Move the arm64 implementations of SHA-3 and POLYVAL to the newly
introduced scoped ksimd API, which replaces kernel_neon_begin() and
kernel_neon_end(). On arm64, this is needed because the latter API
will change in an incompatible manner.
Signed-off-by: Ard Biesheuvel <ardb@kernel.org>
Signed-off-by: Eric Biggers <ebiggers@kernel.org>
Even though ARM's versions of kernel_neon_begin()/_end() are not being
changed, update the newly migrated ARM blake2b to the scoped ksimd API
so that all ARM and arm64 in lib/crypto remains consistent in this
manner.
Signed-off-by: Ard Biesheuvel <ardb@kernel.org>
Signed-off-by: Eric Biggers <ebiggers@kernel.org>
Introduce a more strict replacement API for
kernel_neon_begin()/kernel_neon_end() on both ARM and arm64, and replace
occurrences of the latter pair appearing in lib/crypto
-----BEGIN PGP SIGNATURE-----
iHUEABYKAB0WIQQQm/3uucuRGn1Dmh0wbglWLn0tXAUCaRRKpgAKCRAwbglWLn0t
XAKXAQD/L/XErOIGgSnvJnxG9sF+V2S+id1u9aoEJApbqMvW/gD9Fnvjqa7mRM7f
jSZeDCMB++24SS2zL0/BFiRMmEl5/gc=
=0IKE
-----END PGP SIGNATURE-----
Merge tag 'scoped-ksimd-for-arm-arm64' into libcrypto-fpsimd-on-stack
Pull scoped ksimd API for ARM and arm64 from Ard Biesheuvel:
"Introduce a more strict replacement API for
kernel_neon_begin()/kernel_neon_end() on both ARM and arm64, and
replace occurrences of the latter pair appearing in lib/crypto"
Signed-off-by: Eric Biggers <ebiggers@kernel.org>
Commit aefbab8e77
("arm64: fpsimd: Preserve/restore kernel mode NEON at context switch")
added a 'kernel_fpsimd_state' field to struct thread_struct, which is
the arch-specific portion of struct task_struct, and is allocated for
each task in the system. The size of this field is 528 bytes, resulting
in non-negligible bloat of task_struct, and the resulting memory
overhead may impact performance on systems with many processes.
This allocation is only used if the task is scheduled out or interrupted
by a softirq while using the FP/SIMD unit in kernel mode, and so it is
possible to transparently allocate this buffer on the caller's stack
instead.
So tweak the 'ksimd' scoped guard implementation so that a stack buffer
is allocated and passed to both kernel_neon_begin() and
kernel_neon_end(), and either record it in the task struct, or use it
directly to preserve the task mode kernel FP/SIMD when running in
softirq context. Passing the address to both functions, and checking the
addresses for consistency ensures that callers of the updated bare
begin/end API use it in a manner that is consistent with the new context
switch semantics.
Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Ard Biesheuvel <ardb@kernel.org>
The generic kernel mode FPU API, which is used by the AMDGPU driver to
perform floating point calculations, is modeled after the most
restrictive architecture that supports it. This means it doesn't support
preemption, and can only be used from task context.
The arm64 implementation is a bit more flexible, but supporting that in
the generic API complicates matters slightly, and for no good reason,
given that the only user does not need it.
So enforce that kernel_fpu_begin() can only be called from task context,
and [redundantly] disable preemption. This removes the need for users of
this API to provide a kernel mode FP/SIMD state after a future patch
that makes that compulsory for preemptible task context.
Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Ard Biesheuvel <ardb@kernel.org>
Instead of calling kernel_neon_begin/end directly, switch to the scoped
guard API which encapsulates those calls. This is needed because the
prototypes of those APIs are going to be modified and will require a
kernel mode FP/SIMD buffer to be provided, which the scoped guard API
will do transparently.
Cc: Saeed Mahameed <saeedm@nvidia.com>
Cc: Leon Romanovsky <leon@kernel.org>
Cc: Tariq Toukan <tariqt@nvidia.com>
Cc: Mark Bloch <mbloch@nvidia.com>
Reviewed-by: Jonathan Cameron <jonathan.cameron@huawei.com>
Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Ard Biesheuvel <ardb@kernel.org>
Switch to the more abstract 'scoped_ksimd()' API, which will be modified
in a future patch to transparently allocate a kernel mode FP/SIMD state
buffer on the stack, so that kernel mode FP/SIMD code remains
preemptible in principe, but without the memory overhead that adds 528
bytes to the size of struct task_struct.
Reviewed-by: Eric Biggers <ebiggers@kernel.org>
Reviewed-by: Jonathan Cameron <jonathan.cameron@huawei.com>
Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Ard Biesheuvel <ardb@kernel.org>
Switch to the more abstract 'scoped_ksimd()' API, which will be modified
in a future patch to transparently allocate a kernel mode FP/SIMD state
buffer on the stack, so that kernel mode FP/SIMD code remains
preemptible in principle, but without the memory overhead that adds 528
bytes to the size of struct task_struct.
Reviewed-by: Eric Biggers <ebiggers@kernel.org>
Reviewed-by: Jonathan Cameron <jonathan.cameron@huawei.com>
Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Ard Biesheuvel <ardb@kernel.org>
Switch to the more abstract 'scoped_ksimd()' API, which will be modified
in a future patch to transparently allocate a kernel mode FP/SIMD state
buffer on the stack, so that kernel mode FP/SIMD code remains
preemptible in principle, but without the memory overhead that adds 528
bytes to the size of struct task_struct.
Reviewed-by: Eric Biggers <ebiggers@kernel.org>
Reviewed-by: Jonathan Cameron <jonathan.cameron@huawei.com>
Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Ard Biesheuvel <ardb@kernel.org>
Switch to the more abstract 'scoped_ksimd()' API, which will be modified
in a future patch to transparently allocate a kernel mode FP/SIMD state
buffer on the stack, so that kernel mode FP/SIMD code remains
preemptible in principe, but without the memory overhead that adds 528
bytes to the size of struct task_struct.
Reviewed-by: Eric Biggers <ebiggers@kernel.org>
Reviewed-by: Jonathan Cameron <jonathan.cameron@huawei.com>
Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Ard Biesheuvel <ardb@kernel.org>
Switch to the more abstract 'scoped_ksimd()' API, which will be modified
in a future patch to transparently allocate a kernel mode FP/SIMD state
buffer on the stack, so that kernel mode FP/SIMD code remains
preemptible in principe, but without the memory overhead that adds 528
bytes to the size of struct task_struct.
Reviewed-by: Eric Biggers <ebiggers@kernel.org>
Reviewed-by: Jonathan Cameron <jonathan.cameron@huawei.com>
Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Ard Biesheuvel <ardb@kernel.org>
Switch to the more abstract 'scoped_ksimd()' API, which will be modified
in a future patch to transparently allocate a kernel mode FP/SIMD state
buffer on the stack, so that kernel mode FP/SIMD code remains
preemptible in principe, but without the memory overhead that adds 528
bytes to the size of struct task_struct.
Reviewed-by: Eric Biggers <ebiggers@kernel.org>
Reviewed-by: Jonathan Cameron <jonathan.cameron@huawei.com>
Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Ard Biesheuvel <ardb@kernel.org>
Switch to the more abstract 'scoped_ksimd()' API, which will be modified
in a future patch to transparently allocate a kernel mode FP/SIMD state
buffer on the stack, so that kernel mode FP/SIMD code remains
preemptible in principe, but without the memory overhead that adds 528
bytes to the size of struct task_struct.
Reviewed-by: Eric Biggers <ebiggers@kernel.org>
Reviewed-by: Jonathan Cameron <jonathan.cameron@huawei.com>
Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Ard Biesheuvel <ardb@kernel.org>
Switch to the more abstract 'scoped_ksimd()' API, which will be modified
in a future patch to transparently allocate a kernel mode FP/SIMD state
buffer on the stack, so that kernel mode FP/SIMD code remains
preemptible in principe, but without the memory overhead that adds 528
bytes to the size of struct task_struct.
Reviewed-by: Eric Biggers <ebiggers@kernel.org>
Reviewed-by: Jonathan Cameron <jonathan.cameron@huawei.com>
Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Ard Biesheuvel <ardb@kernel.org>
Switch to the more abstract 'scoped_ksimd()' API, which will be modified
in a future patch to transparently allocate a kernel mode FP/SIMD state
buffer on the stack, so that kernel mode FP/SIMD code remains
preemptible in principe, but without the memory overhead that adds 528
bytes to the size of struct task_struct.
Reviewed-by: Eric Biggers <ebiggers@kernel.org>
Reviewed-by: Jonathan Cameron <jonathan.cameron@huawei.com>
Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Ard Biesheuvel <ardb@kernel.org>
Move away from calling kernel_neon_begin() and kernel_neon_end()
directly, and instead, use the newly introduced scoped_ksimd() API. This
permits arm64 to modify the kernel mode NEON API without affecting code
that is shared between ARM and arm64.
Reviewed-by: Eric Biggers <ebiggers@kernel.org>
Reviewed-by: Jonathan Cameron <jonathan.cameron@huawei.com>
Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Ard Biesheuvel <ardb@kernel.org>
Move away from calling kernel_neon_begin() and kernel_neon_end()
directly, and instead, use the newly introduced scoped_ksimd() API. This
permits arm64 to modify the kernel mode NEON API without affecting code
that is shared between ARM and arm64.
Reviewed-by: Eric Biggers <ebiggers@kernel.org>
Reviewed-by: Jonathan Cameron <jonathan.cameron@huawei.com>
Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Ard Biesheuvel <ardb@kernel.org>
Kernel mode NEON sections are now preemptible on arm64, and so there is
no need to yield it when calling APIs that may sleep.
Also, move the calls to kernel_neon_end() to the same scope as
kernel_neon_begin(). This is needed for a subsequent change where a
stack buffer is allocated transparently and passed to
kernel_neon_begin().
While at it, simplify the logic.
Reviewed-by: Eric Biggers <ebiggers@kernel.org>
Acked-by: Herbert Xu <herbert@gondor.apana.org.au>
Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Ard Biesheuvel <ardb@kernel.org>
Kernel mode NEON sections are now preemptible on arm64, and so there is
no need to yield it when calling APIs that may sleep.
Also, move the calls to kernel_neon_end() to the same scope as
kernel_neon_begin(). This is needed for a subsequent change where a
stack buffer is allocated transparently and passed to
kernel_neon_begin().
Reviewed-by: Eric Biggers <ebiggers@kernel.org>
Acked-by: Herbert Xu <herbert@gondor.apana.org.au>
Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Ard Biesheuvel <ardb@kernel.org>
Kernel mode NEON sections are now preemptible on arm64, and so there is
no need to yield it explicitly in order to prevent scheduling latency
spikes.
Reviewed-by: Eric Biggers <ebiggers@kernel.org>
Acked-by: Herbert Xu <herbert@gondor.apana.org.au>
Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Ard Biesheuvel <ardb@kernel.org>
Before modifying the prototypes of kernel_neon_begin() and
kernel_neon_end() to accommodate kernel mode FP/SIMD state buffers
allocated on the stack, move arm64 to the new 'ksimd' scoped guard API,
which encapsulates the calls to those functions.
For symmetry, do the same for 32-bit ARM too.
Reviewed-by: Eric Biggers <ebiggers@kernel.org>
Reviewed-by: Jonathan Cameron <jonathan.cameron@huawei.com>
Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Ard Biesheuvel <ardb@kernel.org>
Before modifying the prototypes of kernel_neon_begin() and
kernel_neon_end() to accommodate kernel mode FP/SIMD state buffers
allocated on the stack, move arm64 to the new 'ksimd' scoped guard API,
which encapsulates the calls to those functions.
For symmetry, do the same for 32-bit ARM too.
Reviewed-by: Eric Biggers <ebiggers@kernel.org>
Reviewed-by: Jonathan Cameron <jonathan.cameron@huawei.com>
Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Ard Biesheuvel <ardb@kernel.org>
Implement the ksimd scoped guard API so that it can be used by code that
supports both ARM and arm64.
Reviewed-by: Kees Cook <kees@kernel.org>
Reviewed-by: Eric Biggers <ebiggers@kernel.org>
Reviewed-by: Jonathan Cameron <jonathan.cameron@huawei.com>
Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Ard Biesheuvel <ardb@kernel.org>
Encapsulate kernel_neon_begin() and kernel_neon_end() using a 'ksimd'
cleanup guard. This hides the prototype of those functions, allowing
them to be changed for arm64 but not ARM, without breaking code that is
shared between those architectures (RAID6, AEGIS-128)
It probably makes sense to expose this API more widely across
architectures, as it affords more flexibility to the arch code to
plumb it in, while imposing more rigid rules regarding the start/end
bookends appearing in matched pairs.
Reviewed-by: Kees Cook <kees@kernel.org>
Reviewed-by: Mark Brown <broonie@kernel.org>
Reviewed-by: Eric Biggers <ebiggers@kernel.org>
Reviewed-by: Jonathan Cameron <jonathan.cameron@huawei.com>
Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Ard Biesheuvel <ardb@kernel.org>
CONFIG_CRYPTO_POLYVAL_ARM64_CE and CONFIG_CRYPTO_POLYVAL_CLMUL_NI no
longer exist. The architecture-optimized POLYVAL code is now just
enabled automatically when HCTR2 support is enabled. Update the fscrypt
documentation accordingly.
Reviewed-by: Ard Biesheuvel <ardb@kernel.org>
Link: https://lore.kernel.org/r/20251109234726.638437-10-ebiggers@kernel.org
Signed-off-by: Eric Biggers <ebiggers@kernel.org>
Remove polyval support from crypto_shash. It no longer has any user now
that the HCTR2 code uses the POLYVAL library instead.
Reviewed-by: Ard Biesheuvel <ardb@kernel.org>
Link: https://lore.kernel.org/r/20251109234726.638437-8-ebiggers@kernel.org
Signed-off-by: Eric Biggers <ebiggers@kernel.org>
The "hash function" in hctr2 is fixed at POLYVAL; it can never vary.
Just use the POLYVAL library, which is much easier to use than the
crypto_shash API. It's faster, uses fixed-size structs, and never fails
(all the functions return void).
Note that this eliminates the only known user of the polyval support in
crypto_shash. A later commit will remove support for polyval from
crypto_shash, given that the library API is sufficient.
Reviewed-by: Ard Biesheuvel <ardb@kernel.org>
Link: https://lore.kernel.org/r/20251109234726.638437-7-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>
Migrate the arm64 implementation of POLYVAL into lib/crypto/, wiring it
up to the POLYVAL library interface. This makes the POLYVAL library be
properly optimized on arm64.
This drops the arm64 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.
Reviewed-by: Ard Biesheuvel <ardb@kernel.org>
Link: https://lore.kernel.org/r/20251109234726.638437-5-ebiggers@kernel.org
Signed-off-by: Eric Biggers <ebiggers@kernel.org>
Add support for POLYVAL to lib/crypto/.
This will replace the polyval crypto_shash algorithm and its use in the
hctr2 template, simplifying the code and reducing overhead.
Specifically, this commit introduces the POLYVAL library API and a
generic implementation of it. Later commits will migrate the existing
architecture-optimized implementations of POLYVAL into lib/crypto/ and
add a KUnit test suite.
I've also rewritten the generic implementation completely, using a more
modern approach instead of the traditional table-based approach. It's
now constant-time, requires no precomputation or dynamic memory
allocations, decreases the per-key memory usage from 4096 bytes to 16
bytes, and is faster than the old polyval-generic even on bulk data
reusing the same key (at least on x86_64, where I measured 15% faster).
We should do this for GHASH too, but for now just do it for POLYVAL.
Reviewed-by: Ard Biesheuvel <ardb@kernel.org>
Tested-by: Ard Biesheuvel <ardb@kernel.org>
Link: https://lore.kernel.org/r/20251109234726.638437-3-ebiggers@kernel.org
Signed-off-by: Eric Biggers <ebiggers@kernel.org>
Rename polyval_init() and polyval_update(), in preparation for adding
library functions with the same name to <crypto/polyval.h>.
Note that polyval-generic.c will be removed later, as it will be
superseded by the library. This commit just keeps the kernel building
for the initial introduction of the library.
Reviewed-by: Ard Biesheuvel <ardb@kernel.org>
Link: https://lore.kernel.org/r/20251109234726.638437-2-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>
Following the usual practice, prefix the names of the data labels with
".L" so that the assembler treats them as truly local. This more
clearly expresses the intent and is less error-prone.
Reviewed-by: Ard Biesheuvel <ardb@kernel.org>
Link: https://lore.kernel.org/r/20251102234209.62133-4-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>
Remove self-references to filenames from assembly files in
lib/crypto/arm/ and lib/crypto/arm64/. This follows the recommended
practice and eliminates an outdated reference to sha2-ce-core.S.
Reviewed-by: Ard Biesheuvel <ardb@kernel.org>
Link: https://lore.kernel.org/r/20251102014809.170713-1-ebiggers@kernel.org
Signed-off-by: Eric Biggers <ebiggers@kernel.org>
The SHA-3 library now utilizes the same s390 SHA-3 acceleration
capabilities as the arch/s390/crypto/ SHA-3 crypto_shash algorithms.
Moreover, crypto/sha3.c now uses the SHA-3 library. The result is that
all SHA-3 APIs are now s390-accelerated without any need for the old
SHA-3 code in arch/s390/crypto/. Remove this superseded code.
Also update the s390 defconfig and debug_defconfig files to enable
CONFIG_CRYPTO_SHA3 instead of CONFIG_CRYPTO_SHA3_256_S390 and
CONFIG_CRYPTO_SHA3_512_S390. This makes it so that the s390-optimized
SHA-3 continues to be built when either of these defconfigs is used.
Tested-by: Harald Freudenberger <freude@linux.ibm.com>
Reviewed-by: Ard Biesheuvel <ardb@kernel.org>
Link: https://lore.kernel.org/r/20251026055032.1413733-16-ebiggers@kernel.org
Signed-off-by: Eric Biggers <ebiggers@kernel.org>
Replace sha3_generic.c with a new file sha3.c which implements the SHA-3
crypto_shash algorithms on top of the SHA-3 library API.
Change the driver name suffix from "-generic" to "-lib" to reflect that
these algorithms now just use the (possibly arch-optimized) library.
This closely mirrors crypto/{md5,sha1,sha256,sha512,blake2b}.c.
Implement export_core and import_core, since crypto/hmac.c expects these
to be present. (Note that there is no security purpose in wrapping
SHA-3 with HMAC. HMAC was designed for older algorithms that don't
resist length extension attacks. But since someone could be using
"hmac(sha3-*)" via crypto_shash anyway, keep supporting it for now.)
Reviewed-by: Ard Biesheuvel <ardb@kernel.org>
Tested-by: Harald Freudenberger <freude@linux.ibm.com>
Link: https://lore.kernel.org/r/20251026055032.1413733-15-ebiggers@kernel.org
Signed-off-by: Eric Biggers <ebiggers@kernel.org>
Make jitterentropy use "sha3-256" instead of "sha3-256-generic", as the
ability to explicitly request the generic code is going away. It's not
worth providing a special generic API just for jitterentropy. There are
many other solutions available to it, such as doing more iterations or
using a more effective jitter collection method.
Moreover, the status quo is that SHA-3 is quite slow anyway. Currently
only arm64 and s390 have architecture-optimized SHA-3 code. I'm not
familiar with the performance of the s390 one, but the arm64 one isn't
actually that much faster than the generic code anyway.
Note that jitterentropy should just use the library API instead of
crypto_shash. But that belongs in a separate change later.
Reviewed-by: Ard Biesheuvel <ardb@kernel.org>
Tested-by: Harald Freudenberger <freude@linux.ibm.com>
Link: https://lore.kernel.org/r/20251026055032.1413733-14-ebiggers@kernel.org
Signed-off-by: Eric Biggers <ebiggers@kernel.org>
Some z/Architecture processors can compute a SHA-3 digest in a single
instruction. arch/s390/crypto/ already uses this capability to optimize
the SHA-3 crypto_shash algorithms.
Use this capability to implement the sha3_224(), sha3_256(), sha3_384(),
and sha3_512() library functions too.
SHA3-256 benchmark results provided by Harald Freudenberger
(https://lore.kernel.org/r/4188d18bfcc8a64941c5ebd8de10ede2@linux.ibm.com/)
on a z/Architecture machine with "facility 86" (MSA level 12):
Length (bytes) Before (MB/s) After (MB/s)
============== ============= ============
16 212 225
64 820 915
256 1850 3350
1024 5400 8300
4096 11200 11300
Note: the original data from Harald was given in the form of a graph for
each length, showing the distribution of throughputs from 500 runs. I
guesstimated the peak of each one.
Harald also reported that the generic SHA-3 code was at most 259 MB/s
(https://lore.kernel.org/r/c39f6b6c110def0095e5da5becc12085@linux.ibm.com/).
So as expected, the earlier commit that optimized sha3_absorb_blocks()
and sha3_keccakf() is the more important one; it optimized the Keccak
permutation which is the most performance-critical part of SHA-3.
Still, this additional commit does notably improve performance further
on some lengths.
Reviewed-by: Ard Biesheuvel <ardb@kernel.org>
Tested-by: Harald Freudenberger <freude@linux.ibm.com>
Link: https://lore.kernel.org/r/20251026055032.1413733-13-ebiggers@kernel.org
Signed-off-by: Eric Biggers <ebiggers@kernel.org>
Add support for architecture-specific overrides of sha3_224(),
sha3_256(), sha3_384(), and sha3_512(). This will be used to implement
these functions more efficiently on s390 than is possible via the usual
init + update + final flow.
Reviewed-by: Ard Biesheuvel <ardb@kernel.org>
Tested-by: Harald Freudenberger <freude@linux.ibm.com>
Link: https://lore.kernel.org/r/20251026055032.1413733-12-ebiggers@kernel.org
Signed-off-by: Eric Biggers <ebiggers@kernel.org>
Implement sha3_absorb_blocks() and sha3_keccakf() using the hardware-
accelerated SHA-3 support in Message-Security-Assist Extension 6.
This accelerates the SHA3-224, SHA3-256, SHA3-384, SHA3-512, and
SHAKE256 library functions.
Note that arch/s390/crypto/ already has SHA-3 code that uses this
extension, but it is exposed only via crypto_shash. This commit brings
the same acceleration to the SHA-3 library. The arch/s390/crypto/
version will become redundant and be removed in later changes.
Tested-by: Harald Freudenberger <freude@linux.ibm.com>
Reviewed-by: Ard Biesheuvel <ardb@kernel.org>
Link: https://lore.kernel.org/r/20251026055032.1413733-11-ebiggers@kernel.org
Signed-off-by: Eric Biggers <ebiggers@kernel.org>
Instead of exposing the arm64-optimized SHA-3 code via arm64-specific
crypto_shash algorithms, instead just implement the sha3_absorb_blocks()
and sha3_keccakf() library functions. This is much simpler, it makes
the SHA-3 library functions be arm64-optimized, and it fixes the
longstanding issue where the arm64-optimized SHA-3 code was disabled by
default. SHA-3 still remains available through crypto_shash, but
individual architectures no longer need to handle it.
Note: to see the diff from arch/arm64/crypto/sha3-ce-glue.c to
lib/crypto/arm64/sha3.h, view this commit with 'git show -M10'.
Reviewed-by: Ard Biesheuvel <ardb@kernel.org>
Link: https://lore.kernel.org/r/20251026055032.1413733-10-ebiggers@kernel.org
Signed-off-by: Eric Biggers <ebiggers@kernel.org>
- Use size_t lengths, to match the library.
- Pass the block size instead of digest size, and add support for the
block size that SHAKE128 uses. This allows the code to be used with
SHAKE128 and SHAKE256, which don't have the concept of a digest size.
SHAKE256 has the same block size as SHA3-256, but SHAKE128 has a
unique block size. Thus, there are now 5 supported block sizes.
Don't bother changing the "glue" code arm64_sha3_update() too much, as
it gets deleted when the SHA-3 code is migrated into lib/crypto/ anyway.
Reviewed-by: Ard Biesheuvel <ardb@kernel.org>
Link: https://lore.kernel.org/r/20251026055032.1413733-9-ebiggers@kernel.org
Signed-off-by: Eric Biggers <ebiggers@kernel.org>
Since the SHA-3 algorithms are FIPS-approved, add the boot-time
self-test which is apparently required. This closely follows the
corresponding SHA-1, SHA-256, and SHA-512 tests.
Tested-by: Harald Freudenberger <freude@linux.ibm.com>
Reviewed-by: Ard Biesheuvel <ardb@kernel.org>
Link: https://lore.kernel.org/r/20251026055032.1413733-8-ebiggers@kernel.org
Signed-off-by: Eric Biggers <ebiggers@kernel.org>
In crypto/sha3_generic.c, the keccakf() function calls keccakf_round()
to do four of Keccak-f's five step mappings. However, it does not do
the Iota step mapping - presumably because that is dependent on round
number, whereas Theta, Rho, Pi and Chi are not.
Note that the keccakf_round() function needs to be explicitly
non-inlined on certain architectures as gcc's produced output will (or
used to) use over 1KiB of stack space if inlined.
Now, this code was copied more or less verbatim into lib/crypto/sha3.c,
so that has the same aesthetic issue. Fix this there by passing the
round number into sha3_keccakf_one_round_generic() and doing the Iota
step mapping there.
crypto/sha3_generic.c is left untouched as that will be converted to use
lib/crypto/sha3.c at some point.
Suggested-by: Eric Biggers <ebiggers@kernel.org>
Signed-off-by: David Howells <dhowells@redhat.com>
Tested-by: Harald Freudenberger <freude@linux.ibm.com>
Reviewed-by: Ard Biesheuvel <ardb@kernel.org>
Link: https://lore.kernel.org/r/20251026055032.1413733-5-ebiggers@kernel.org
Signed-off-by: Eric Biggers <ebiggers@kernel.org>
