Instead of providing crypto_shash algorithms for the arch-optimized
SHA-256 code, instead implement the SHA-256 library. This is much
simpler, it makes the SHA-256 library functions be arch-optimized, and
it fixes the longstanding issue where the arch-optimized SHA-256 was
disabled by default. SHA-256 still remains available through
crypto_shash, but individual architectures no longer need to handle it.
To match sha256_blocks_arch(), change the type of the nblocks parameter
of the assembly functions from int to size_t. The assembly functions
actually already treated it as size_t.
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Continue disentangling the crypto library functions from the generic
crypto infrastructure by moving the x86 BLAKE2s, ChaCha, and Poly1305
library functions into a new directory arch/x86/lib/crypto/ that does
not depend on CRYPTO. This mirrors the distinction between crypto/ and
lib/crypto/.
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
arch/x86/crypto/Kconfig is sourced only when CONFIG_X86=y, so there is
no need for the symbols defined inside it to depend on X86.
In the case of CRYPTO_TWOFISH_586 and CRYPTO_TWOFISH_X86_64, the
dependency was actually on '(X86 || UML_X86)', which suggests that these
two symbols were intended to be available under user-mode Linux as well.
Yet, again these symbols were defined only when CONFIG_X86=y, so that
was not the case. Just remove this redundant dependency.
Acked-by: Ard Biesheuvel <ardb@kernel.org>
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Now that all sha256_base users have been converted to use the API
partial block handling, remove the partial block helpers.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Though the module_exit functions are now no-ops, they should still be
defined, since otherwise the modules become unremovable.
Fixes: 1f81c58279 ("crypto: arm/poly1305 - remove redundant shash algorithm")
Fixes: f4b1a73aec ("crypto: arm64/poly1305 - remove redundant shash algorithm")
Fixes: 378a337ab4 ("crypto: powerpc/poly1305 - implement library instead of shash")
Fixes: 21969da642 ("crypto: x86/poly1305 - remove redundant shash algorithm")
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Though the module_exit functions are now no-ops, they should still be
defined, since otherwise the modules become unremovable.
Fixes: 08820553f3 ("crypto: arm/chacha - remove the redundant skcipher algorithms")
Fixes: 8c28abede1 ("crypto: arm64/chacha - remove the skcipher algorithms")
Fixes: f7915484c0 ("crypto: powerpc/chacha - remove the skcipher algorithms")
Fixes: ceba0eda83 ("crypto: riscv/chacha - implement library instead of skcipher")
Fixes: 632ab0978f ("crypto: x86/chacha - remove the skcipher algorithms")
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
The x86 Poly1305 code never falls back to the generic code, so selecting
CRYPTO_LIB_POLY1305_GENERIC is unnecessary.
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Since crypto/poly1305.c now registers a poly1305-$(ARCH) shash algorithm
that uses the architecture's Poly1305 library functions, individual
architectures no longer need to do the same. Therefore, remove the
redundant shash algorithm from the arch-specific code and leave just the
library functions there.
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Following the example of the crc32, crc32c, and chacha code, make the
crypto subsystem register both generic and architecture-optimized
poly1305 shash algorithms, both implemented on top of the appropriate
library functions. This eliminates the need for every architecture to
implement the same shash glue code.
Note that the poly1305 shash requires that the key be prepended to the
data, which differs from the library functions where the key is simply a
parameter to poly1305_init(). Previously this was handled at a fairly
low level, polluting the library code with shash-specific code.
Reorganize things so that the shash code handles this quirk itself.
Also, to register the architecture-optimized shashes only when
architecture-optimized code is actually being used, add a function
poly1305_is_arch_optimized() and make each arch implement it. Change
each architecture's Poly1305 module_init function to arch_initcall so
that the CPU feature detection is guaranteed to run before
poly1305_is_arch_optimized() gets called by crypto/poly1305.c. (In
cases where poly1305_is_arch_optimized() just returns true
unconditionally, using arch_initcall is not strictly needed, but it's
still good to be consistent across architectures.)
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
The chacha_use_simd static branch is required for x86 machines that
lack SSSE3 support. Restore it and the generic fallback code.
Reported-by: Eric Biggers <ebiggers@kernel.org>
Fixes: 9b4400215e ("crypto: x86/chacha - Remove SIMD fallback path")
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Since crypto/chacha.c now registers chacha20-$(ARCH), xchacha20-$(ARCH),
and xchacha12-$(ARCH) skcipher algorithms that use the architecture's
ChaCha and HChaCha library functions, individual architectures no longer
need to do the same. Therefore, remove the redundant skcipher
algorithms and leave just the library functions.
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Following the example of the crc32 and crc32c code, make the crypto
subsystem register both generic and architecture-optimized chacha20,
xchacha20, and xchacha12 skcipher algorithms, all implemented on top of
the appropriate library functions. This eliminates the need for every
architecture to implement the same skcipher glue code.
To register the architecture-optimized skciphers only when
architecture-optimized code is actually being used, add a function
chacha_is_arch_optimized() and make each arch implement it. Change each
architecture's ChaCha module_init function to arch_initcall so that the
CPU feature detection is guaranteed to run before
chacha_is_arch_optimized() gets called by crypto/chacha.c. In the case
of s390, remove the CPU feature based module autoloading, which is no
longer needed since the module just gets pulled in via function linkage.
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Optimize the AVX-512 version of _compute_first_set_of_tweaks by using
vectorized shifts to compute the first vector of tweak blocks, and by
using byte-aligned shifts when multiplying by x^8.
AES-XTS performance on AMD Ryzen 9 9950X (Zen 5) improves by about 2%
for 4096-byte messages or 6% for 512-byte messages. AES-XTS performance
on Intel Sapphire Rapids improves by about 1% for 4096-byte messages or
3% for 512-byte messages. Code size decreases by 75 bytes which
outweighs the increase in rodata size of 16 bytes.
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Stop wrapping skcipher and aead algorithms with the crypto SIMD helper
(crypto/simd.c). The only purpose of doing so was to work around x86
not always supporting kernel-mode FPU in softirqs. Specifically, if a
hardirq interrupted a task context kernel-mode FPU section and then a
softirqs were run at the end of that hardirq, those softirqs could not
use kernel-mode FPU. This has now been fixed. In combination with the
fact that the skcipher and aead APIs only support task and softirq
contexts, these can now just use kernel-mode FPU unconditionally on x86.
This simplifies the code and improves performance.
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Stop wrapping skcipher and aead algorithms with the crypto SIMD helper
(crypto/simd.c). The only purpose of doing so was to work around x86
not always supporting kernel-mode FPU in softirqs. Specifically, if a
hardirq interrupted a task context kernel-mode FPU section and then a
softirqs were run at the end of that hardirq, those softirqs could not
use kernel-mode FPU. This has now been fixed. In combination with the
fact that the skcipher and aead APIs only support task and softirq
contexts, these can now just use kernel-mode FPU unconditionally on x86.
This simplifies the code and improves performance.
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Stop wrapping skcipher and aead algorithms with the crypto SIMD helper
(crypto/simd.c). The only purpose of doing so was to work around x86
not always supporting kernel-mode FPU in softirqs. Specifically, if a
hardirq interrupted a task context kernel-mode FPU section and then a
softirqs were run at the end of that hardirq, those softirqs could not
use kernel-mode FPU. This has now been fixed. In combination with the
fact that the skcipher and aead APIs only support task and softirq
contexts, these can now just use kernel-mode FPU unconditionally on x86.
This simplifies the code and improves performance.
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Stop wrapping skcipher and aead algorithms with the crypto SIMD helper
(crypto/simd.c). The only purpose of doing so was to work around x86
not always supporting kernel-mode FPU in softirqs. Specifically, if a
hardirq interrupted a task context kernel-mode FPU section and then a
softirqs were run at the end of that hardirq, those softirqs could not
use kernel-mode FPU. This has now been fixed. In combination with the
fact that the skcipher and aead APIs only support task and softirq
contexts, these can now just use kernel-mode FPU unconditionally on x86.
This simplifies the code and improves performance.
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Stop wrapping skcipher and aead algorithms with the crypto SIMD helper
(crypto/simd.c). The only purpose of doing so was to work around x86
not always supporting kernel-mode FPU in softirqs. Specifically, if a
hardirq interrupted a task context kernel-mode FPU section and then a
softirqs were run at the end of that hardirq, those softirqs could not
use kernel-mode FPU. This has now been fixed. In combination with the
fact that the skcipher and aead APIs only support task and softirq
contexts, these can now just use kernel-mode FPU unconditionally on x86.
This simplifies the code and improves performance.
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Stop wrapping skcipher and aead algorithms with the crypto SIMD helper
(crypto/simd.c). The only purpose of doing so was to work around x86
not always supporting kernel-mode FPU in softirqs. Specifically, if a
hardirq interrupted a task context kernel-mode FPU section and then a
softirqs were run at the end of that hardirq, those softirqs could not
use kernel-mode FPU. This has now been fixed. In combination with the
fact that the skcipher and aead APIs only support task and softirq
contexts, these can now just use kernel-mode FPU unconditionally on x86.
This simplifies the code and improves performance.
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Stop wrapping skcipher and aead algorithms with the crypto SIMD helper
(crypto/simd.c). The only purpose of doing so was to work around x86
not always supporting kernel-mode FPU in softirqs. Specifically, if a
hardirq interrupted a task context kernel-mode FPU section and then a
softirqs were run at the end of that hardirq, those softirqs could not
use kernel-mode FPU. This has now been fixed. In combination with the
fact that the skcipher and aead APIs only support task and softirq
contexts, these can now just use kernel-mode FPU unconditionally on x86.
This simplifies the code and improves performance.
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Stop wrapping skcipher and aead algorithms with the crypto SIMD helper
(crypto/simd.c). The only purpose of doing so was to work around x86
not always supporting kernel-mode FPU in softirqs. Specifically, if a
hardirq interrupted a task context kernel-mode FPU section and then a
softirqs were run at the end of that hardirq, those softirqs could not
use kernel-mode FPU. This has now been fixed. In combination with the
fact that the skcipher and aead APIs only support task and softirq
contexts, these can now just use kernel-mode FPU unconditionally on x86.
This simplifies the code and improves performance.
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Intel made a late change to the AVX10 specification that removes support
for a 256-bit maximum vector length and enumeration of the maximum
vector length. AVX10 will imply a maximum vector length of 512 bits.
I.e. there won't be any such thing as AVX10/256 or AVX10/512; there will
just be AVX10, and it will essentially just consolidate AVX512 features.
As a result of this new development, my strategy of providing both
*_avx10_256 and *_avx10_512 functions didn't turn out to be that useful.
The only remaining motivation for the 256-bit AVX512 / AVX10 functions
is to avoid downclocking on older Intel CPUs. But in the case of
AES-XTS and AES-CTR, I already wrote *_avx2 code too (primarily to
support CPUs without AVX512), which performs almost as well as
*_avx10_256. So we should just use that.
Therefore, remove the *_avx10_256 AES-XTS and AES-CTR functions and
algorithms, and rename the *_avx10_512 AES-XTS and AES-CTR functions and
algorithms to *_avx512. Make Ice Lake and Tiger Lake use *_avx2 instead
of *_avx10_256 which they previously used.
I've left AES-GCM unchanged for now. There is no VAES+AVX2 optimized
AES-GCM in the kernel yet, so the path forward for that is not as clear.
However, I did write a VAES+AVX2 optimized AES-GCM for BoringSSL. So
one option is to port that to the kernel and then do the same cleanup.
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Pull CRC updates from Eric Biggers:
"Another set of improvements to the kernel's CRC (cyclic redundancy
check) code:
- Rework the CRC64 library functions to be directly optimized, like
what I did last cycle for the CRC32 and CRC-T10DIF library
functions
- Rewrite the x86 PCLMULQDQ-optimized CRC code, and add VPCLMULQDQ
support and acceleration for crc64_be and crc64_nvme
- Rewrite the riscv Zbc-optimized CRC code, and add acceleration for
crc_t10dif, crc64_be, and crc64_nvme
- Remove crc_t10dif and crc64_rocksoft from the crypto API, since
they are no longer needed there
- Rename crc64_rocksoft to crc64_nvme, as the old name was incorrect
- Add kunit test cases for crc64_nvme and crc7
- Eliminate redundant functions for calculating the Castagnoli CRC32,
settling on just crc32c()
- Remove unnecessary prompts from some of the CRC kconfig options
- Further optimize the x86 crc32c code"
* tag 'crc-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/ebiggers/linux: (36 commits)
x86/crc: drop the avx10_256 functions and rename avx10_512 to avx512
lib/crc: remove unnecessary prompt for CONFIG_CRC64
lib/crc: remove unnecessary prompt for CONFIG_LIBCRC32C
lib/crc: remove unnecessary prompt for CONFIG_CRC8
lib/crc: remove unnecessary prompt for CONFIG_CRC7
lib/crc: remove unnecessary prompt for CONFIG_CRC4
lib/crc7: unexport crc7_be_syndrome_table
lib/crc_kunit.c: update comment in crc_benchmark()
lib/crc_kunit.c: add test and benchmark for crc7_be()
x86/crc32: optimize tail handling for crc32c short inputs
riscv/crc64: add Zbc optimized CRC64 functions
riscv/crc-t10dif: add Zbc optimized CRC-T10DIF function
riscv/crc32: reimplement the CRC32 functions using new template
riscv/crc: add "template" for Zbc optimized CRC functions
x86/crc: add ANNOTATE_NOENDBR to suppress objtool warnings
x86/crc32: improve crc32c_arch() code generation with clang
x86/crc64: implement crc64_be and crc64_nvme using new template
x86/crc-t10dif: implement crc_t10dif using new template
x86/crc32: implement crc32_le using new template
x86/crc: add "template" for [V]PCLMULQDQ based CRC functions
...
All implementations of chacha_init_arch() just call
chacha_init_generic(), so it is pointless. Just delete it, and replace
chacha_init() with what was previously chacha_init_generic().
Signed-off-by: Eric Biggers <ebiggers@google.com>
Acked-by: Ard Biesheuvel <ardb@kernel.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Mark the src.virt.addr field in struct skcipher_walk as a pointer
to const data. This guarantees that the user won't modify the data
which should be done through dst.virt.addr to ensure that flushing
is done when necessary.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Rather than returning the address and storing the length into an
argument pointer, add an address field to the walk struct and use
that to store the address. The length is returned directly.
Change the done functions to use this stored address instead of
getting them from the caller.
Split the address into two using a union. The user should only
access the const version so that it is never changed.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
The ARCH_MAY_HAVE patch missed arm64, mips and s390. But it may
also lead to arch options being enabled but ineffective because
of modular/built-in conflicts.
As the primary user of all these options wireguard is selecting
the arch options anyway, make the same selections at the lib/crypto
option level and hide the arch options from the user.
Instead of selecting them centrally from lib/crypto, simply set
the default of each arch option as suggested by Eric Biggers.
Change the Crypto API generic algorithms to select the top-level
lib/crypto options instead of the generic one as otherwise there
is no way to enable the arch options (Eric Biggers). Introduce a
set of INTERNAL options to work around dependency cycles on the
CONFIG_CRYPTO symbol.
Fixes: 1047e21aec ("crypto: lib/Kconfig - Fix lib built-in failure when arch is modular")
Reported-by: kernel test robot <lkp@intel.com>
Reported-by: Arnd Bergmann <arnd@kernel.org>
Closes: https://lore.kernel.org/oe-kbuild-all/202502232152.JC84YDLp-lkp@intel.com/
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
In crypto_aegis128_aesni_process_ad(), use scatterwalk_next() which
consolidates scatterwalk_clamp() and scatterwalk_map(). Use
scatterwalk_done_src() which consolidates scatterwalk_unmap(),
scatterwalk_advance(), and scatterwalk_done().
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
In gcm_process_assoc(), use scatterwalk_next() which consolidates
scatterwalk_clamp() and scatterwalk_map(). Use scatterwalk_done_src()
which consolidates scatterwalk_unmap(), scatterwalk_advance(), and
scatterwalk_done().
Also rename some variables to avoid implying that anything is actually
mapped (it's not), or that the loop is going page by page (it is for
now, but nothing actually requires that to be the case).
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Rather than copying a request by hand with memcpy, use the correct
API helpers to setup the new request. This will matter once the
API helpers start setting up chained requests as a simple memcpy
will break chaining.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
The HAVE_ARCH Kconfig options in lib/crypto try to solve the
modular versus built-in problem, but it still fails when the
the LIB option (e.g., CRYPTO_LIB_CURVE25519) is selected externally.
Fix this by introducing a level of indirection with ARCH_MAY_HAVE
Kconfig options, these then go on to select the ARCH_HAVE options
if the ARCH Kconfig options matches that of the LIB option.
Reported-by: kernel test robot <lkp@intel.com>
Closes: https://lore.kernel.org/oe-kbuild-all/202501230223.ikroNDr1-lkp@intel.com/
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
As with the other AES modes I've implemented, I've received interest in
my AES-XTS assembly code being reused in other projects. Therefore,
change the license to Apache-2.0 OR BSD-2-Clause like what I used for
AES-GCM. Apache-2.0 is the license of OpenSSL and BoringSSL.
Note that it is difficult to *directly* share code between the kernel,
OpenSSL, and BoringSSL for various reasons such as perlasm vs. plain
asm, Windows ABI support, different divisions of responsibility between
C and asm in each project, etc. So whether that will happen instead of
just doing ports is still TBD. But this dual license should at least
make it possible to port changes between the projects.
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Delete aes_ctrby8_avx-x86_64.S and add a new assembly file
aes-ctr-avx-x86_64.S which follows a similar approach to
aes-xts-avx-x86_64.S in that it uses a "template" to provide AESNI+AVX,
VAES+AVX2, VAES+AVX10/256, and VAES+AVX10/512 code, instead of just
AESNI+AVX. Wire it up to the crypto API accordingly.
This greatly improves the performance of AES-CTR and AES-XCTR on
VAES-capable CPUs, with the best case being AMD Zen 5 where an over 230%
increase in throughput is seen on long messages. Performance on
non-VAES-capable CPUs remains about the same, and the non-AVX AES-CTR
code (aesni_ctr_enc) is also kept as-is for now. There are some slight
regressions (less than 10%) on some short message lengths on some CPUs;
these are difficult to avoid, given how the previous code was so heavily
unrolled by message length, and they are not particularly important.
Detailed performance results are given in the tables below.
Both CTR and XCTR support is retained. The main loop remains
8-vector-wide, which differs from the 4-vector-wide main loops that are
used in the XTS and GCM code. A wider loop is appropriate for CTR and
XCTR since they have fewer other instructions (such as vpclmulqdq) to
interleave with the AES instructions.
Similar to what was the case for AES-GCM, the new assembly code also has
a much smaller binary size, as it fixes the excessive unrolling by data
length and key length present in the old code. Specifically, the new
assembly file compiles to about 9 KB of text vs. 28 KB for the old file.
This is despite 4x as many implementations being included.
The tables below show the detailed performance results. The tables show
percentage improvement in single-threaded throughput for repeated
encryption of the given message length; an increase from 6000 MB/s to
12000 MB/s would be listed as 100%. They were collected by directly
measuring the Linux crypto API performance using a custom kernel module.
The tested CPUs were all server processors from Google Compute Engine
except for Zen 5 which was a Ryzen 9 9950X desktop processor.
Table 1: AES-256-CTR throughput improvement,
CPU microarchitecture vs. message length in bytes:
| 16384 | 4096 | 4095 | 1420 | 512 | 500 |
---------------------+-------+-------+-------+-------+-------+-------+
AMD Zen 5 | 232% | 203% | 212% | 143% | 71% | 95% |
Intel Emerald Rapids | 116% | 116% | 117% | 91% | 78% | 79% |
Intel Ice Lake | 109% | 103% | 107% | 81% | 54% | 56% |
AMD Zen 4 | 109% | 91% | 100% | 70% | 43% | 59% |
AMD Zen 3 | 92% | 78% | 87% | 57% | 32% | 43% |
AMD Zen 2 | 9% | 8% | 14% | 12% | 8% | 21% |
Intel Skylake | 7% | 7% | 8% | 5% | 3% | 8% |
| 300 | 200 | 64 | 63 | 16 |
---------------------+-------+-------+-------+-------+-------+
AMD Zen 5 | 57% | 39% | -9% | 7% | -7% |
Intel Emerald Rapids | 37% | 42% | -0% | 13% | -8% |
Intel Ice Lake | 39% | 30% | -1% | 14% | -9% |
AMD Zen 4 | 42% | 38% | -0% | 18% | -3% |
AMD Zen 3 | 38% | 35% | 6% | 31% | 5% |
AMD Zen 2 | 24% | 23% | 5% | 30% | 3% |
Intel Skylake | 9% | 1% | -4% | 10% | -7% |
Table 2: AES-256-XCTR throughput improvement,
CPU microarchitecture vs. message length in bytes:
| 16384 | 4096 | 4095 | 1420 | 512 | 500 |
---------------------+-------+-------+-------+-------+-------+-------+
AMD Zen 5 | 240% | 201% | 216% | 151% | 75% | 108% |
Intel Emerald Rapids | 100% | 99% | 102% | 91% | 94% | 104% |
Intel Ice Lake | 93% | 89% | 92% | 74% | 50% | 64% |
AMD Zen 4 | 86% | 75% | 83% | 60% | 41% | 52% |
AMD Zen 3 | 73% | 63% | 69% | 45% | 21% | 33% |
AMD Zen 2 | -2% | -2% | 2% | 3% | -1% | 11% |
Intel Skylake | -1% | -1% | 1% | 2% | -1% | 9% |
| 300 | 200 | 64 | 63 | 16 |
---------------------+-------+-------+-------+-------+-------+
AMD Zen 5 | 78% | 56% | -4% | 38% | -2% |
Intel Emerald Rapids | 61% | 55% | 4% | 32% | -5% |
Intel Ice Lake | 57% | 42% | 3% | 44% | -4% |
AMD Zen 4 | 35% | 28% | -1% | 17% | -3% |
AMD Zen 3 | 26% | 23% | -3% | 11% | -6% |
AMD Zen 2 | 13% | 24% | -1% | 14% | -3% |
Intel Skylake | 16% | 8% | -4% | 35% | -3% |
Signed-off-by: Eric Biggers <ebiggers@google.com>
Reviewed-by: Ard Biesheuvel <ardb@kernel.org>
Tested-by: Ard Biesheuvel <ardb@kernel.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
With the introduction of kCFI the addition of ENDBR to
SYM_FUNC_START* no longer suffices to make the function indirectly
callable. This now requires the use of SYM_TYPED_FUNC_START.
As such, remove the implicit ENDBR from SYM_FUNC_START* and add some
explicit annotations to fix things up again.
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Sami Tolvanen <samitolvanen@google.com>
Link: https://lore.kernel.org/r/20250207122546.409116003@infradead.org
The expectation is that all EXPORT'ed symbols are free to have their
address taken and called indirectly. The majority of the assembly
defined functions currently violate this expectation.
Make then all use SYM_TYPED_FUNC_START() in order to emit the proper
kCFI preamble.
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Sami Tolvanen <samitolvanen@google.com>
Link: https://lore.kernel.org/r/20250207122546.302679189@infradead.org
Lift zmm_exclusion_list in aesni-intel_glue.c into the x86 CPU setup
code, and add a new x86 CPU feature flag X86_FEATURE_PREFER_YMM that is
set when the CPU is on this list.
This allows other code in arch/x86/, such as the CRC library code, to
apply the same exclusion list when deciding whether to execute 256-bit
or 512-bit optimized functions.
Note that full AVX512 support including ZMM registers is still exposed
to userspace and is still supported for in-kernel use. This flag just
indicates whether in-kernel code should prefer to use YMM registers.
Acked-by: Ard Biesheuvel <ardb@kernel.org>
Acked-by: Ingo Molnar <mingo@kernel.org>
Acked-by: Keith Busch <kbusch@kernel.org>
Reviewed-by: "Martin K. Petersen" <martin.petersen@oracle.com>
Link: https://lore.kernel.org/r/20250210174540.161705-2-ebiggers@kernel.org
Signed-off-by: Eric Biggers <ebiggers@google.com>
Remove 32-bit support from the fast path in xts_crypt(). Then optimize
it for 64-bit, and simplify the code, by switching to sg_virt() and
removing the now-unnecessary checks for crossing a page boundary.
The result is simpler code that is slightly smaller and faster in the
case that actually matters (64-bit).
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
