Tweak the round key logic so that they can be loaded using a single
branchless sequence using overlapping loads. This is shorter and
simpler, and puts the conditional branches based on the key size further
apart, which might benefit microarchitectures that cannot record taken
branches at every instruction. For these branches, use test-bit-branch
instructions that don't clobber the condition flags.
Note that none of this has any impact on performance, positive or
otherwise (and the branch prediction benefit would only benefit AES-192
which nobody uses). It does make for nicer code, though.
While at it, use \@ to generate the labels inside the macros, which is
more robust than using fixed numbers, which could clash inadvertently.
Also, bring aes-neon.S in line with these changes, including the switch
to test-and-branch instructions, to avoid surprises in the future when
we might start relying on the condition flags being preserved in the
chaining mode wrappers in aes-modes.S
Signed-off-by: Ard Biesheuvel <ardb@kernel.org>
Reviewed-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
API:
- Avoid unnecessary copying in scomp for trivial SG lists.
Algorithms:
- Optimise NEON CCM implementation on ARM64.
Drivers:
- Add queue stop/query debugfs support in hisilicon/qm.
-
-----BEGIN PGP SIGNATURE-----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=Uw2R
-----END PGP SIGNATURE-----
Merge tag 'v6.9-p1' of git://git.kernel.org/pub/scm/linux/kernel/git/herbert/crypto-2.6
Pull crypto updates from Herbert Xu:
"API:
- Avoid unnecessary copying in scomp for trivial SG lists
Algorithms:
- Optimise NEON CCM implementation on ARM64
Drivers:
- Add queue stop/query debugfs support in hisilicon/qm
- Intel qat updates and cleanups"
* tag 'v6.9-p1' of git://git.kernel.org/pub/scm/linux/kernel/git/herbert/crypto-2.6: (79 commits)
Revert "crypto: remove CONFIG_CRYPTO_STATS"
crypto: scomp - remove memcpy if sg_nents is 1 and pages are lowmem
crypto: tcrypt - add ffdhe2048(dh) test
crypto: iaa - fix the missing CRYPTO_ALG_ASYNC in cra_flags
crypto: hisilicon/zip - fix the missing CRYPTO_ALG_ASYNC in cra_flags
hwrng: hisi - use dev_err_probe
MAINTAINERS: Remove T Ambarus from few mchp entries
crypto: iaa - Fix comp/decomp delay statistics
crypto: iaa - Fix async_disable descriptor leak
dt-bindings: rng: atmel,at91-trng: add sam9x7 TRNG
dt-bindings: crypto: add sam9x7 in Atmel TDES
dt-bindings: crypto: add sam9x7 in Atmel SHA
dt-bindings: crypto: add sam9x7 in Atmel AES
crypto: remove CONFIG_CRYPTO_STATS
crypto: dh - Make public key test FIPS-only
crypto: rockchip - fix to check return value
crypto: jitter - fix CRYPTO_JITTERENTROPY help text
crypto: qat - make ring to service map common for QAT GEN4
crypto: qat - fix ring to service map for dcc in 420xx
crypto: qat - fix ring to service map for dcc in 4xxx
...
The bit-sliced implementation of AES-CTR operates on blocks of 128
bytes, and will fall back to the plain NEON version for tail blocks or
inputs that are shorter than 128 bytes to begin with.
It will call straight into the plain NEON asm helper, which performs all
memory accesses in granules of 16 bytes (the size of a NEON register).
For this reason, the associated plain NEON glue code will copy inputs
shorter than 16 bytes into a temporary buffer, given that this is a rare
occurrence and it is not worth the effort to work around this in the asm
code.
The fallback from the bit-sliced NEON version fails to take this into
account, potentially resulting in out-of-bounds accesses. So clone the
same workaround, and use a temp buffer for short in/outputs.
Fixes: fc074e1300 ("crypto: arm64/aes-neonbs-ctr - fallback to plain NEON for final chunk")
Cc: <stable@vger.kernel.org>
Reported-by: syzbot+f1ceaa1a09ab891e1934@syzkaller.appspotmail.com
Reviewed-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Ard Biesheuvel <ardb@kernel.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
The C glue code already infers whether or not the current iteration is
the final one, by comparing walk.nbytes with walk.total. This means we
can easily inform the asm helpers of this as well, by conditionally
passing a pointer to the original IV, which is used in the finalization
of the MAC. This removes the need for a separate call into the asm code
to perform the finalization.
Signed-off-by: Ard Biesheuvel <ardb@kernel.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
The encryption and decryption code paths are mostly identical, except
for a small difference where the plaintext input into the MAC is taken
from either the input or the output block.
We can factor this in quite easily using a vector bit select, and a few
additional XORs, without the need for branches. This way, we can use the
same tail handling logic on the encrypt and decrypt code paths, allowing
further consolidation of the asm helpers in a subsequent patch.
(In the main loop, adding just a handful of ALU instructions results in
a noticeable performance hit [around 5% on Apple M2], so those routines
are kept separate)
Signed-off-by: Ard Biesheuvel <ardb@kernel.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
The CCM code as originally written attempted to use as few NEON
registers as possible, to avoid having to eagerly preserve/restore the
entire NEON register file at every call to kernel_neon_begin/end. At
that time, this API took a number of NEON registers as a parameter, and
only preserved that many registers.
Today, the NEON register file is restored lazily, and the old API is
long gone. This means we can use as many NEON registers as we can make
meaningful use of, which means in the AES case that we can keep all
round keys in registers rather than reloading each of them for each AES
block processed.
On Cortex-A53, this results in a speedup of more than 50%. (From 4
cycles per byte to 2.6 cycles per byte)
Signed-off-by: Ard Biesheuvel <ardb@kernel.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
CCM combines the counter (CTR) encryption mode with a MAC based on the
same block cipher. This MAC construction is a bit clunky: it invokes the
block cipher in a way that cannot be parallelized, resulting in poor CPU
pipeline efficiency.
The arm64 CCM code mitigates this by interleaving the encryption and MAC
at the AES round level, resulting in a substantial speedup. But this
approach does not apply to the additional authenticated data (AAD) which
is not encrypted.
This means the special asm routine dealing with the AAD is not any
better than the MAC update routine used by the arm64 AES block
encryption driver, so let's reuse that, and drop the special AES-CCM
version.
Signed-off-by: Ard Biesheuvel <ardb@kernel.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Implement the CCM tail handling using a single sequence that uses
permute vectors and overlapping loads and stores, rather than going over
the tail byte by byte in a loop, and using scalar operations. This is
more efficient, even though the measured speedup is only around 1-2% on
the CPUs I have tried.
Signed-off-by: Ard Biesheuvel <ardb@kernel.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
In preparation for optimizing the CCM core asm code using permutation
vectors and overlapping loads and stores, ensure that inputs shorter
than the size of a AES block are passed via a buffer on the stack, in a
way that positions the data at the end of a 16 byte buffer. This removes
the need for the asm code to reason about a rare corner case where the
tail of the data cannot be read/written using a single NEON load/store
instruction.
While at it, tweak the copyright header and authorship to bring it up to
date.
Signed-off-by: Ard Biesheuvel <ardb@kernel.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Now that kernel mode NEON no longer disables preemption, we no longer
have to take care to disable and re-enable use of the NEON when calling
into the skcipher walk API. So just keep it enabled until done.
Signed-off-by: Ard Biesheuvel <ardb@kernel.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
This reverts commit 57ead1bf1c, which updated the CCM code to only
rely on walk.nbytes to check for failures returned from the skcipher
walk API, mostly for the common good rather than to fix a particular
problem in the code.
This change introduces a problem of its own: the skcipher walk is
started with the 'atomic' argument set to false, which means that the
skcipher walk API is permitted to sleep. Subsequently, it invokes
skcipher_walk_done() with preemption disabled on the final iteration of
the loop. This appears to work by accident, but it is arguably a bad
example, and providing a better example was the point of the original
patch.
Given that future changes to the CCM code will rely on the original
behavior of entering the loop even for zero sized inputs, let's just
revert this change entirely, and proceed from there.
Signed-off-by: Ard Biesheuvel <ardb@kernel.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
In the Linux kernel, a function whose name has two leading underscores
is conventionally called by the same-named function without leading
underscores -- not the other way around. __sha512_block_data_order()
got this backwards. Fix this, albeit without changing the name in the
perlasm since that is OpenSSL code. No change in behavior.
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
In the Linux kernel, a function whose name has two leading underscores
is conventionally called by the same-named function without leading
underscores -- not the other way around. __sha256_block_data_order()
and __sha256_block_neon() got this backwards. Fix this, albeit without
changing the names in the perlasm since that is OpenSSL code. No change
in behavior.
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
In the Linux kernel, a function whose name has two leading underscores
is conventionally called by the same-named function without leading
underscores -- not the other way around. __sha512_ce_transform() and
__sha512_block_data_order() got this backwards. Fix this, albeit
without changing "sha512_block_data_order" in the perlasm since that is
OpenSSL code. No change in behavior.
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
In the Linux kernel, a function whose name has two leading underscores
is conventionally called by the same-named function without leading
underscores -- not the other way around. __sha2_ce_transform() and
__sha256_block_data_order() got this backwards. Fix this, albeit
without changing "sha256_block_data_order" in the perlasm since that is
OpenSSL code. No change in behavior.
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
In the Linux kernel, a function whose name has two leading underscores
is conventionally called by the same-named function without leading
underscores -- not the other way around. __sha1_ce_transform() got this
backwards. Fix this. No change in behavior.
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Implement the ->digest method to improve performance on single-page
messages by reducing the number of indirect calls.
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Implement a ->digest function for sha256-ce. This improves the
performance of crypto_shash_digest() with this algorithm by reducing the
number of indirect calls that are made. This only adds ~112 bytes of
code, mostly for the inlined init, as the finup function is tail-called.
For now, don't bother with this for sha224, since sha224 is rarely used.
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Do it more simiply. This also fixes single target builds.
[before]
$ make ARCH=arm64 CROSS_COMPILE=aarch64-linux-gnu- arch/arm64/crypto/aes-glue-ce.i
[snip]
make[4]: *** No rule to make target 'arch/arm64/crypto/aes-glue-ce.i'. Stop.
[after]
$ make ARCH=arm64 CROSS_COMPILE=aarch64-linux-gnu- arch/arm64/crypto/aes-glue-ce.i
[snip]
CPP arch/arm64/crypto/aes-glue-ce.i
Signed-off-by: Masahiro Yamada <masahiroy@kernel.org>
Acked-by: Will Deacon <will@kernel.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Include module.h in arch/arm64/crypto/sha256-glue.c as it uses
various macros (such as MODULE_AUTHOR) that are defined there.
Also fix the ordering of types.h.
Reported-by: kernel test robot <lkp@intel.com>
Closes: https://lore.kernel.org/oe-kbuild-all/202305191953.PIB1w80W-lkp@intel.com/
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
aesbs_ecb_encrypt(), aesbs_ecb_decrypt(), aesbs_xts_encrypt(), and
aesbs_xts_decrypt() are called via indirect function calls. Therefore
they need to use SYM_TYPED_FUNC_START instead of SYM_FUNC_START to cause
their type hashes to be emitted when the kernel is built with
CONFIG_CFI_CLANG=y. Otherwise, the code crashes with a CFI failure if
the compiler doesn't happen to optimize out the indirect calls.
Fixes: c50d32859e ("arm64: Add types to indirect called assembly functions")
Cc: stable@vger.kernel.org
Signed-off-by: Eric Biggers <ebiggers@google.com>
Reviewed-by: Ard Biesheuvel <ardb@kernel.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
An often overlooked aspect of the skcipher walker API is that an
error is not just indicated by a non-zero return value, but by the
fact that walk->nbytes is zero.
Thus it is an error to call skcipher_walk_done after getting back
walk->nbytes == 0 from the previous interaction with the walker.
This is because when walk->nbytes is zero the walker is left in
an undefined state and any further calls to it may try to free
uninitialised stack memory.
The sm4 arm64 ccm code gets this wrong and ends up calling
skcipher_walk_done even when walk->nbytes is zero.
This patch rewrites the loop in a form that resembles other callers.
Reported-by: Tianjia Zhang <tianjia.zhang@linux.alibaba.com>
Fixes: ae1b83c7d5 ("crypto: arm64/sm4 - add CE implementation for GCM mode")
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Tested-by: Tianjia Zhang <tianjia.zhang@linux.alibaba.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
The fact that an error in the skcipher walker API are indicated
not only by a non-zero return value, but also by the fact that
walk->nbytes is zero, causes the layout of the skcipher walker
loop to be sufficiently different from the usual layout, which
is not a problem in itself, but it is likely to cause reading
confusion and difficulty in code maintenance.
This patch rewrites skcipher walker loop, and separates the
last chunk cryption from the loop to avoid wrong calls to the
skcipher walker API. In addition to following the usual convention
of checking walk->nbytes, it also makes the loop execute logic
clearer and easier to understand.
Signed-off-by: Tianjia Zhang <tianjia.zhang@linux.alibaba.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
An often overlooked aspect of the skcipher walker API is that an
error is not just indicated by a non-zero return value, but by the
fact that walk->nbytes is zero.
Thus it is an error to call skcipher_walk_done after getting back
walk->nbytes == 0 from the previous interaction with the walker.
This is because when walk->nbytes is zero the walker is left in
an undefined state and any further calls to it may try to free
uninitialised stack memory.
The arm64 ccm code has to deal with zero-length messages, and
it needs to process data even when walk->nbytes == 0 is returned.
It doesn't have this bug because there is an explicit check for
walk->nbytes != 0 prior to the skcipher_walk_done call.
However, the loop is still sufficiently different from the usual
layout and it appears to have been copied into other code which
then ended up with this bug. This patch rewrites it to follow the
usual convention of checking walk->nbytes.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Tested-by: Ard Biesheuvel <ardb@kernel.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Add support for RFC4106 ESP encapsulation to the accelerated GCM
implementation. This results in a ~10% speedup for IPsec frames of
typical size (~1420 bytes) on Cortex-A53.
Signed-off-by: Ard Biesheuvel <ardb@kernel.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
The SM4 CCM/GCM assembly functions for encryption and decryption is
called via indirect function calls. Therefore they need to use
SYM_TYPED_FUNC_START instead of SYM_FUNC_START to cause its type hash
to be emitted when the kernel is built with CONFIG_CFI_CLANG=y.
Otherwise, the code crashes with a CFI failure (if the compiler didn't
happen to optimize out the indirect call).
Fixes: 67fa3a7fdf ("crypto: arm64/sm4 - add CE implementation for CCM mode")
Fixes: ae1b83c7d5 ("crypto: arm64/sm4 - add CE implementation for GCM mode")
Signed-off-by: Tianjia Zhang <tianjia.zhang@linux.alibaba.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Use the frame_push and frame_pop macros to set up the stack frame so
that return address protections will be enabled automically when
configured.
Signed-off-by: Ard Biesheuvel <ardb@kernel.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Use the frame_push and frame_pop macros to set up the stack frame so
that return address protections will be enabled automically when
configured.
Signed-off-by: Ard Biesheuvel <ardb@kernel.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Use the frame_push and frame_pop macros to create the stack frames in
the AES chaining mode wrappers so that they will get PAC and/or shadow
call stack protection when configured.
Signed-off-by: Ard Biesheuvel <ardb@kernel.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Use the frame_push and frame_pop macros consistently to create the stack
frame, so that we will get PAC and/or shadow call stack handling as well
when enabled.
Signed-off-by: Ard Biesheuvel <ardb@kernel.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
The helper crypto_tfm_ctx is only used by the Crypto API algorithm
code and should really be in algapi.h. However, for historical
reasons many files relied on it to be in crypto.h. This patch
changes those files to use algapi.h instead in prepartion for a
move.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
sm3_neon_transform() is called via indirect function calls. Therefore
it needs to use SYM_TYPED_FUNC_START instead of SYM_FUNC_START to cause
its type hash to be emitted when the kernel is built with
CONFIG_CFI_CLANG=y. Otherwise, the code crashes with a CFI failure (if
the compiler didn't happen to optimize out the indirect call).
Fixes: c50d32859e ("arm64: Add types to indirect called assembly functions")
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Sami Tolvanen <samitolvanen@google.com>
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Since the CFI implementation now supports indirect calls to assembly
functions, take advantage of that rather than use a wrapper function.
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Sami Tolvanen <samitolvanen@google.com>
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
The cpu feature defined by MODULE_DEVICE_TABLE is only referenced when
compiling as a module, and the warning of unused variable will be
encountered when compiling with intree. The warning can be removed by
adding the __maybe_unused flag.
Fixes: 03c9a333fe ("crypto: arm64/ghash - add NEON accelerated fallback for 64-bit PMULL")
Fixes: ae1b83c7d5 ("crypto: arm64/sm4 - add CE implementation for GCM mode")
Reported-by: kernel test robot <lkp@intel.com>
Signed-off-by: Tianjia Zhang <tianjia.zhang@linux.alibaba.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
The gf128mul library does not depend on the crypto API at all, so it can
be moved into lib/crypto. This will allow us to use it in other library
code in a subsequent patch without having to depend on CONFIG_CRYPTO.
While at it, change the Kconfig symbol name to align with other crypto
library implementations. However, the source file name is retained, as
it is reflected in the module .ko filename, and changing this might
break things for users.
Signed-off-by: Ard Biesheuvel <ardb@kernel.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
This patch is a CE-optimized assembly implementation for GCM mode.
Benchmark on T-Head Yitian-710 2.75 GHz, the data comes from the 224 and 224
modes of tcrypt, and compared the performance before and after this patch (the
driver used before this patch is gcm_base(ctr-sm4-ce,ghash-generic)).
The abscissas are blocks of different lengths. The data is tabulated and the
unit is Mb/s:
Before (gcm_base(ctr-sm4-ce,ghash-generic)):
gcm(sm4) | 16 64 256 512 1024 1420 4096 8192
-------------+---------------------------------------------------------------------
GCM enc | 25.24 64.65 104.66 116.69 123.81 125.12 129.67 130.62
GCM dec | 25.40 64.80 104.74 116.70 123.81 125.21 129.68 130.59
GCM mb enc | 24.95 64.06 104.20 116.38 123.55 124.97 129.63 130.61
GCM mb dec | 24.92 64.00 104.13 116.34 123.55 124.98 129.56 130.48
After:
gcm-sm4-ce | 16 64 256 512 1024 1420 4096 8192
-------------+---------------------------------------------------------------------
GCM enc | 108.62 397.18 971.60 1283.92 1522.77 1513.39 1777.00 1806.96
GCM dec | 116.36 398.14 1004.27 1319.11 1624.21 1635.43 1932.54 1974.20
GCM mb enc | 107.13 391.79 962.05 1274.94 1514.76 1508.57 1769.07 1801.58
GCM mb dec | 113.40 389.36 988.51 1307.68 1619.10 1631.55 1931.70 1970.86
Signed-off-by: Tianjia Zhang <tianjia.zhang@linux.alibaba.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
This patch is a CE-optimized assembly implementation for CCM mode.
Benchmark on T-Head Yitian-710 2.75 GHz, the data comes from the 223 and 225
modes of tcrypt, and compared the performance before and after this patch (the
driver used before this patch is ccm_base(ctr-sm4-ce,cbcmac-sm4-ce)).
The abscissas are blocks of different lengths. The data is tabulated and the
unit is Mb/s:
Before (rfc4309(ccm_base(ctr-sm4-ce,cbcmac-sm4-ce))):
ccm(sm4) | 16 64 256 512 1024 1420 4096 8192
-------------+---------------------------------------------------------------
CCM enc | 35.07 125.40 336.47 468.17 581.97 619.18 712.56 736.01
CCM dec | 34.87 124.40 335.08 466.75 581.04 618.81 712.25 735.89
CCM mb enc | 34.71 123.96 333.92 465.39 579.91 617.49 711.45 734.92
CCM mb dec | 34.42 122.80 331.02 462.81 578.28 616.42 709.88 734.19
After (rfc4309(ccm-sm4-ce)):
ccm-sm4-ce | 16 64 256 512 1024 1420 4096 8192
-------------+---------------------------------------------------------------
CCM enc | 77.12 249.82 569.94 725.17 839.27 867.71 952.87 969.89
CCM dec | 75.90 247.26 566.29 722.12 836.90 865.95 951.74 968.57
CCM mb enc | 75.98 245.25 562.91 718.99 834.76 864.70 950.17 967.90
CCM mb dec | 75.06 243.78 560.58 717.13 833.68 862.70 949.35 967.11
Signed-off-by: Tianjia Zhang <tianjia.zhang@linux.alibaba.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
This patch is a CE-optimized assembly implementation for cmac/xcbc/cbcmac.
Benchmark on T-Head Yitian-710 2.75 GHz, the data comes from the 300 mode of
tcrypt, and compared the performance before and after this patch (the driver
used before this patch is XXXmac(sm4-ce)). The abscissas are blocks of
different lengths. The data is tabulated and the unit is Mb/s:
Before:
update-size | 16 64 256 1024 2048 4096 8192
---------------+--------------------------------------------------------
cmac(sm4-ce) | 293.33 403.69 503.76 527.78 531.10 535.46 535.81
xcbc(sm4-ce) | 292.83 402.50 504.02 529.08 529.87 536.55 538.24
cbcmac(sm4-ce) | 318.42 415.79 497.12 515.05 523.15 521.19 523.01
After:
update-size | 16 64 256 1024 2048 4096 8192
---------------+--------------------------------------------------------
cmac-sm4-ce | 371.99 675.28 903.56 971.65 980.57 990.40 991.04
xcbc-sm4-ce | 372.11 674.55 903.47 971.61 980.96 990.42 991.10
cbcmac-sm4-ce | 371.63 675.33 903.23 972.07 981.42 990.93 991.45
Signed-off-by: Tianjia Zhang <tianjia.zhang@linux.alibaba.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
This patch is a CE-optimized assembly implementation for XTS mode.
Benchmark on T-Head Yitian-710 2.75 GHz, the data comes from the 218 mode of
tcrypt, and compared the performance before and after this patch (the driver
used before this patch is xts(ecb-sm4-ce)). The abscissas are blocks of
different lengths. The data is tabulated and the unit is Mb/s:
Before:
xts(ecb-sm4-ce) | 16 64 128 256 1024 1420 4096
----------------+--------------------------------------------------------------
XTS enc | 117.17 430.56 732.92 1134.98 2007.03 2136.23 2347.20
XTS dec | 116.89 429.02 733.40 1132.96 2006.13 2130.50 2347.92
After:
xts-sm4-ce | 16 64 128 256 1024 1420 4096
----------------+--------------------------------------------------------------
XTS enc | 224.68 798.91 1248.08 1714.60 2413.73 2467.84 2612.62
XTS dec | 229.85 791.34 1237.79 1720.00 2413.30 2473.84 2611.95
Signed-off-by: Tianjia Zhang <tianjia.zhang@linux.alibaba.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
This patch is a CE-optimized assembly implementation for CTS-CBC mode.
Benchmark on T-Head Yitian-710 2.75 GHz, the data comes from the 218 mode of
tcrypt, and compared the performance before and after this patch (the driver
used before this patch is cts(cbc-sm4-ce)). The abscissas are blocks of
different lengths. The data is tabulated and the unit is Mb/s:
Before:
cts(cbc-sm4-ce) | 16 64 128 256 1024 1420 4096
----------------+--------------------------------------------------------------
CTS-CBC enc | 286.09 297.17 457.97 627.75 868.58 900.80 957.69
CTS-CBC dec | 286.67 285.63 538.35 947.08 2241.03 2577.32 3391.14
After:
cts-cbc-sm4-ce | 16 64 128 256 1024 1420 4096
----------------+--------------------------------------------------------------
CTS-CBC enc | 288.19 428.80 593.57 741.04 911.73 931.80 950.00
CTS-CBC dec | 292.22 468.99 838.23 1380.76 2741.17 3036.42 3409.62
Signed-off-by: Tianjia Zhang <tianjia.zhang@linux.alibaba.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
In the accelerated implementation of the SM4 algorithm using the Crypto
Extension instructions, there are some functions that can be reused in
the upcoming accelerated implementation of the GCM/CCM mode, and the
CBC/CFB encryption is reused in the optimized implementation of SVESM4.
Signed-off-by: Tianjia Zhang <tianjia.zhang@linux.alibaba.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Use a 128-bit swap mask and tbl instruction to simplify the implementation
for generating SM4 rkey_dec.
Also fixed the issue of not being wrapped by kernel_neon_begin/end() when
using the sm4_ce_expand_key() function.
Signed-off-by: Tianjia Zhang <tianjia.zhang@linux.alibaba.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
This patch does not add new features, but only refactors and simplifies the
implementation of the Crypto Extension acceleration of the SM4 algorithm:
Extract the macro optimized by SM4 Crypto Extension for reuse in the
subsequent optimization of CCM/GCM modes.
Encryption in CBC and CFB modes processes four blocks at a time instead of
one, allowing the ld1 instruction to load 64 bytes of data at a time, which
will reduces unnecessary memory accesses.
CBC/CFB/CTR makes full use of free registers to reduce redundant memory
accesses, and rearranges some instructions to improve out-of-order execution
capabilities.
Signed-off-by: Tianjia Zhang <tianjia.zhang@linux.alibaba.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
This patch does not add new features. The main work is to refactor and
simplify the implementation of SM4 NEON, which is reflected in the
following aspects:
The accelerated implementation supports the arbitrary number of blocks,
not just multiples of 8, which simplifies the implementation and brings
some optimization acceleration for data that is not aligned by 8 blocks.
When loading the input data, use the ld4 instruction to replace the
original ld1 instruction as much as possible, which will save the cost
of matrix transposition of the input data.
Use 8-block parallelism whenever possible to speed up matrix transpose
and rotation operations, instead of up to 4-block parallelism.
Signed-off-by: Tianjia Zhang <tianjia.zhang@linux.alibaba.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
This patch adds the NEON acceleration implementation of the SM3 hash
algorithm. The main algorithm is based on SM3 NEON accelerated work of
the libgcrypt project.
Benchmark on T-Head Yitian-710 2.75 GHz, the data comes from the 326 mode
of tcrypt, and compares the performance data of sm3-generic and sm3-ce.
The abscissas are blocks of different lengths. The data is tabulated and
the unit is Mb/s:
update-size | 16 64 256 1024 2048 4096 8192
---------------+--------------------------------------------------------
sm3-generic | 185.24 221.28 301.26 307.43 300.83 308.82 308.91
sm3-neon | 171.81 220.20 322.94 339.28 334.09 343.61 343.87
sm3-ce | 227.48 333.48 502.62 527.87 520.45 534.91 535.40
Signed-off-by: Tianjia Zhang <tianjia.zhang@linux.alibaba.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Raise the priority of the sm3-ce algorithm from 200 to 400, this is
to make room for the implementation of sm3-neon.
Signed-off-by: Tianjia Zhang <tianjia.zhang@linux.alibaba.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
With CONFIG_CFI_CLANG, assembly functions indirectly called from C
code must be annotated with type identifiers to pass CFI checking. Use
SYM_TYPED_FUNC_START for the indirectly called functions, and ensure
we emit `bti c` also with SYM_TYPED_FUNC_START.
Signed-off-by: Sami Tolvanen <samitolvanen@google.com>
Reviewed-by: Kees Cook <keescook@chromium.org>
Tested-by: Kees Cook <keescook@chromium.org>
Tested-by: Nathan Chancellor <nathan@kernel.org>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Tested-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Kees Cook <keescook@chromium.org>
Link: https://lore.kernel.org/r/20220908215504.3686827-10-samitolvanen@google.com
Commit 3f342a2325 ("crypto: Kconfig - simplify hash entries") makes
various changes to the config descriptions as part of some consolidation
and clean-up, but among all those changes, it also accidently renames
CRYPTO_SHA1_ARM64_CE to CRYPTO_SHA1_ARM64.
Revert this unintended config name change.
See Link for the author's confirmation of this happening accidently.
Fixes: 3f342a2325 ("crypto: Kconfig - simplify hash entries")
Link: https://lore.kernel.org/all/MW5PR84MB18424AB8C095BFC041AE33FDAB479@MW5PR84MB1842.NAMPRD84.PROD.OUTLOOK.COM/
Signed-off-by: Lukas Bulwahn <lukas.bulwahn@gmail.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
