Instead of exposing the x86-optimized SM3 code via an x86-specific
crypto_shash algorithm, instead just implement the sm3_blocks() library
function. This is much simpler, it makes the SM3 library functions be
x86-optimized, and it fixes the longstanding issue where the
x86-optimized SM3 code was disabled by default. SM3 still remains
available through crypto_shash, but individual architectures no longer
need to handle it.
Tweak the prototype of sm3_transform_avx() to match what the library
expects, including changing the block count to size_t. Note that the
assembly code actually already treated this argument as size_t.
Acked-by: Ard Biesheuvel <ardb@kernel.org>
Link: https://lore.kernel.org/r/20260321040935.410034-10-ebiggers@kernel.org
Signed-off-by: Eric Biggers <ebiggers@kernel.org>
Remove the "ghash-pclmulqdqni" crypto_shash algorithm. Move the
corresponding assembly code into lib/crypto/, and wire it up to the
GHASH library.
This makes the GHASH library be optimized with x86's carryless
multiplication instructions. It also greatly reduces the amount of
x86-specific glue code that is needed, and it fixes the issue where this
GHASH optimization was disabled by default.
Rename and adjust the prototypes of the assembly functions to make them
fit better with the library. Remove the byte-swaps (pshufb
instructions) that are no longer necessary because the library keeps the
accumulator in POLYVAL format rather than GHASH format.
Rename clmul_ghash_mul() to polyval_mul_pclmul() to reflect that it
really does a POLYVAL style multiplication. Wire it up to both
ghash_mul_arch() and polyval_mul_arch().
Acked-by: Ard Biesheuvel <ardb@kernel.org>
Link: https://lore.kernel.org/r/20260319061723.1140720-15-ebiggers@kernel.org
Signed-off-by: Eric Biggers <ebiggers@kernel.org>
Currently, some architectures (arm64 and x86) have optimized code for
both GHASH and POLYVAL. Others (arm, powerpc, riscv, and s390) have
optimized code only for GHASH. While POLYVAL support could be
implemented on these other architectures, until then we need to support
the case where arch-optimized functions are present only for GHASH.
Therefore, update the support for arch-optimized POLYVAL functions to
allow architectures to opt into supporting these functions individually.
The new meaning of CONFIG_CRYPTO_LIB_GF128HASH_ARCH is that some level
of GHASH and/or POLYVAL acceleration is provided.
Also provide an implementation of polyval_mul() based on
polyval_blocks_arch(), for when polyval_mul_arch() isn't implemented.
Acked-by: Ard Biesheuvel <ardb@kernel.org>
Link: https://lore.kernel.org/r/20260319061723.1140720-3-ebiggers@kernel.org
Signed-off-by: Eric Biggers <ebiggers@kernel.org>
Currently, the standalone GHASH code is coupled with crypto_shash. This
has resulted in unnecessary complexity and overhead, as well as the code
being unavailable to library code such as the AES-GCM library. Like was
done with POLYVAL, it needs to find a new home in lib/crypto/.
GHASH and POLYVAL are closely related and can each be implemented in
terms of each other. Optimized code for one can be reused with the
other. But also since GHASH tends to be difficult to implement directly
due to its unnatural bit order, most modern GHASH implementations
(including the existing arm, arm64, powerpc, and x86 optimized GHASH
code, and the new generic GHASH code I'll be adding) actually
reinterpret the GHASH computation as an equivalent POLYVAL computation,
pre and post-processing the inputs and outputs to map to/from POLYVAL.
Given this close relationship, it makes sense to group the GHASH and
POLYVAL code together in the same module. This gives us a wide range of
options for implementing them, reusing code between the two and properly
utilizing whatever instructions each architecture provides.
Thus, GHASH support will be added to the library module that is
currently called "polyval". Rename it to an appropriate name:
"gf128hash". Rename files, options, functions, etc. where appropriate
to reflect the upcoming sharing with GHASH. (Note: polyval_kunit is not
renamed, as ghash_kunit will be added alongside it instead.)
Acked-by: Ard Biesheuvel <ardb@kernel.org>
Link: https://lore.kernel.org/r/20260319061723.1140720-2-ebiggers@kernel.org
Signed-off-by: Eric Biggers <ebiggers@kernel.org>
Optimize the AES library with x86 AES-NI instructions.
The relevant existing assembly functions, aesni_set_key(), aesni_enc(),
and aesni_dec(), are a bit difficult to extract into the library:
- They're coupled to the code for the AES modes.
- They operate on struct crypto_aes_ctx. The AES library now uses
different structs.
- They assume the key is 16-byte aligned. The AES library only
*prefers* 16-byte alignment; it doesn't require it.
Moreover, they're not all that great in the first place:
- They use unrolled loops, which isn't a great choice on x86.
- They use the 'aeskeygenassist' instruction, which is unnecessary, is
slow on Intel CPUs, and forces the loop to be unrolled.
- They have special code for AES-192 key expansion, despite that being
kind of useless. AES-128 and AES-256 are the ones used in practice.
These are small functions anyway.
Therefore, I opted to just write replacements of these functions for the
library. They address all the above issues.
Acked-by: Ard Biesheuvel <ardb@kernel.org>
Link: https://lore.kernel.org/r/20260112192035.10427-18-ebiggers@kernel.org
Signed-off-by: Eric Biggers <ebiggers@kernel.org>
Migrate the x86_64 implementations of NH into lib/crypto/. This makes
the nh() function be optimized on x86_64 kernels.
Note: this temporarily makes the adiantum template not utilize the
x86_64 optimized NH code. This is resolved in a later commit that
converts the adiantum template to use nh() instead of "nhpoly1305".
Link: https://lore.kernel.org/r/20251211011846.8179-6-ebiggers@kernel.org
Signed-off-by: Eric Biggers <ebiggers@kernel.org>
Migrate the x86_64 implementation of POLYVAL into lib/crypto/, wiring it
up to the POLYVAL library interface. This makes the POLYVAL library be
properly optimized on x86_64.
This drops the x86_64 optimizations of polyval in the crypto_shash API.
That's fine, since polyval will be removed from crypto_shash entirely
since it is unneeded there. But even if it comes back, the crypto_shash
API could just be implemented on top of the library API, as usual.
Adjust the names and prototypes of the assembly functions to align more
closely with the rest of the library code.
Also replace a movaps instruction with movups to remove the assumption
that the key struct is 16-byte aligned. Users can still align the key
if they want (and at least in this case, movups is just as fast as
movaps), but it's inconvenient to require it.
Reviewed-by: Ard Biesheuvel <ardb@kernel.org>
Link: https://lore.kernel.org/r/20251109234726.638437-6-ebiggers@kernel.org
Signed-off-by: Eric Biggers <ebiggers@kernel.org>
AVX-512 supports 3-input XORs via the vpternlogd (or vpternlogq)
instruction with immediate 0x96. This approach, vs. the alternative of
two vpxor instructions, is already used in the CRC, AES-GCM, and AES-XTS
code, since it reduces the instruction count and is faster on some CPUs.
Make blake2s_compress_avx512() take advantage of it too.
Reviewed-by: Ard Biesheuvel <ardb@kernel.org>
Link: https://lore.kernel.org/r/20251102234209.62133-7-ebiggers@kernel.org
Signed-off-by: Eric Biggers <ebiggers@kernel.org>
Just before returning, blake2s_compress_ssse3() and
blake2s_compress_avx512() store updated values to the 'h', 't', and 'f'
fields of struct blake2s_ctx. But 'f' is always unchanged (which is
correct; only the C code changes it). So, there's no need to write to
'f'. Use 64-bit stores (movq and vmovq) instead of 128-bit stores
(movdqu and vmovdqu) so that only 't' is written.
Reviewed-by: Ard Biesheuvel <ardb@kernel.org>
Link: https://lore.kernel.org/r/20251102234209.62133-6-ebiggers@kernel.org
Signed-off-by: Eric Biggers <ebiggers@kernel.org>
Various cleanups for readability. No change to the generated code:
- Add some comments
- Add #defines for arguments
- Rename some labels
- Use decimal constants instead of hex where it makes sense.
(The pshufd immediates intentionally remain as hex.)
- Add blank lines when there's a logical break
The round loop still could use some work, but this is at least a start.
Reviewed-by: Ard Biesheuvel <ardb@kernel.org>
Link: https://lore.kernel.org/r/20251102234209.62133-5-ebiggers@kernel.org
Signed-off-by: Eric Biggers <ebiggers@kernel.org>
Since blake2s_compress() is always passed nblocks != 0, remove the
unnecessary check for nblocks == 0 from blake2s_compress_ssse3().
Note that this makes it consistent with blake2s_compress_avx512() in the
same file as well as the arm32 blake2s_compress().
Reviewed-by: Ard Biesheuvel <ardb@kernel.org>
Link: https://lore.kernel.org/r/20251102234209.62133-3-ebiggers@kernel.org
Signed-off-by: Eric Biggers <ebiggers@kernel.org>
In the C code, the 'inc' argument to the assembly functions
blake2s_compress_ssse3() and blake2s_compress_avx512() is declared with
type u32, matching blake2s_compress(). The assembly code then reads it
from the 64-bit %rcx. However, the ABI doesn't guarantee zero-extension
to 64 bits, nor do gcc or clang guarantee it. Therefore, fix these
functions to read this argument from the 32-bit %ecx.
In theory, this bug could have caused the wrong 'inc' value to be used,
causing incorrect BLAKE2s hashes. In practice, probably not: I've fixed
essentially this same bug in many other assembly files too, but there's
never been a real report of it having caused a problem. In x86_64, all
writes to 32-bit registers are zero-extended to 64 bits. That results
in zero-extension in nearly all situations. I've only been able to
demonstrate a lack of zero-extension with a somewhat contrived example
involving truncation, e.g. when the C code has a u64 variable holding
0x1234567800000040 and passes it as a u32 expecting it to be truncated
to 0x40 (64). But that's not what the real code does, of course.
Fixes: ed0356eda1 ("crypto: blake2s - x86_64 SIMD implementation")
Cc: stable@vger.kernel.org
Reviewed-by: Ard Biesheuvel <ardb@kernel.org>
Link: https://lore.kernel.org/r/20251102234209.62133-2-ebiggers@kernel.org
Signed-off-by: Eric Biggers <ebiggers@kernel.org>
A couple more small cleanups to the BLAKE2s code before these things get
propagated into the BLAKE2b code:
- Drop 'const' from some non-pointer function parameters. It was a bit
excessive and not conventional.
- Rename 'block' argument of blake2s_compress*() to 'data'. This is for
consistency with the SHA-* code, and also to avoid the implication
that it points to a singular "block".
No functional changes.
Reviewed-by: Ard Biesheuvel <ardb@kernel.org>
Link: https://lore.kernel.org/r/20251018043106.375964-4-ebiggers@kernel.org
Signed-off-by: Eric Biggers <ebiggers@kernel.org>
For consistency with the SHA-1, SHA-2, SHA-3 (in development), and MD5
library APIs, rename blake2s_state to blake2s_ctx.
As a refresher, the ctx name:
- Is a bit shorter.
- Avoids confusion with the compression function state, which is also
often called the state (but is just part of the full context).
- Is consistent with OpenSSL.
Not a big deal, of course. But consistency is nice. With a BLAKE2b
library API about to be added, this is a convenient time to update this.
Reviewed-by: Ard Biesheuvel <ardb@kernel.org>
Link: https://lore.kernel.org/r/20251018043106.375964-3-ebiggers@kernel.org
Signed-off-by: Eric Biggers <ebiggers@kernel.org>
Pull interleaved SHA-256 hashing support from Eric Biggers:
"Optimize fsverity with 2-way interleaved hashing
Add support for 2-way interleaved SHA-256 hashing to lib/crypto/, and
make fsverity use it for faster file data verification. This improves
fsverity performance on many x86_64 and arm64 processors.
Later, I plan to make dm-verity use this too"
* tag 'fsverity-for-linus' of git://git.kernel.org/pub/scm/fs/fsverity/linux:
fsverity: Use 2-way interleaved SHA-256 hashing when supported
fsverity: Remove inode parameter from fsverity_hash_block()
lib/crypto: tests: Add tests and benchmark for sha256_finup_2x()
lib/crypto: x86/sha256: Add support for 2-way interleaved hashing
lib/crypto: arm64/sha256: Add support for 2-way interleaved hashing
lib/crypto: sha256: Add support for 2-way interleaved hashing
Add an implementation of sha256_finup_2x_arch() for x86_64. It
interleaves the computation of two SHA-256 hashes using the x86 SHA-NI
instructions. dm-verity and fs-verity will take advantage of this for
greatly improved performance on capable CPUs.
This increases the throughput of SHA-256 hashing 4096-byte messages by
the following amounts on the following CPUs:
Intel Ice Lake (server): 4%
Intel Sapphire Rapids: 38%
Intel Emerald Rapids: 38%
AMD Zen 1 (Threadripper 1950X): 84%
AMD Zen 4 (EPYC 9B14): 98%
AMD Zen 5 (Ryzen 9 9950X): 64%
For now, this seems to benefit AMD more than Intel. This seems to be
because current AMD CPUs support concurrent execution of the SHA-NI
instructions, but unfortunately current Intel CPUs don't, except for the
sha256msg2 instruction. Hopefully future Intel CPUs will support SHA-NI
on more execution ports. Zen 1 supports 2 concurrent sha256rnds2, and
Zen 4 supports 4 concurrent sha256rnds2, which suggests that even better
performance may be achievable on Zen 4 by interleaving more than two
hashes. However, doing so poses a number of trade-offs, and furthermore
Zen 5 goes back to supporting "only" 2 concurrent sha256rnds2.
Reviewed-by: Ard Biesheuvel <ardb@kernel.org>
Link: https://lore.kernel.org/r/20250915160819.140019-4-ebiggers@kernel.org
Signed-off-by: Eric Biggers <ebiggers@kernel.org>
Reorganize the Curve25519 library code:
- Build a single libcurve25519 module, instead of up to three modules:
libcurve25519, libcurve25519-generic, and an arch-specific module.
- Move the arch-specific Curve25519 code from arch/$(SRCARCH)/crypto/ to
lib/crypto/$(SRCARCH)/. Centralize the build rules into
lib/crypto/Makefile and lib/crypto/Kconfig.
- Include the arch-specific code directly in lib/crypto/curve25519.c via
a header, rather than using a separate .c file.
- Eliminate the entanglement with CRYPTO. CRYPTO_LIB_CURVE25519 no
longer selects CRYPTO, and the arch-specific Curve25519 code no longer
depends on CRYPTO.
This brings Curve25519 in line with the latest conventions for
lib/crypto/, used by other algorithms. The exception is that I kept the
generic code in separate translation units for now. (Some of the
function names collide between the x86 and generic Curve25519 code. And
the Curve25519 functions are very long anyway, so inlining doesn't
matter as much for Curve25519 as it does for some other algorithms.)
Link: https://lore.kernel.org/r/20250906213523.84915-11-ebiggers@kernel.org
Signed-off-by: Eric Biggers <ebiggers@kernel.org>
As was done with the other algorithms, reorganize the BLAKE2s code so
that the generic implementation and the arch-specific "glue" code is
consolidated into a single translation unit, so that the compiler will
inline the functions and automatically decide whether to include the
generic code in the resulting binary or not.
Similarly, also consolidate the build rules into
lib/crypto/{Makefile,Kconfig}. This removes the last uses of
lib/crypto/{arm,x86}/{Makefile,Kconfig}, so remove those too.
Don't keep the !KMSAN dependency. It was needed only for other
algorithms such as ChaCha that initialize memory from assembly code.
Reviewed-by: Ard Biesheuvel <ardb@kernel.org>
Link: https://lore.kernel.org/r/20250827151131.27733-12-ebiggers@kernel.org
Signed-off-by: Eric Biggers <ebiggers@kernel.org>
When support for a crypto algorithm is enabled, the arch-optimized
implementation of that algorithm should be enabled too. We've learned
this the hard way many times over the years: people regularly forget to
enable the arch-optimized implementations of the crypto algorithms,
resulting in significant performance being left on the table.
Currently, BLAKE2s support is always enabled ('obj-y'), since random.c
uses it. Therefore, the arch-optimized BLAKE2s code, which exists for
ARM and x86_64, should be always enabled too. Let's do that.
Note that the effect on kernel image size is very small and should not
be a concern. On ARM, enabling CRYPTO_BLAKE2S_ARM actually *shrinks*
the kernel size by about 1200 bytes, since the ARM-optimized
blake2s_compress() completely replaces the generic blake2s_compress().
On x86_64, enabling CRYPTO_BLAKE2S_X86 increases the kernel size by
about 1400 bytes, as the generic blake2s_compress() is still included as
a fallback; however, for context, that is only about a quarter the size
of the generic blake2s_compress(). The x86_64 optimized BLAKE2s code
uses much less icache at runtime than the generic code.
Reviewed-by: Ard Biesheuvel <ardb@kernel.org>
Link: https://lore.kernel.org/r/20250827151131.27733-10-ebiggers@kernel.org
Signed-off-by: Eric Biggers <ebiggers@kernel.org>
Save 480 bytes of .rodata by replacing the .long constants with .bytes,
and using the vpmovzxbd instruction to expand them.
Also update the code to do the loads before incrementing %rax rather
than after. This avoids the need for the first load to use an offset.
Reviewed-by: Ard Biesheuvel <ardb@kernel.org>
Link: https://lore.kernel.org/r/20250827151131.27733-8-ebiggers@kernel.org
Signed-off-by: Eric Biggers <ebiggers@kernel.org>
Consolidate the ChaCha code into a single module (excluding
chacha-block-generic.c which remains always built-in for random.c),
similar to various other algorithms:
- Each arch now provides a header file lib/crypto/$(SRCARCH)/chacha.h,
replacing lib/crypto/$(SRCARCH)/chacha*.c. The header defines
chacha_crypt_arch() and hchacha_block_arch(). It is included by
lib/crypto/chacha.c, and thus the code gets built into the single
libchacha module, with improved inlining in some cases.
- Whether arch-optimized ChaCha is buildable is now controlled centrally
by lib/crypto/Kconfig instead of by lib/crypto/$(SRCARCH)/Kconfig.
The conditions for enabling it remain the same as before, and it
remains enabled by default.
- Any additional arch-specific translation units for the optimized
ChaCha code, such as assembly files, are now compiled by
lib/crypto/Makefile instead of lib/crypto/$(SRCARCH)/Makefile.
This removes the last use for the Makefile and Kconfig files in the
arm64, mips, powerpc, riscv, and s390 subdirectories of lib/crypto/. So
also remove those files and the references to them.
Reviewed-by: Ard Biesheuvel <ardb@kernel.org>
Link: https://lore.kernel.org/r/20250827151131.27733-7-ebiggers@kernel.org
Signed-off-by: Eric Biggers <ebiggers@kernel.org>
Consolidate the Poly1305 code into a single module, similar to various
other algorithms (SHA-1, SHA-256, SHA-512, etc.):
- Each arch now provides a header file lib/crypto/$(SRCARCH)/poly1305.h,
replacing lib/crypto/$(SRCARCH)/poly1305*.c. The header defines
poly1305_block_init(), poly1305_blocks(), poly1305_emit(), and
optionally poly1305_mod_init_arch(). It is included by
lib/crypto/poly1305.c, and thus the code gets built into the single
libpoly1305 module, with improved inlining in some cases.
- Whether arch-optimized Poly1305 is buildable is now controlled
centrally by lib/crypto/Kconfig instead of by
lib/crypto/$(SRCARCH)/Kconfig. The conditions for enabling it remain
the same as before, and it remains enabled by default. (The PPC64 one
remains unconditionally disabled due to 'depends on BROKEN'.)
- Any additional arch-specific translation units for the optimized
Poly1305 code, such as assembly files, are now compiled by
lib/crypto/Makefile instead of lib/crypto/$(SRCARCH)/Makefile.
A special consideration is needed because the Adiantum code uses the
poly1305_core_*() functions directly. For now, just carry forward that
approach. This means retaining the CRYPTO_LIB_POLY1305_GENERIC kconfig
symbol, and keeping the poly1305_core_*() functions in separate
translation units. So it's not quite as streamlined I've done with the
other hash functions, but we still get a single libpoly1305 module.
Note: to see the diff from the arm, arm64, and x86 .c files to the new
.h files, view this commit with 'git show -M10'.
Reviewed-by: Ard Biesheuvel <ardb@kernel.org>
Link: https://lore.kernel.org/r/20250829152513.92459-3-ebiggers@kernel.org
Signed-off-by: Eric Biggers <ebiggers@kernel.org>
Eric Biggers
AlanSong-ocandEric Biggers
Linus Torvalds