01:54:31 +0000 (11:54 +1000)]
crypto: vmx - Document CTR mode counter width quirks

The CTR code comes from OpenSSL, where it does a 32-bit counter.
The kernel has a 128-bit counter. This difference has lead to
issues.

Document it.

Signed-off-by: Daniel Axtens <dja@axtens.net>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: caam - update IV using HW support
Horia Geantă [Mon, 10 Jun 2019 13:30:59 +0000 (16:30 +0300)]
crypto: caam - update IV using HW support

Modify drivers to perform skcipher IV update using the crypto engine,
instead of performing the operation in SW.

Besides being more efficient, this also fixes IV update for CTR mode.

Output HW S/G table is appended with an entry pointing to the same
IV buffer used as input (which is now mapped BIDIRECTIONAL).

AS (Algorithm State) parameter of the OPERATION command is changed
from INIFINAL to INIT in descriptors used by ctr(aes), cbc(aes).
This is needed since in case FINAL bit is set, HW skips IV updating
in the Context Register for the last data block.

Signed-off-by: Horia Geantă <horia.geanta@nxp.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: caam - use len instead of nents for bulding HW S/G table
Horia Geantă [Mon, 10 Jun 2019 13:30:58 +0000 (16:30 +0300)]
crypto: caam - use len instead of nents for bulding HW S/G table

Currently, conversion of SW S/G table into HW S/G layout relies on
nents returned by sg_nents_for_len(sg, len).
However this leaves the possibility of HW S/G referencing more data
then needed: since buffer length in HW S/G entries is filled using
sg_dma_len(sg), the last entry in HW S/G table might have a length
that is bigger than needed for the crypto request.

This way of S/G table conversion is fine, unless after converting a table
more entries have to be appended to the HW S/G table.
In this case, crypto engine would access data from the S/G entry having
the incorrect length, instead of advancing in the S/G table.
This situation doesn't exist, but the upcoming implementation of
IV update for skcipher algorithms needs to add a S/G entry after
req->dst S/G (corresponding to output IV).

Signed-off-by: Horia Geantă <horia.geanta@nxp.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agohwrng: iproc-rng200 - Add support for 7211
Florian Fainelli [Fri, 10 May 2019 17:31:11 +0000 (10:31 -0700)]
hwrng: iproc-rng200 - Add support for 7211

BCM7211 features a RNG200 hardware random number generator block, add
support for this chip by matching the chip-specific compatible string.

Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agodt-bindings: rng: Document BCM7211 RNG compatible string
Florian Fainelli [Fri, 10 May 2019 17:31:10 +0000 (10:31 -0700)]
dt-bindings: rng: Document BCM7211 RNG compatible string

BCM7211 features a RNG200 block, document its compatible string.

Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
Reviewed-by: Rob Herring <robh@kernel.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/herbert/crypto-2.6
Herbert Xu [Thu, 20 Jun 2019 06:17:24 +0000 (14:17 +0800)]
Merge git://git./linux/kernel/git/herbert/crypto-2.6

Merge crypto tree to pick up vmx changes.

5 years agocrypto: sahara - Use devm_platform_ioremap_resource()
Fabio Estevam [Thu, 6 Jun 2019 16:13:49 +0000 (13:13 -0300)]
crypto: sahara - Use devm_platform_ioremap_resource()

Use devm_platform_ioremap_resource() to simplify the code a bit.

Signed-off-by: Fabio Estevam <festevam@gmail.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: mxs-dcp - Use devm_platform_ioremap_resource()
Fabio Estevam [Thu, 6 Jun 2019 16:13:48 +0000 (13:13 -0300)]
crypto: mxs-dcp - Use devm_platform_ioremap_resource()

Use devm_platform_ioremap_resource() to simplify the code a bit.

Signed-off-by: Fabio Estevam <festevam@gmail.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: hisilicon - Use the correct style for SPDX License Identifier
Nishad Kamdar [Thu, 6 Jun 2019 15:06:17 +0000 (20:36 +0530)]
crypto: hisilicon - Use the correct style for SPDX License Identifier

This patch corrects the SPDX License Identifier style
in header file related to Crypto Drivers for Hisilicon
SEC Engine in Hip06 and Hip07.
For C header files Documentation/process/license-rules.rst
mandates C-like comments (opposed to C source files where
C++ style should be used)

Changes made by using a script provided by Joe Perches here:
https://lkml.org/lkml/2019/2/7/46

Suggested-by: Joe Perches <joe@perches.com>
Signed-off-by: Nishad Kamdar <nishadkamdar@gmail.com>
Acked-by: Jonathan Cameron <Jonathan.Cameron@huawei.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: qat - use struct_size() helper
Gustavo A. R. Silva [Thu, 6 Jun 2019 14:57:13 +0000 (09:57 -0500)]
crypto: qat - use struct_size() helper

One of the more common cases of allocation size calculations is finding
the size of a structure that has a zero-sized array at the end, along
with memory for some number of elements for that array. For example:

struct qat_alg_buf_list {
...
        struct qat_alg_buf bufers[];
} __packed __aligned(64);

Make use of the struct_size() helper instead of an open-coded version
in order to avoid any potential type mistakes.

So, replace the following form:

sizeof(struct qat_alg_buf_list) + ((1 + n) * sizeof(struct qat_alg_buf))

with:

struct_size(bufl, bufers, n + 1)

This code was detected with the help of Coccinelle.

Signed-off-by: Gustavo A. R. Silva <gustavo@embeddedor.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agoARM: dts: imx7ulp: add crypto support
Iuliana Prodan [Thu, 6 Jun 2019 08:02:55 +0000 (11:02 +0300)]
ARM: dts: imx7ulp: add crypto support

Add crypto node in device tree for CAAM support.

Noteworthy is that on 7ulp the interrupt line is shared
between the two job rings.

Signed-off-by: Iuliana Prodan <iuliana.prodan@nxp.com>
Signed-off-by: Franck LENORMAND <franck.lenormand@nxp.com>
Signed-off-by: Horia Geantă <horia.geanta@nxp.com>
Acked-by: Shawn Guo <shawnguo@kernel.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: cavium/nitrox - Use the correct style for SPDX License Identifier
Nishad Kamdar [Wed, 5 Jun 2019 13:21:53 +0000 (18:51 +0530)]
crypto: cavium/nitrox - Use the correct style for SPDX License Identifier

This patch corrects the SPDX License Identifier style
in header files related to Crypto Drivers for Cavium Nitrox
family CNN55XX devices.
For C header files Documentation/process/license-rules.rst
mandates C-like comments (opposed to C source files where
C++ style should be used)

Changes made by using a script provided by Joe Perches here:
https://lkml.org/lkml/2019/2/7/46

Suggested-by: Joe Perches <joe@perches.com>
Signed-off-by: Nishad Kamdar <nishadkamdar@gmail.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: bcm - Make some symbols static
YueHaibing [Tue, 4 Jun 2019 14:53:51 +0000 (22:53 +0800)]
crypto: bcm - Make some symbols static

Fix sparse warnings:

drivers/crypto/bcm/cipher.c:99:6: warning: symbol 'BCMHEADER' was not declared. Should it be static?
drivers/crypto/bcm/cipher.c:2096:6: warning: symbol 'spu_no_incr_hash' was not declared. Should it be static?
drivers/crypto/bcm/cipher.c:4823:5: warning: symbol 'bcm_spu_probe' was not declared. Should it be static?
drivers/crypto/bcm/cipher.c:4867:5: warning: symbol 'bcm_spu_remove' was not declared. Should it be static?
drivers/crypto/bcm/spu2.c:52:6: warning: symbol 'spu2_cipher_type_names' was not declared. Should it be static?
drivers/crypto/bcm/spu2.c:56:6: warning: symbol 'spu2_cipher_mode_names' was not declared. Should it be static?
drivers/crypto/bcm/spu2.c:60:6: warning: symbol 'spu2_hash_type_names' was not declared. Should it be static?
drivers/crypto/bcm/spu2.c:66:6: warning: symbol 'spu2_hash_mode_names' was not declared. Should it be static?

Reported-by: Hulk Robot <hulkci@huawei.com>
Signed-off-by: YueHaibing <yuehaibing@huawei.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: chacha - constify ctx and iv arguments
Eric Biggers [Mon, 3 Jun 2019 05:47:14 +0000 (22:47 -0700)]
crypto: chacha - constify ctx and iv arguments

Constify the ctx and iv arguments to crypto_chacha_init() and the
various chacha*_stream_xor() functions.  This makes it clear that they
are not modified.

Signed-off-by: Eric Biggers <ebiggers@google.com>
Acked-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: chacha20poly1305 - a few cleanups
Eric Biggers [Mon, 3 Jun 2019 05:46:34 +0000 (22:46 -0700)]
crypto: chacha20poly1305 - a few cleanups

- Use sg_init_one() instead of sg_init_table() then sg_set_buf().

- Remove unneeded calls to sg_init_table() prior to scatterwalk_ffwd().

- Simplify initializing the poly tail block.

- Simplify computing padlen.

This doesn't change any actual behavior.

Cc: Martin Willi <martin@strongswan.org>
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: skcipher - make chunksize and walksize accessors internal
Eric Biggers [Mon, 3 Jun 2019 05:46:11 +0000 (22:46 -0700)]
crypto: skcipher - make chunksize and walksize accessors internal

The 'chunksize' and 'walksize' properties of skcipher algorithms are
implementation details that users of the skcipher API should not be
looking at.  So move their accessor functions from <crypto/skcipher.h>
to <crypto/internal/skcipher.h>.

Signed-off-by: Eric Biggers <ebiggers@google.com>
Acked-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: skcipher - un-inline encrypt and decrypt functions
Eric Biggers [Mon, 3 Jun 2019 05:45:51 +0000 (22:45 -0700)]
crypto: skcipher - un-inline encrypt and decrypt functions

crypto_skcipher_encrypt() and crypto_skcipher_decrypt() have grown to be
more than a single indirect function call.  They now also check whether
a key has been set, and with CONFIG_CRYPTO_STATS=y they also update the
crypto statistics.  That can add up to a lot of bloat at every call
site.  Moreover, these always involve a function call anyway, which
greatly limits the benefits of inlining.

So change them to be non-inline.

Signed-off-by: Eric Biggers <ebiggers@google.com>
Acked-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: aead - un-inline encrypt and decrypt functions
Eric Biggers [Mon, 3 Jun 2019 05:45:16 +0000 (22:45 -0700)]
crypto: aead - un-inline encrypt and decrypt functions

crypto_aead_encrypt() and crypto_aead_decrypt() have grown to be more
than a single indirect function call.  They now also check whether a key
has been set, the decryption side checks whether the input is at least
as long as the authentication tag length, and with CONFIG_CRYPTO_STATS=y
they also update the crypto statistics.  That can add up to a lot of
bloat at every call site.  Moreover, these always involve a function
call anyway, which greatly limits the benefits of inlining.

So change them to be non-inline.

Signed-off-by: Eric Biggers <ebiggers@google.com>
Acked-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: x86/aesni - remove unused internal cipher algorithm
Eric Biggers [Mon, 3 Jun 2019 05:44:50 +0000 (22:44 -0700)]
crypto: x86/aesni - remove unused internal cipher algorithm

Since commit 944585a64f5e ("crypto: x86/aes-ni - remove special handling
of AES in PCBC mode"), the "__aes-aesni" internal cipher algorithm is no
longer used.  So remove it too.

Signed-off-by: Eric Biggers <ebiggers@google.com>
Acked-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: doc - improve the skcipher API example code
Eric Biggers [Mon, 3 Jun 2019 05:44:08 +0000 (22:44 -0700)]
crypto: doc - improve the skcipher API example code

Rewrite the skcipher API example, changing it to encrypt a buffer with
AES-256-XTS.  This addresses various problems with the previous example:

- It requests a specific driver "cbc-aes-aesni", which is unusual.
  Normally users ask for "cbc(aes)", not a specific driver.

- It encrypts only a single AES block.  For the reader, that doesn't
  clearly distinguish the "skcipher" API from the "cipher" API.

- Showing how to encrypt something with bare CBC is arguably a poor
  choice of example, as it doesn't follow modern crypto trends.  Now,
  usually authenticated encryption is recommended, in which case the
  user would use the AEAD API, not skcipher.  Disk encryption is still a
  legitimate use for skcipher, but for that usually XTS is recommended.

- Many other bugs and poor coding practices, such as not setting
  CRYPTO_TFM_REQ_MAY_SLEEP, unnecessarily allocating a heap buffer for
  the IV, unnecessary NULL checks, using a pointless wrapper struct, and
  forgetting to set an error code in one case.

Signed-off-by: Eric Biggers <ebiggers@google.com>
Acked-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: testmgr - add some more preemption points
Eric Biggers [Mon, 3 Jun 2019 05:42:33 +0000 (22:42 -0700)]
crypto: testmgr - add some more preemption points

Call cond_resched() after each fuzz test iteration.  This avoids stall
warnings if fuzz_iterations is set very high for testing purposes.

While we're at it, also call cond_resched() after finishing testing each
test vector.

Signed-off-by: Eric Biggers <ebiggers@google.com>
Acked-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: algapi - require cra_name and cra_driver_name
Eric Biggers [Mon, 3 Jun 2019 05:40:58 +0000 (22:40 -0700)]
crypto: algapi - require cra_name and cra_driver_name

Now that all algorithms explicitly set cra_driver_name, make it required
for algorithm registration and remove the code that generated a default
cra_driver_name.

Also add an explicit check that cra_name is set too, since that's
obviously required too, yet it didn't seem to be checked anywhere.

Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: make all generic algorithms set cra_driver_name
Eric Biggers [Mon, 3 Jun 2019 05:40:57 +0000 (22:40 -0700)]
crypto: make all generic algorithms set cra_driver_name

Most generic crypto algorithms declare a driver name ending in
"-generic".  The rest don't declare a driver name and instead rely on
the crypto API automagically appending "-generic" upon registration.

Having multiple conventions is unnecessarily confusing and makes it
harder to grep for all generic algorithms in the kernel source tree.
But also, allowing NULL driver names is problematic because sometimes
people fail to set it, e.g. the case fixed by commit 417980364300
("crypto: cavium/zip - fix collision with generic cra_driver_name").

Of course, people can also incorrectly name their drivers "-generic".
But that's much easier to notice / grep for.

Therefore, let's make cra_driver_name mandatory.  In preparation for
this, this patch makes all generic algorithms set cra_driver_name.

Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: chacha20poly1305 - fix atomic sleep when using async algorithm
Eric Biggers [Fri, 31 May 2019 18:12:30 +0000 (11:12 -0700)]
crypto: chacha20poly1305 - fix atomic sleep when using async algorithm

Clear the CRYPTO_TFM_REQ_MAY_SLEEP flag when the chacha20poly1305
operation is being continued from an async completion callback, since
sleeping may not be allowed in that context.

This is basically the same bug that was recently fixed in the xts and
lrw templates.  But, it's always been broken in chacha20poly1305 too.
This was found using syzkaller in combination with the updated crypto
self-tests which actually test the MAY_SLEEP flag now.

Reproducer:

    python -c 'import socket; socket.socket(socket.AF_ALG, 5, 0).bind(
            ("aead", "rfc7539(cryptd(chacha20-generic),poly1305-generic)"))'

Kernel output:

    BUG: sleeping function called from invalid context at include/crypto/algapi.h:426
    in_atomic(): 1, irqs_disabled(): 0, pid: 1001, name: kworker/2:2
    [...]
    CPU: 2 PID: 1001 Comm: kworker/2:2 Not tainted 5.2.0-rc2 #5
    Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.12.0-20181126_142135-anatol 04/01/2014
    Workqueue: crypto cryptd_queue_worker
    Call Trace:
     __dump_stack lib/dump_stack.c:77 [inline]
     dump_stack+0x4d/0x6a lib/dump_stack.c:113
     ___might_sleep kernel/sched/core.c:6138 [inline]
     ___might_sleep.cold.19+0x8e/0x9f kernel/sched/core.c:6095
     crypto_yield include/crypto/algapi.h:426 [inline]
     crypto_hash_walk_done+0xd6/0x100 crypto/ahash.c:113
     shash_ahash_update+0x41/0x60 crypto/shash.c:251
     shash_async_update+0xd/0x10 crypto/shash.c:260
     crypto_ahash_update include/crypto/hash.h:539 [inline]
     poly_setkey+0xf6/0x130 crypto/chacha20poly1305.c:337
     poly_init+0x51/0x60 crypto/chacha20poly1305.c:364
     async_done_continue crypto/chacha20poly1305.c:78 [inline]
     poly_genkey_done+0x15/0x30 crypto/chacha20poly1305.c:369
     cryptd_skcipher_complete+0x29/0x70 crypto/cryptd.c:279
     cryptd_skcipher_decrypt+0xcd/0x110 crypto/cryptd.c:339
     cryptd_queue_worker+0x70/0xa0 crypto/cryptd.c:184
     process_one_work+0x1ed/0x420 kernel/workqueue.c:2269
     worker_thread+0x3e/0x3a0 kernel/workqueue.c:2415
     kthread+0x11f/0x140 kernel/kthread.c:255
     ret_from_fork+0x1f/0x30 arch/x86/entry/entry_64.S:352

Fixes: 71ebc4d1b27d ("crypto: chacha20poly1305 - Add a ChaCha20-Poly1305 AEAD construction, RFC7539")
Cc: <stable@vger.kernel.org> # v4.2+
Cc: Martin Willi <martin@strongswan.org>
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: atmel-i2c - Fix build error while CRC16 set to m
YueHaibing [Fri, 31 May 2019 12:17:49 +0000 (20:17 +0800)]
crypto: atmel-i2c - Fix build error while CRC16 set to m

If CRYPTO_DEV_ATMEL_ECC is set m, which select CRC16 to m,
while CRYPTO_DEV_ATMEL_SHA204A is set to y, building fails.

drivers/crypto/atmel-i2c.o: In function 'atmel_i2c_checksum':
atmel-i2c.c:(.text+0x16): undefined reference to 'crc16'

Add CRC16 dependency to CRYPTO_DEV_ATMEL_SHA204A

Reported-by: Hulk Robot <hulkci@huawei.com>
Fixes: da001fb651b0 ("crypto: atmel-i2c - add support for SHA204A random number generator")
Signed-off-by: YueHaibing <yuehaibing@huawei.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: caam - disable some clock checks for iMX7ULP
Iuliana Prodan [Fri, 31 May 2019 11:06:34 +0000 (14:06 +0300)]
crypto: caam - disable some clock checks for iMX7ULP

Disabled the check and set of 'mem' and 'emi_slow'
clocks, since these are not available for iMX7ULP.

Signed-off-by: Iuliana Prodan <iuliana.prodan@nxp.com>
Reviewed-by: Horia Geantă <horia.geanta@nxp.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: caam - limit output IV to CBC to work around CTR mode DMA issue
Ard Biesheuvel [Fri, 31 May 2019 08:13:06 +0000 (10:13 +0200)]
crypto: caam - limit output IV to CBC to work around CTR mode DMA issue

The CAAM driver currently violates an undocumented and slightly
controversial requirement imposed by the crypto stack that a buffer
referred to by the request structure via its virtual address may not
be modified while any scatterlists passed via the same request
structure are mapped for inbound DMA.

This may result in errors like

  alg: aead: decryption failed on test 1 for gcm_base(ctr-aes-caam,ghash-generic): ret=74
  alg: aead: Failed to load transform for gcm(aes): -2

on non-cache coherent systems, due to the fact that the GCM driver
passes an IV buffer by virtual address which shares a cacheline with
the auth_tag buffer passed via a scatterlist, resulting in corruption
of the auth_tag when the IV is updated while the DMA mapping is live.

Since the IV that is returned to the caller is only valid for CBC mode,
and given that the in-kernel users of CBC (such as CTS) don't trigger the
same issue as the GCM driver, let's just disable the output IV generation
for all modes except CBC for the time being.

Fixes: 854b06f76879 ("crypto: caam - properly set IV after {en,de}crypt")
Cc: Horia Geanta <horia.geanta@nxp.com>
Cc: Iuliana Prodan <iuliana.prodan@nxp.com>
Reported-by: Sascha Hauer <s.hauer@pengutronix.de>
Cc: <stable@vger.kernel.org>
Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Reviewed-by: Horia Geanta <horia.geanta@nxp.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: atmel - Fix sparse endianness warnings
Herbert Xu [Fri, 31 May 2019 06:39:51 +0000 (14:39 +0800)]
crypto: atmel - Fix sparse endianness warnings

The param2 member in atmel_i2c_cmd is supposed to be little-endian
but was marked as u16.  This patch changes it to a __le16 which
reveals a missing endian swap in atmel_i2c_init_read_cmd.

Another missing little-endian marking is also added in
atmel_i2c_checksum.

Fixes: 11105693fa05 ("crypto: atmel-ecc - introduce Microchip...")
Reported-by: kbuild test robot <lkp@intel.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: lrw - use correct alignmask
Eric Biggers [Thu, 30 May 2019 17:53:08 +0000 (10:53 -0700)]
crypto: lrw - use correct alignmask

Commit c778f96bf347 ("crypto: lrw - Optimize tweak computation")
incorrectly reduced the alignmask of LRW instances from
'__alignof__(u64) - 1' to '__alignof__(__be32) - 1'.

However, xor_tweak() and setkey() assume that the data and key,
respectively, are aligned to 'be128', which has u64 alignment.

Fix the alignmask to be at least '__alignof__(be128) - 1'.

Fixes: c778f96bf347 ("crypto: lrw - Optimize tweak computation")
Cc: <stable@vger.kernel.org> # v4.20+
Cc: Ondrej Mosnacek <omosnace@redhat.com>
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: ghash - fix unaligned memory access in ghash_setkey()
Eric Biggers [Thu, 30 May 2019 17:50:39 +0000 (10:50 -0700)]
crypto: ghash - fix unaligned memory access in ghash_setkey()

Changing ghash_mod_init() to be subsys_initcall made it start running
before the alignment fault handler has been installed on ARM.  In kernel
builds where the keys in the ghash test vectors happened to be
misaligned in the kernel image, this exposed the longstanding bug that
ghash_setkey() is incorrectly casting the key buffer (which can have any
alignment) to be128 for passing to gf128mul_init_4k_lle().

Fix this by memcpy()ing the key to a temporary buffer.

Don't fix it by setting an alignmask on the algorithm instead because
that would unnecessarily force alignment of the data too.

Fixes: 2cdc6899a88e ("crypto: ghash - Add GHASH digest algorithm for GCM")
Reported-by: Peter Robinson <pbrobinson@gmail.com>
Cc: stable@vger.kernel.org
Signed-off-by: Eric Biggers <ebiggers@google.com>
Tested-by: Peter Robinson <pbrobinson@gmail.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: xxhash - Implement xxhash support
Nikolay Borisov [Thu, 30 May 2019 06:52:57 +0000 (09:52 +0300)]
crypto: xxhash - Implement xxhash support

xxhash is currently implemented as a self-contained module in /lib.
This patch enables that module to be used as part of the generic kernel
crypto framework. It adds a simple wrapper to the 64bit version.

I've also added test vectors (with help from Nick Terrell). The upstream
xxhash code is tested by running hashing operation on random 222 byte
data with seed values of 0 and a prime number. The upstream test
suite can be found at https://github.com/Cyan4973/xxHash/blob/cf46e0c/xxhsum.c#L664

Essentially hashing is run on data of length 0,1,14,222 with the
aforementioned seed values 0 and prime 2654435761. The particular random
222 byte string was provided to me by Nick Terrell by reading
/dev/random and the checksums were calculated by the upstream xxsum
utility with the following bash script:

dd if=/dev/random of=TEST_VECTOR bs=1 count=222

for a in 0 1; do
for l in 0 1 14 222; do
for s in 0 2654435761; do
echo algo $a length $l seed $s;
head -c $l TEST_VECTOR | ~/projects/kernel/xxHash/xxhsum -H$a -s$s
done
done
done

This produces output as follows:

algo 0 length 0 seed 0
02cc5d05  stdin
algo 0 length 0 seed 2654435761
02cc5d05  stdin
algo 0 length 1 seed 0
25201171  stdin
algo 0 length 1 seed 2654435761
25201171  stdin
algo 0 length 14 seed 0
c1d95975  stdin
algo 0 length 14 seed 2654435761
c1d95975  stdin
algo 0 length 222 seed 0
b38662a6  stdin
algo 0 length 222 seed 2654435761
b38662a6  stdin
algo 1 length 0 seed 0
ef46db3751d8e999  stdin
algo 1 length 0 seed 2654435761
ac75fda2929b17ef  stdin
algo 1 length 1 seed 0
27c3f04c2881203a  stdin
algo 1 length 1 seed 2654435761
4a15ed26415dfe4d  stdin
algo 1 length 14 seed 0
3d33dc700231dfad  stdin
algo 1 length 14 seed 2654435761
ea5f7ddef9a64f80  stdin
algo 1 length 222 seed 0
5f3d3c08ec2bef34  stdin
algo 1 length 222 seed 2654435761
6a9df59664c7ed62  stdin

algo 1 is xx64 variant, algo 0 is the 32 bit variant which is currently
not hooked up.

Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Reviewed-by: Eric Biggers <ebiggers@kernel.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: jitter - update implementation to 2.1.2
Stephan Müller [Wed, 29 May 2019 19:24:25 +0000 (21:24 +0200)]
crypto: jitter - update implementation to 2.1.2

The Jitter RNG implementation is updated to comply with upstream version
2.1.2. The change covers the following aspects:

* Time variation measurement is conducted over the LFSR operation
instead of the XOR folding

* Invcation of stuck test during initialization

* Removal of the stirring functionality and the Von-Neumann
unbiaser as the LFSR using a primitive and irreducible polynomial
generates an identical distribution of random bits

This implementation was successfully used in FIPS 140-2 validations
as well as in German BSI evaluations.

This kernel implementation was tested as follows:

* The unchanged kernel code file jitterentropy.c is compiled as part
of user space application to generate raw unconditioned noise
data. That data is processed with the NIST SP800-90B non-IID test
tool to verify that the kernel code exhibits an equal amount of noise
as the upstream Jitter RNG version 2.1.2.

* Using AF_ALG with the libkcapi tool of kcapi-rng the Jitter RNG was
output tested with dieharder to verify that the output does not
exhibit statistical weaknesses. The following command was used:
kcapi-rng -n "jitterentropy_rng" -b 100000000000 | dieharder -a -g 200

* The unchanged kernel code file jitterentropy.c is compiled as part
of user space application to test the LFSR implementation. The
LFSR is injected a monotonically increasing counter as input and
the output is fed into dieharder to verify that the LFSR operation
does not exhibit statistical weaknesses.

* The patch was tested on the Muen separation kernel which returns
a more coarse time stamp to verify that the Jitter RNG does not cause
regressions with its initialization test considering that the Jitter
RNG depends on a high-resolution timer.

Tested-by: Reto Buerki <reet@codelabs.ch>
Signed-off-by: Stephan Mueller <smueller@chronox.de>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: testmgr - test the shash API
Eric Biggers [Tue, 28 May 2019 16:40:55 +0000 (09:40 -0700)]
crypto: testmgr - test the shash API

For hash algorithms implemented using the "shash" algorithm type, test
both the ahash and shash APIs, not just the ahash API.

Testing the ahash API already tests the shash API indirectly, which is
normally good enough.  However, there have been corner cases where there
have been shash bugs that don't get exposed through the ahash API.  So,
update testmgr to test the shash API too.

This would have detected the arm64 SHA-1 and SHA-2 bugs for which fixes
were just sent out (https://patchwork.kernel.org/patch/10964843/ and
https://patchwork.kernel.org/patch/10965089/):

    alg: shash: sha1-ce test failed (wrong result) on test vector 0, cfg="init+finup aligned buffer"
    alg: shash: sha224-ce test failed (wrong result) on test vector 0, cfg="init+finup aligned buffer"
    alg: shash: sha256-ce test failed (wrong result) on test vector 0, cfg="init+finup aligned buffer"

This also would have detected the bugs fixed by commit 307508d10729
("crypto: crct10dif-generic - fix use via crypto_shash_digest()") and
commit dec3d0b1071a
("crypto: x86/crct10dif-pcl - fix use via crypto_shash_digest()").

Signed-off-by: Eric Biggers <ebiggers@google.com>
Acked-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: arm64/sha2-ce - correct digest for empty data in finup
Elena Petrova [Tue, 28 May 2019 14:35:06 +0000 (15:35 +0100)]
crypto: arm64/sha2-ce - correct digest for empty data in finup

The sha256-ce finup implementation for ARM64 produces wrong digest
for empty input (len=0). Expected: the actual digest, result: initial
value of SHA internal state. The error is in sha256_ce_finup:
for empty data `finalize` will be 1, so the code is relying on
sha2_ce_transform to make the final round. However, in
sha256_base_do_update, the block function will not be called when
len == 0.

Fix it by setting finalize to 0 if data is empty.

Fixes: 03802f6a80b3a ("crypto: arm64/sha2-ce - move SHA-224/256 ARMv8 implementation to base layer")
Cc: stable@vger.kernel.org
Signed-off-by: Elena Petrova <lenaptr@google.com>
Reviewed-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: arm64/sha1-ce - correct digest for empty data in finup
Elena Petrova [Tue, 28 May 2019 12:41:52 +0000 (13:41 +0100)]
crypto: arm64/sha1-ce - correct digest for empty data in finup

The sha1-ce finup implementation for ARM64 produces wrong digest
for empty input (len=0). Expected: da39a3ee..., result: 67452301...
(initial value of SHA internal state). The error is in sha1_ce_finup:
for empty data `finalize` will be 1, so the code is relying on
sha1_ce_transform to make the final round. However, in
sha1_base_do_update, the block function will not be called when
len == 0.

Fix it by setting finalize to 0 if data is empty.

Fixes: 07eb54d306f4 ("crypto: arm64/sha1-ce - move SHA-1 ARMv8 implementation to base layer")
Cc: stable@vger.kernel.org
Signed-off-by: Elena Petrova <lenaptr@google.com>
Reviewed-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: caam - strip input without changing crypto request
Iuliana Prodan [Tue, 28 May 2019 09:52:11 +0000 (12:52 +0300)]
crypto: caam - strip input without changing crypto request

For rsa and pkcs1pad, CAAM expects an input of modulus size.
For this we strip the leading zeros in case the size is more than modulus.
This commit avoids modifying the crypto request while stripping zeros from
input, to comply with the crypto API requirement. This is done by adding
a fixup input pointer and length.

Signed-off-by: Iuliana Prodan <iuliana.prodan@nxp.com>
Reviewed-by: Horia Geanta <horia.geanta@nxp.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: caam - fix pkcs1pad(rsa-caam, sha256) failure because of invalid input
Iuliana Prodan [Tue, 28 May 2019 09:52:10 +0000 (12:52 +0300)]
crypto: caam - fix pkcs1pad(rsa-caam, sha256) failure because of invalid input

The problem is with the input data size sent to CAAM for encrypt/decrypt.
Pkcs1pad is failing due to pkcs1 padding done in SW starting with0x01
instead of 0x00 0x01.
CAAM expects an input of modulus size. For this we strip the leading
zeros in case the size is more than modulus or pad the input with zeros
until the modulus size is reached.

Signed-off-by: Iuliana Prodan <iuliana.prodan@nxp.com>
Reviewed-by: Horia Geantă <horia.geanta@nxp.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: inside-secure - do not rely on the hardware last bit for result descriptors
Antoine Tenart [Mon, 27 May 2019 14:51:06 +0000 (16:51 +0200)]
crypto: inside-secure - do not rely on the hardware last bit for result descriptors

When performing a transformation the hardware is given result
descriptors to save the result data. Those result descriptors are
batched using a 'first' and a 'last' bit. There are cases were more
descriptors than needed are given to the engine, leading to the engine
only using some of them, and not setting the last bit on the last
descriptor we gave. This causes issues were the driver and the hardware
aren't in sync anymore about the number of result descriptors given (as
the driver do not give a pool of descriptor to use for any
transformation, but a pool of descriptors to use *per* transformation).

This patch fixes it by attaching the number of given result descriptors
to the requests, and by using this number instead of the 'last' bit
found on the descriptors to process them.

Signed-off-by: Antoine Tenart <antoine.tenart@bootlin.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: inside-secure - fix use of the SG list
Antoine Tenart [Mon, 27 May 2019 14:51:05 +0000 (16:51 +0200)]
crypto: inside-secure - fix use of the SG list

Replace sg_nents_for_len by sg_nents when DMA mapping/unmapping buffers
and when looping over the SG entries. This fix cases where the SG
entries aren't used fully, which would in such cases led to using fewer
SG entries than needed (and thus the engine wouldn't have access to the
full input data and the result would be wrong).

Signed-off-by: Antoine Tenart <antoine.tenart@bootlin.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: inside-secure - add support for HMAC updates
Antoine Tenart [Mon, 27 May 2019 14:51:04 +0000 (16:51 +0200)]
crypto: inside-secure - add support for HMAC updates

This patch adds support for HMAC updates in the Inside Secure SafeXcel
crypto engine driver. Updates were supported for hash algorithms, but
were never enabled for HMAC ones. This fixes boot time test issues.

Signed-off-by: Antoine Tenart <antoine.tenart@bootlin.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: inside-secure - implement IV retrieval
Antoine Tenart [Mon, 27 May 2019 14:51:03 +0000 (16:51 +0200)]
crypto: inside-secure - implement IV retrieval

This patch adds support for retrieving intermediate IV from the crypto
engine when using the CBC block mode with AES and (3)DES. The retrieved
IV is copied to the request IV buffer, as requested by the kernel crypto
API.

This fix boot tests added by
commit 8efd972ef96a ("crypto: testmgr - support checking skcipher output IV").

Signed-off-by: Antoine Tenart <antoine.tenart@bootlin.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: inside-secure - fix queued len computation
Antoine Tenart [Mon, 27 May 2019 14:51:02 +0000 (16:51 +0200)]
crypto: inside-secure - fix queued len computation

This patch fixes the queued len computation, which could theoretically
be wrong if req->len[1] - req->processed[1] > 1. Be future-proof here,
and fix it.

Fixes: b460edb6230a ("crypto: inside-secure - sha512 support")
Signed-off-by: Antoine Tenart <antoine.tenart@bootlin.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: inside-secure - fix zeroing of the request in ahash_exit_inv
Antoine Tenart [Mon, 27 May 2019 14:51:01 +0000 (16:51 +0200)]
crypto: inside-secure - fix zeroing of the request in ahash_exit_inv

A request is zeroed in safexcel_ahash_exit_inv(). This request total
size is EIP197_AHASH_REQ_SIZE while the memset zeroing it uses
sizeof(struct ahash_request), which happens to be less than
EIP197_AHASH_REQ_SIZE. This patch fixes it.

Fixes: f6beaea30487 ("crypto: inside-secure - authenc(hmac(sha256), cbc(aes)) support")
Signed-off-by: Antoine Tenart <antoine.tenart@bootlin.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: inside-secure - unify cache reset
Antoine Tenart [Mon, 27 May 2019 14:51:00 +0000 (16:51 +0200)]
crypto: inside-secure - unify cache reset

This patch unify the way the cache related data is zeroed when the cache
buffer is DMA unmapped. It should not change the driver behaviour, but
improves the code safety and readability.

Signed-off-by: Antoine Tenart <antoine.tenart@bootlin.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: inside-secure - enable context reuse
Antoine Tenart [Mon, 27 May 2019 14:50:59 +0000 (16:50 +0200)]
crypto: inside-secure - enable context reuse

The context given to the crypto engine can be reused over time. While
the driver was designed to allow this, the feature wasn't enabled in the
hardware engine. This patch enables it.

Signed-off-by: Antoine Tenart <antoine.tenart@bootlin.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: inside-secure - change returned error when a descriptor reports an error
Antoine Tenart [Mon, 27 May 2019 14:50:58 +0000 (16:50 +0200)]
crypto: inside-secure - change returned error when a descriptor reports an error

This patch changes the error reported by the Inside Secure SafeXcel
driver when a result descriptor reports an error, from -EIO to -EINVAL,
as this is what the crypto framework expects. This was found while
running the crypto extra tests.

Signed-off-by: Antoine Tenart <antoine.tenart@bootlin.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: inside-secure - improve the result error format when displayed
Antoine Tenart [Mon, 27 May 2019 14:50:57 +0000 (16:50 +0200)]
crypto: inside-secure - improve the result error format when displayed

The result descriptors contain errors, which are represented as a
bitmap. This patch updates the error message to not treat the error as a
decimal value, but as an hexadecimal one. This helps in knowing the
value does not have a direct meaning (the set bits themselves have).

Signed-off-by: Antoine Tenart <antoine.tenart@bootlin.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: inside-secure - remove useless check
Antoine Tenart [Mon, 27 May 2019 14:50:56 +0000 (16:50 +0200)]
crypto: inside-secure - remove useless check

When sending an ahash request, the code checks for the extra variable
not to be 0. This check is useless as the extra variable can't be 0 at
this point (it is checked on the line just before).

This patch does not modify the driver behaviour in any way.

Signed-off-by: Antoine Tenart <antoine.tenart@bootlin.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: inside-secure - fix coding style for a condition
Antoine Tenart [Mon, 27 May 2019 14:50:55 +0000 (16:50 +0200)]
crypto: inside-secure - fix coding style for a condition

This cosmetic patch fixes a cosmetic issue with if brackets.

Signed-off-by: Antoine Tenart <antoine.tenart@bootlin.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: inside-secure - move comment
Antoine Tenart [Mon, 27 May 2019 14:50:54 +0000 (16:50 +0200)]
crypto: inside-secure - move comment

This cosmetic patch moves a comment before the condition it is related
to. The patch does not change the driver behaviour in any way.

Signed-off-by: Antoine Tenart <antoine.tenart@bootlin.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: inside-secure - remove empty line
Antoine Tenart [Mon, 27 May 2019 14:50:53 +0000 (16:50 +0200)]
crypto: inside-secure - remove empty line

Cosmetic patch removing an empty line in the skcipher token creation
routine.

Signed-off-by: Antoine Tenart <antoine.tenart@bootlin.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agodt-bindings: move Atmel ECC508A I2C crypto processor to trivial-devices
Ard Biesheuvel [Fri, 24 May 2019 16:26:51 +0000 (18:26 +0200)]
dt-bindings: move Atmel ECC508A I2C crypto processor to trivial-devices

Move the binding for the discrete Atmel I2C Elliptic Curve h/w crypto
module to trivial-devices.yaml, as it doesn't belong in atmel-crypto
which describes unrelated on-SoC peripherals.

Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Reviewed-by: Rob Herring <robh@kernel.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agodt-bindings: add Atmel SHA204A I2C crypto processor
Ard Biesheuvel [Fri, 24 May 2019 16:26:50 +0000 (18:26 +0200)]
dt-bindings: add Atmel SHA204A I2C crypto processor

Add a compatible string for the Atmel SHA204A I2C crypto processor.

Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Reviewed-by: Rob Herring <robh@kernel.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: atmel-i2c - add support for SHA204A random number generator
Ard Biesheuvel [Fri, 24 May 2019 16:26:49 +0000 (18:26 +0200)]
crypto: atmel-i2c - add support for SHA204A random number generator

The Linaro/96boards Secure96 mezzanine contains (among other things)
an Atmel SHA204A symmetric crypto processor. This chip implements a
number of different functionalities, but one that is highly useful
for many different 96boards platforms is the random number generator.

So let's implement a driver for the SHA204A, and for the time being,
implement support for the random number generator only.

Reviewed-by: Linus Walleij <linus.walleij@linaro.org>
Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: atmel-ecc - factor out code that can be shared
Ard Biesheuvel [Fri, 24 May 2019 16:26:48 +0000 (18:26 +0200)]
crypto: atmel-ecc - factor out code that can be shared

In preparation of adding support for the random number generator in
Atmel atsha204a devices, refactor the existing atmel-ecc driver (which
drives hardware that is closely related) so we can share the basic
I2C and command queuing routines.

Reviewed-by: Linus Walleij <linus.walleij@linaro.org>
Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: atmel-ecc - add support for ACPI probing on non-AT91 platforms
Ard Biesheuvel [Fri, 24 May 2019 16:26:47 +0000 (18:26 +0200)]
crypto: atmel-ecc - add support for ACPI probing on non-AT91 platforms

The Atmel/Microchip EC508A is a I2C device that could be wired into
any platform, and is being used on the Linaro/96boards Secure96
mezzanine adapter. This means it could be found on any platform, even
on ones that use ACPI enumeration (via PRP0001 devices). So update the
code to enable this use case.

This involves tweaking the bus rate discovery code to take ACPI probing
into account, which records the maximum bus rate as a property of the
slave device. For the atmel-ecc code, this means that the effective bus
rate should never exceed the maximum rate, unless we are dealing with
buggy firmware. Nonetheless, let's just use the existing plumbing to
discover the bus rate and keep the existing logic intact.

Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Reviewed-by: Linus Walleij <linus.walleij@linaro.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agoi2c: acpi: permit bus speed to be discovered after enumeration
Ard Biesheuvel [Fri, 24 May 2019 16:26:46 +0000 (18:26 +0200)]
i2c: acpi: permit bus speed to be discovered after enumeration

Currently, the I2C ACPI enumeration code only permits the max bus rate
to be discovered before enumerating the slaves on the bus. In some
cases, drivers for slave devices may require this information, e.g.,
some ATmel crypto drivers need to generate a so-called wake token
of a fixed duration, regardless of the bus rate.

So tweak the code so i2c_acpi_lookup_speed() is able to obtain this
information after enumeration as well.

Acked-by: Mika Westerberg <mika.westerberg@linux.intel.com>
Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: caam - print messages in caam_dump_sg at debug level
Sascha Hauer [Thu, 23 May 2019 08:50:30 +0000 (10:50 +0200)]
crypto: caam - print messages in caam_dump_sg at debug level

caam_dump_sg() is only compiled in when DEBUG is defined, hence the
messages are debug messages. Remove the @level argument from
caam_dump_sg() and print all messages at debug level.

Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: caam - print debug messages at debug level
Sascha Hauer [Thu, 23 May 2019 08:50:29 +0000 (10:50 +0200)]
crypto: caam - print debug messages at debug level

The CAAM driver used to put its debug messages inside #ifdef DEBUG and
then prints the messages at KERN_ERR level. Replace this with proper
functions printing at KERN_DEBUG level. The #ifdef DEBUG gets
unnecessary when the right functions are used.

This replaces:

- print_hex_dump(KERN_ERR ...) inside #ifdef DEBUG with
  print_hex_dump_debug(...)
- dev_err() inside #ifdef DEBUG with dev_dbg()
- printk(KERN_ERR ...) inside #ifdef DEBUG with dev_dbg()

Some parts of the driver use these functions already, so it is only
consequent to use the debug function consistently.

Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
5 years agocrypto: caam - remove unused defines
Sascha Hauer [Thu, 23 May 2019 08:50:28 +0000 (10:50 +0200)]
crypto: caam - remove unused defines

The CAAM driver defines its own debug() macro, but it is unused. Remove
it.

Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>
Reviewed-by: Horia Geantă <horia.geanta@nxp.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>