The existing mv_cesa driver supports some features of the CESA IP but is
quite limited, and reworking it to support new features (like involving the
TDMA engine to offload the CPU) is almost impossible.
This driver has been rewritten from scratch to take those new features into
account.
This commit introduce the base infrastructure allowing us to add support
for DMA optimization.
It also includes support for one hash (SHA1) and one cipher (AES)
algorithm, and enable those features on the Armada 370 SoC.
Other algorithms and platforms will be added later on.
Signed-off-by: Boris Brezillon <boris.brezillon@free-electrons.com>
Signed-off-by: Arnaud Ebalard <arno@natisbad.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
We are about to add a new driver to support new features like using the
TDMA engine to offload the CPU.
Orion, Dove and Kirkwood platforms are already using the mv_cesa driver,
but Orion SoCs do not embed the TDMA engine, which means we will have to
differentiate them if we want to get TDMA support on Dove and Kirkwood.
In the other hand, the migration from the old driver to the new one is not
something all people are willing to do without first auditing the new
driver.
Hence we have to support the new compatible in the mv_cesa driver so that
new platforms with updated DTs can still attach their crypto engine device
to this driver.
Signed-off-by: Boris Brezillon <boris.brezillon@free-electrons.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
The mv_cesa driver currently expects the SRAM memory region to be passed
as a platform device resource.
This approach implies two drawbacks:
- the DT representation is wrong
- the only one that can access the SRAM is the crypto engine
The last point is particularly annoying in some cases: for example on
armada 370, a small region of the crypto SRAM is used to implement the
cpuidle, which means you would not be able to enable both cpuidle and the
CESA driver.
To address that problem, we explicitly define the SRAM device in the DT
and then reference the sram node from the crypto engine node.
Also note that the old way of retrieving the SRAM memory region is still
supported, or in other words, backward compatibility is preserved.
Signed-off-by: Boris Brezillon <boris.brezillon@free-electrons.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
On Dove platforms, the crypto engine requires a clock. Document this
clocks property in the mv_cesa bindings doc.
Signed-off-by: Boris Brezillon <boris.brezillon@free-electrons.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Merge the mvebu/drivers branch of the arm-soc tree which contains
just a single patch bfa1ce5f38 ("bus:
mvebu-mbus: add mv_mbus_dram_info_nooverlap()") that happens to be
a prerequisite of the new marvell/cesa crypto driver.
Update the "IBM Power in-Nest Crypto Acceleration" and
"IBM Power 842 compression accelerator" sections to specify the correct
files.
The "IBM Power in-Nest Crypto Acceleration" was originally the only
NX driver, and so its section listed all drivers/crypto/nx/ files,
but now there is also the 842 driver which has its own section. This
lists explicitly what files are owned by the Crypto driver and which
files are owned by the 842 compression driver.
Signed-off-by: Dan Streetman <ddstreet@ieee.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Add support to the nx-842-pseries.c driver for running in little endian
mode.
The pSeries platform NX 842 driver currently only works as big endian.
This adds cpu_to_be*() and be*_to_cpu() in the appropriate places to
work in LE mode also.
Signed-off-by: Dan Streetman <ddstreet@ieee.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
The new aead_edesc_alloc left out the bit indicating the last
entry on the source SG list. This patch fixes it.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
I incorrectly removed DESC_MAX_USED_BYTES when enlarging the size
of the shared descriptor buffers, thus making it four times larger
than what is necessary. This patch restores the division by four
calculation.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
The struct aead_instance is meant to extend struct crypto_instance
by incorporating the extra members of struct aead_alg. However,
the current layout which is copied from shash/ahash does not specify
the struct fully. In particular only aead_alg is present.
For shash/ahash this works because users there add extra headroom
to sizeof(struct crypto_instance) when allocating the instance.
Unfortunately for aead, this bit was lost when the new aead_instance
was added.
Rather than fixing it like shash/ahash, this patch simply expands
struct aead_instance to contain what is supposed to be there, i.e.,
adding struct crypto_instance.
In order to not break existing AEAD users, this is done through an
anonymous union.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
The struct crypto_alg is embedded into various type-specific structs
such as aead_alg. This is then used as part of instances such as
struct aead_instance. It is also embedded into the generic struct
crypto_instance. In order to ensure that struct aead_instance can
be converted to struct crypto_instance when necessary, we need to
ensure that crypto_alg is aligned properly.
This patch adds an alignment attribute to struct crypto_alg to
ensure this.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
This patch fixes a number of problems in crypto driver Kconfig
entries:
1. Select BLKCIPHER instead of BLKCIPHER2. The latter is internal
and should not be used outside of the crypto API itself.
2. Do not select ALGAPI unless you use a legacy type like
CRYPTO_ALG_TYPE_CIPHER.
3. Select the algorithm type that you are implementing, e.g., AEAD.
4. Do not select generic C code such as CBC/ECB unless you use them
as a fallback.
5. Remove default n since that is the default default.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
The patch
crypto: caam - Add definition of rd/wr_reg64 for little endian platform
added support for little endian platforms to the CAAM driver. Namely a
write and read function for 64 bit registers.
The only user of this functions is the Job Ring driver (drivers/crypto/caam/jr.c).
It uses the functions to set the DMA addresses for the input/output rings.
However, at least in the default configuration, the least significant 32 bits are
always in the base+0x0004 address; independent of the endianness of the bytes itself.
That means the addresses do not change with the system endianness.
DMA addresses are only 32 bits wide on non-64-bit systems, writing the upper 32 bits
of this value to the register for the least significant bits results in the DMA address
being set to 0.
Fix this by always writing the registers in the same way.
Suggested-by: Russell King <linux@arm.linux.org.uk>
Signed-off-by: Steffen Trumtrar <s.trumtrar@pengutronix.de>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Add a new rsa generic SW implementation.
This implements only cryptographic primitives.
Signed-off-by: Tadeusz Struk <tadeusz.struk@intel.com>
Added select on ASN1.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Add Public Key Encryption API.
Signed-off-by: Tadeusz Struk <tadeusz.struk@intel.com>
Made CRYPTO_AKCIPHER invisible like other type config options.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
The Poly1305 authenticator requires a unique key for each generated tag. This
implies that we can't set the key per tfm, as multiple users set individual
keys. Instead we pass a desc specific key as the first two blocks of the
message to authenticate in update().
Signed-off-by: Martin Willi <martin@strongswan.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
This reverts commit 9b9f9296a7b73fbafe0a0a6f2494eaadd97f9f73 as
all in-kernel implementations of GCM have been converted to the
new AEAD interface, meaning that they should now pass the updated
rfc4543 test.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
This patch converts the caam GCM implementations to the new AEAD
interface. This is compile-tested only.
Note that all IV generation for GCM algorithms have been removed.
The reason is that the current generation uses purely random IVs
which is not appropriate for counter-based algorithms where we
first and foremost require uniqueness.
Of course there is no reason why you couldn't implement seqiv or
seqniv within caam since all they do is xor the sequence number
with a salt, but since I can't test this on actual hardware I'll
leave it alone for now.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Currently dma_map_sg_chained does not handle errors from the
underlying dma_map_sg calls. This patch adds rollback in case
of an error by simply calling dma_unmap_sg_chained for the ones
that we've already mapped.
All current callers ignore the return value so this should have
no impact on them.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
This patch converts the nx GCM implementations to the new AEAD
interface. This is compile-tested only.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
This patch updates the rfc4543 test vectors to the new format
where the IV is part of the AD. For now these vectors are still
unused. They will be reactivated once all rfc4543 implementations
have migrated.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
This patch converts generic gcm and its associated transforms to
the new AEAD interface. The biggest reward is in code reduction
for rfc4543 where it used to do IV stitching which is no longer
needed as the IV is already part of the AD on input.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Because the old rfc4543 implementation always injected an IV into
the AD, while the new one does not, we have to disable the test
while it is converted over to the new AEAD interface.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Fix a "Trying to vfree() nonexistent vm area" error when unloading the CAAM
controller module by providing the correct pointer value to iounmap().
Signed-off-by: Victoria Milhoan <vicki.milhoan@freescale.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
The CAAM driver uses two data buffers to store data for a hashing operation,
with one buffer defined as active. This change forces switching of the
active buffer when executing a hashing operation to avoid a later DMA unmap
using the length of the opposite buffer.
Signed-off-by: Victoria Milhoan <vicki.milhoan@freescale.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Added a mpi_read_buf() helper function to export MPI to a buf provided by
the user, and a mpi_get_size() helper, that tells the user how big the buf is.
Changed mpi_free to use kzfree instead of kfree because it is used to free
crypto keys.
Signed-off-by: Tadeusz Struk <tadeusz.struk@intel.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
The top-level CRYPTO_DEV_VMX option already depends on PPC64 so
there is no need to depend on it again at CRYPTO_DEV_VMX_ENCRYPT.
This patch also removes a redundant "default n".
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
The '__init aesni_init()' function calls the '__exit crypto_fpu_exit()'
function directly. Since they are in different sections, this generates
a warning.
make CONFIG_DEBUG_SECTION_MISMATCH=y
...
WARNING: arch/x86/crypto/aesni-intel.o(.init.text+0x12b): Section
mismatch in reference from the function init_module() to the function
.exit.text:crypto_fpu_exit()
The function __init init_module() references
a function __exit crypto_fpu_exit().
This is often seen when error handling in the init function
uses functionality in the exit path.
The fix is often to remove the __exit annotation of
crypto_fpu_exit() so it may be used outside an exit section.
Fix the warning by removing the __exit annotation.
Signed-off-by: Jeremiah Mahler <jmmahler@gmail.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Replace the NX842_MEM_COMPRESS define with a function that returns the
specific platform driver's required working memory size.
The common nx-842.c driver refuses to load if there is no platform
driver present, so instead of defining an approximate working memory
size that's the maximum approximate size of both platform driver's
size requirements, the platform driver can directly provide its
specific, i.e. sizeof(struct nx842_workmem), size requirements which
the 842-nx crypto compression driver will use.
This saves memory by both reducing the required size of each driver
to the specific sizeof() amount, as well as using the specific loaded
platform driver's required amount, instead of the maximum of both.
Signed-off-by: Dan Streetman <ddstreet@ieee.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Move the contents of the include/linux/nx842.h header file into the
drivers/crypto/nx/nx-842.h header file. Remove the nx842.h header
file and its entry in the MAINTAINERS file.
The include/linux/nx842.h header originally was there because the
crypto/842.c driver needed it to communicate with the nx-842 hw
driver. However, that crypto compression driver was moved into
the drivers/crypto/nx/ directory, and now can directly include the
nx-842.h header. Nothing else needs the public include/linux/nx842.h
header file, as all use of the nx-842 hardware driver will be through
the "842-nx" crypto compression driver, since the direct nx-842 api is
very limited in the buffer alignments and sizes that it will accept,
and the crypto compression interface handles those limitations and
allows any alignment and size buffers.
Signed-off-by: Dan Streetman <ddstreet@ieee.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Currently the driver assumes that the SG list contains exactly
the number of bytes required. This assumption is incorrect.
Up until now this has been harmless. However with the new AEAD
interface this now breaks as the AD SG list contains more bytes
than just the AD.
This patch fixes this by always clamping the AD SG list by the
specified AD length.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
This patch makes use of the new sg_nents_for_len helper to replace
the custom sg_count function.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
This driver uses SZ_64K so it should include linux/sizes.h rather
than relying on others to pull it in for it.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Be more verbose and also report ->backend_cra_name when
crypto_alloc_shash() or crypto_alloc_cipher() fail in
drbg_init_hash_kernel() or drbg_init_sym_kernel()
correspondingly.
Example
DRBG: could not allocate digest TFM handle: hmac(sha256)
Signed-off-by: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
As required by SP800-90A, the DRBG implements are reseeding threshold.
This threshold is at 2**48 (64 bit) and 2**32 bit (32 bit) as
implemented in drbg_max_requests.
With the recently introduced changes, the DRBG is now always used as a
stdrng which is initialized very early in the boot cycle. To ensure that
sufficient entropy is present, the Jitter RNG is added to even provide
entropy at early boot time.
However, the 2nd seed source, the nonblocking pool, is usually
degraded at that time. Therefore, the DRBG is seeded with the Jitter RNG
(which I believe contains good entropy, which however is questioned by
others) and is seeded with a degradded nonblocking pool. This seed is
now used for quasi the lifetime of the system (2**48 requests is a lot).
The patch now changes the reseed threshold as follows: up until the time
the DRBG obtains a seed from a fully iniitialized nonblocking pool, the
reseeding threshold is lowered such that the DRBG is forced to reseed
itself resonably often. Once it obtains the seed from a fully
initialized nonblocking pool, the reseed threshold is set to the value
required by SP800-90A.
Signed-off-by: Stephan Mueller <smueller@chronox.de>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
This patch removes the kernel blocking API as it has been completely
replaced by the callback API.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
The get_blocking_random_bytes API is broken because the wait can
be arbitrarily long (potentially forever) so there is no safe way
of calling it from within the kernel.
This patch replaces it with the new callback API which does not
have this problem.
The patch also removes the entropy buffer registered with the DRBG
handle in favor of stack variables to hold the seed data.
Signed-off-by: Stephan Mueller <smueller@chronox.de>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
The get_blocking_random_bytes API is broken because the wait can
be arbitrarily long (potentially forever) so there is no safe way
of calling it from within the kernel.
This patch replaces it with a callback API instead. The callback
is invoked potentially from interrupt context so the user needs
to schedule their own work thread if necessary.
In addition to adding callbacks, they can also be removed as
otherwise this opens up a way for user-space to allocate kernel
memory with no bound (by opening algif_rng descriptors and then
closing them).
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
nios2 is the only architecture that does not inline get_cycles
and does not export it. This breaks crypto as it uses get_cycles
in a number of modules.
Reported-by: Guenter Roeck <linux@roeck-us.net>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Replace the global -O0 compiler flag from the Makefile with GCC
pragmas to mark only the functions required to be compiled without
optimizations.
This patch also adds a comment describing the rationale for the
functions chosen to be compiled without optimizations.
Signed-off-by: Stephan Mueller <smueller@chronox.de>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Currently caam assumes that the SG list contains exactly the number
of bytes required. This assumption is incorrect.
Up until now this has been harmless. However with the new AEAD
interface this now breaks as the AD SG list contains more bytes
than just the AD.
This patch fixes this by always clamping the AD SG list by the
specified AD length.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
This patch fixes an issue when building an internal AD representation.
We need to check assoclen and not only blindly loop over assoc sgl.
Signed-off-by: Tadeusz Struk <tadeusz.struk@intel.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
The device doensn't support the default value and will change it to 256, which
will cause performace degradation for biger packets.
Add an explicit write to set it to 1024.
Reported-by: Tianliang Wang <tianliang.wang@intel.com>
Signed-off-by: Tadeusz Struk <tadeusz.struk@intel.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Boris BREZILLON