162306a36Sopenharmony_ci// SPDX-License-Identifier: GPL-2.0 OR MIT 262306a36Sopenharmony_ci/* 362306a36Sopenharmony_ci * Copyright (C) 2015-2019 Jason A. Donenfeld <Jason@zx2c4.com>. All Rights Reserved. 462306a36Sopenharmony_ci */ 562306a36Sopenharmony_ci 662306a36Sopenharmony_ci#include <crypto/internal/blake2s.h> 762306a36Sopenharmony_ci 862306a36Sopenharmony_ci#include <linux/types.h> 962306a36Sopenharmony_ci#include <linux/jump_label.h> 1062306a36Sopenharmony_ci#include <linux/kernel.h> 1162306a36Sopenharmony_ci#include <linux/sizes.h> 1262306a36Sopenharmony_ci 1362306a36Sopenharmony_ci#include <asm/cpufeature.h> 1462306a36Sopenharmony_ci#include <asm/fpu/api.h> 1562306a36Sopenharmony_ci#include <asm/processor.h> 1662306a36Sopenharmony_ci#include <asm/simd.h> 1762306a36Sopenharmony_ci 1862306a36Sopenharmony_ciasmlinkage void blake2s_compress_ssse3(struct blake2s_state *state, 1962306a36Sopenharmony_ci const u8 *block, const size_t nblocks, 2062306a36Sopenharmony_ci const u32 inc); 2162306a36Sopenharmony_ciasmlinkage void blake2s_compress_avx512(struct blake2s_state *state, 2262306a36Sopenharmony_ci const u8 *block, const size_t nblocks, 2362306a36Sopenharmony_ci const u32 inc); 2462306a36Sopenharmony_ci 2562306a36Sopenharmony_cistatic __ro_after_init DEFINE_STATIC_KEY_FALSE(blake2s_use_ssse3); 2662306a36Sopenharmony_cistatic __ro_after_init DEFINE_STATIC_KEY_FALSE(blake2s_use_avx512); 2762306a36Sopenharmony_ci 2862306a36Sopenharmony_civoid blake2s_compress(struct blake2s_state *state, const u8 *block, 2962306a36Sopenharmony_ci size_t nblocks, const u32 inc) 3062306a36Sopenharmony_ci{ 3162306a36Sopenharmony_ci /* SIMD disables preemption, so relax after processing each page. */ 3262306a36Sopenharmony_ci BUILD_BUG_ON(SZ_4K / BLAKE2S_BLOCK_SIZE < 8); 3362306a36Sopenharmony_ci 3462306a36Sopenharmony_ci if (!static_branch_likely(&blake2s_use_ssse3) || !may_use_simd()) { 3562306a36Sopenharmony_ci blake2s_compress_generic(state, block, nblocks, inc); 3662306a36Sopenharmony_ci return; 3762306a36Sopenharmony_ci } 3862306a36Sopenharmony_ci 3962306a36Sopenharmony_ci do { 4062306a36Sopenharmony_ci const size_t blocks = min_t(size_t, nblocks, 4162306a36Sopenharmony_ci SZ_4K / BLAKE2S_BLOCK_SIZE); 4262306a36Sopenharmony_ci 4362306a36Sopenharmony_ci kernel_fpu_begin(); 4462306a36Sopenharmony_ci if (IS_ENABLED(CONFIG_AS_AVX512) && 4562306a36Sopenharmony_ci static_branch_likely(&blake2s_use_avx512)) 4662306a36Sopenharmony_ci blake2s_compress_avx512(state, block, blocks, inc); 4762306a36Sopenharmony_ci else 4862306a36Sopenharmony_ci blake2s_compress_ssse3(state, block, blocks, inc); 4962306a36Sopenharmony_ci kernel_fpu_end(); 5062306a36Sopenharmony_ci 5162306a36Sopenharmony_ci nblocks -= blocks; 5262306a36Sopenharmony_ci block += blocks * BLAKE2S_BLOCK_SIZE; 5362306a36Sopenharmony_ci } while (nblocks); 5462306a36Sopenharmony_ci} 5562306a36Sopenharmony_ciEXPORT_SYMBOL(blake2s_compress); 5662306a36Sopenharmony_ci 5762306a36Sopenharmony_cistatic int __init blake2s_mod_init(void) 5862306a36Sopenharmony_ci{ 5962306a36Sopenharmony_ci if (boot_cpu_has(X86_FEATURE_SSSE3)) 6062306a36Sopenharmony_ci static_branch_enable(&blake2s_use_ssse3); 6162306a36Sopenharmony_ci 6262306a36Sopenharmony_ci if (IS_ENABLED(CONFIG_AS_AVX512) && 6362306a36Sopenharmony_ci boot_cpu_has(X86_FEATURE_AVX) && 6462306a36Sopenharmony_ci boot_cpu_has(X86_FEATURE_AVX2) && 6562306a36Sopenharmony_ci boot_cpu_has(X86_FEATURE_AVX512F) && 6662306a36Sopenharmony_ci boot_cpu_has(X86_FEATURE_AVX512VL) && 6762306a36Sopenharmony_ci cpu_has_xfeatures(XFEATURE_MASK_SSE | XFEATURE_MASK_YMM | 6862306a36Sopenharmony_ci XFEATURE_MASK_AVX512, NULL)) 6962306a36Sopenharmony_ci static_branch_enable(&blake2s_use_avx512); 7062306a36Sopenharmony_ci 7162306a36Sopenharmony_ci return 0; 7262306a36Sopenharmony_ci} 7362306a36Sopenharmony_ci 7462306a36Sopenharmony_cisubsys_initcall(blake2s_mod_init); 75