1/* SPDX-License-Identifier: GPL-2.0 OR MIT */
2/*
3 * Copyright (C) 2015-2019 Jason A. Donenfeld <Jason@zx2c4.com>. All Rights Reserved.
4 */
5
6#include <asm/cpufeature.h>
7#include <asm/fpu/api.h>
8#include <asm/processor.h>
9#include <asm/simd.h>
10#include <linux/jump_label.h>
11#include <linux/kernel.h>
12#include <linux/sizes.h>
13
14asmlinkage void blake2s_compress_ssse3(struct blake2s_state *state,
15 const u8 *block, const size_t nblocks,
16 const u32 inc);
17asmlinkage void blake2s_compress_avx512(struct blake2s_state *state,
18 const u8 *block, const size_t nblocks,
19 const u32 inc);
20
21static __ro_after_init DEFINE_STATIC_KEY_FALSE(blake2s_use_ssse3);
22static __ro_after_init DEFINE_STATIC_KEY_FALSE(blake2s_use_avx512);
23
24static void blake2s_compress(struct blake2s_state *state, const u8 *block,
25 size_t nblocks, const u32 inc)
26{
27 /* SIMD disables preemption, so relax after processing each page. */
28 BUILD_BUG_ON(SZ_4K / BLAKE2S_BLOCK_SIZE < 8);
29
30 if (!static_branch_likely(&blake2s_use_ssse3) || !may_use_simd()) {
31 blake2s_compress_generic(state, block, nblocks, inc);
32 return;
33 }
34
35 do {
36 const size_t blocks = min_t(size_t, nblocks,
37 SZ_4K / BLAKE2S_BLOCK_SIZE);
38
39 kernel_fpu_begin();
40 if (static_branch_likely(&blake2s_use_avx512))
41 blake2s_compress_avx512(state, block, nblocks: blocks, inc);
42 else
43 blake2s_compress_ssse3(state, block, nblocks: blocks, inc);
44 kernel_fpu_end();
45
46 nblocks -= blocks;
47 block += blocks * BLAKE2S_BLOCK_SIZE;
48 } while (nblocks);
49}
50
51#define blake2s_mod_init_arch blake2s_mod_init_arch
52static void blake2s_mod_init_arch(void)
53{
54 if (boot_cpu_has(X86_FEATURE_SSSE3))
55 static_branch_enable(&blake2s_use_ssse3);
56
57 if (boot_cpu_has(X86_FEATURE_AVX) &&
58 boot_cpu_has(X86_FEATURE_AVX2) &&
59 boot_cpu_has(X86_FEATURE_AVX512F) &&
60 boot_cpu_has(X86_FEATURE_AVX512VL) &&
61 cpu_has_xfeatures(XFEATURE_MASK_SSE | XFEATURE_MASK_YMM |
62 XFEATURE_MASK_AVX512, NULL))
63 static_branch_enable(&blake2s_use_avx512);
64}
65