crypto: x86/poly1305 - Add block-only interface

Add block-only interface.

Also remove the unnecessary SIMD fallback path.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
This commit is contained in:
Herbert Xu 2025-04-28 12:56:18 +08:00
parent 14d3197914
commit 318c53ae02
2 changed files with 71 additions and 87 deletions

View file

@ -118,6 +118,19 @@ sub declare_function() {
}
}
sub declare_typed_function() {
my ($name, $align, $nargs) = @_;
if($kernel) {
$code .= "SYM_TYPED_FUNC_START($name)\n";
$code .= ".L$name:\n";
} else {
$code .= ".globl $name\n";
$code .= ".type $name,\@function,$nargs\n";
$code .= ".align $align\n";
$code .= "$name:\n";
}
}
sub end_function() {
my ($name) = @_;
if($kernel) {
@ -128,7 +141,7 @@ sub end_function() {
}
$code.=<<___ if $kernel;
#include <linux/linkage.h>
#include <linux/cfi_types.h>
___
if ($avx) {
@ -236,14 +249,14 @@ ___
$code.=<<___ if (!$kernel);
.extern OPENSSL_ia32cap_P
.globl poly1305_init_x86_64
.hidden poly1305_init_x86_64
.globl poly1305_block_init_arch
.hidden poly1305_block_init_arch
.globl poly1305_blocks_x86_64
.hidden poly1305_blocks_x86_64
.globl poly1305_emit_x86_64
.hidden poly1305_emit_x86_64
___
&declare_function("poly1305_init_x86_64", 32, 3);
&declare_typed_function("poly1305_block_init_arch", 32, 3);
$code.=<<___;
xor %eax,%eax
mov %rax,0($ctx) # initialize hash value
@ -298,7 +311,7 @@ $code.=<<___;
.Lno_key:
RET
___
&end_function("poly1305_init_x86_64");
&end_function("poly1305_block_init_arch");
&declare_function("poly1305_blocks_x86_64", 32, 4);
$code.=<<___;
@ -4105,9 +4118,9 @@ avx_handler:
.section .pdata
.align 4
.rva .LSEH_begin_poly1305_init_x86_64
.rva .LSEH_end_poly1305_init_x86_64
.rva .LSEH_info_poly1305_init_x86_64
.rva .LSEH_begin_poly1305_block_init_arch
.rva .LSEH_end_poly1305_block_init_arch
.rva .LSEH_info_poly1305_block_init_arch
.rva .LSEH_begin_poly1305_blocks_x86_64
.rva .LSEH_end_poly1305_blocks_x86_64
@ -4155,10 +4168,10 @@ ___
$code.=<<___;
.section .xdata
.align 8
.LSEH_info_poly1305_init_x86_64:
.LSEH_info_poly1305_block_init_arch:
.byte 9,0,0,0
.rva se_handler
.rva .LSEH_begin_poly1305_init_x86_64,.LSEH_begin_poly1305_init_x86_64
.rva .LSEH_begin_poly1305_block_init_arch,.LSEH_begin_poly1305_block_init_arch
.LSEH_info_poly1305_blocks_x86_64:
.byte 9,0,0,0

View file

@ -3,34 +3,15 @@
* Copyright (C) 2015-2019 Jason A. Donenfeld <Jason@zx2c4.com>. All Rights Reserved.
*/
#include <crypto/internal/simd.h>
#include <crypto/poly1305.h>
#include <asm/cpu_device_id.h>
#include <asm/fpu/api.h>
#include <crypto/internal/poly1305.h>
#include <linux/jump_label.h>
#include <linux/kernel.h>
#include <linux/module.h>
#include <linux/sizes.h>
#include <linux/string.h>
#include <linux/unaligned.h>
#include <asm/cpu_device_id.h>
#include <asm/simd.h>
asmlinkage void poly1305_init_x86_64(void *ctx,
const u8 key[POLY1305_BLOCK_SIZE]);
asmlinkage void poly1305_blocks_x86_64(void *ctx, const u8 *inp,
const size_t len, const u32 padbit);
asmlinkage void poly1305_emit_x86_64(void *ctx, u8 mac[POLY1305_DIGEST_SIZE],
const u32 nonce[4]);
asmlinkage void poly1305_emit_avx(void *ctx, u8 mac[POLY1305_DIGEST_SIZE],
const u32 nonce[4]);
asmlinkage void poly1305_blocks_avx(void *ctx, const u8 *inp, const size_t len,
const u32 padbit);
asmlinkage void poly1305_blocks_avx2(void *ctx, const u8 *inp, const size_t len,
const u32 padbit);
asmlinkage void poly1305_blocks_avx512(void *ctx, const u8 *inp,
const size_t len, const u32 padbit);
static __ro_after_init DEFINE_STATIC_KEY_FALSE(poly1305_use_avx);
static __ro_after_init DEFINE_STATIC_KEY_FALSE(poly1305_use_avx2);
static __ro_after_init DEFINE_STATIC_KEY_FALSE(poly1305_use_avx512);
struct poly1305_arch_internal {
union {
@ -45,64 +26,50 @@ struct poly1305_arch_internal {
struct { u32 r2, r1, r4, r3; } rn[9];
};
/* The AVX code uses base 2^26, while the scalar code uses base 2^64. If we hit
* the unfortunate situation of using AVX and then having to go back to scalar
* -- because the user is silly and has called the update function from two
* separate contexts -- then we need to convert back to the original base before
* proceeding. It is possible to reason that the initial reduction below is
* sufficient given the implementation invariants. However, for an avoidance of
* doubt and because this is not performance critical, we do the full reduction
* anyway. Z3 proof of below function: https://xn--4db.cc/ltPtHCKN/py
*/
static void convert_to_base2_64(void *ctx)
asmlinkage void poly1305_block_init_arch(
struct poly1305_block_state *state,
const u8 raw_key[POLY1305_BLOCK_SIZE]);
EXPORT_SYMBOL_GPL(poly1305_block_init_arch);
asmlinkage void poly1305_blocks_x86_64(struct poly1305_arch_internal *ctx,
const u8 *inp,
const size_t len, const u32 padbit);
asmlinkage void poly1305_emit_x86_64(const struct poly1305_state *ctx,
u8 mac[POLY1305_DIGEST_SIZE],
const u32 nonce[4]);
asmlinkage void poly1305_emit_avx(const struct poly1305_state *ctx,
u8 mac[POLY1305_DIGEST_SIZE],
const u32 nonce[4]);
asmlinkage void poly1305_blocks_avx(struct poly1305_arch_internal *ctx,
const u8 *inp, const size_t len,
const u32 padbit);
asmlinkage void poly1305_blocks_avx2(struct poly1305_arch_internal *ctx,
const u8 *inp, const size_t len,
const u32 padbit);
asmlinkage void poly1305_blocks_avx512(struct poly1305_arch_internal *ctx,
const u8 *inp,
const size_t len, const u32 padbit);
static __ro_after_init DEFINE_STATIC_KEY_FALSE(poly1305_use_avx);
static __ro_after_init DEFINE_STATIC_KEY_FALSE(poly1305_use_avx2);
static __ro_after_init DEFINE_STATIC_KEY_FALSE(poly1305_use_avx512);
void poly1305_blocks_arch(struct poly1305_block_state *state, const u8 *inp,
unsigned int len, u32 padbit)
{
struct poly1305_arch_internal *state = ctx;
u32 cy;
if (!state->is_base2_26)
return;
cy = state->h[0] >> 26; state->h[0] &= 0x3ffffff; state->h[1] += cy;
cy = state->h[1] >> 26; state->h[1] &= 0x3ffffff; state->h[2] += cy;
cy = state->h[2] >> 26; state->h[2] &= 0x3ffffff; state->h[3] += cy;
cy = state->h[3] >> 26; state->h[3] &= 0x3ffffff; state->h[4] += cy;
state->hs[0] = ((u64)state->h[2] << 52) | ((u64)state->h[1] << 26) | state->h[0];
state->hs[1] = ((u64)state->h[4] << 40) | ((u64)state->h[3] << 14) | (state->h[2] >> 12);
state->hs[2] = state->h[4] >> 24;
#define ULT(a, b) ((a ^ ((a ^ b) | ((a - b) ^ b))) >> (sizeof(a) * 8 - 1))
cy = (state->hs[2] >> 2) + (state->hs[2] & ~3ULL);
state->hs[2] &= 3;
state->hs[0] += cy;
state->hs[1] += (cy = ULT(state->hs[0], cy));
state->hs[2] += ULT(state->hs[1], cy);
#undef ULT
state->is_base2_26 = 0;
}
static void poly1305_simd_init(void *ctx, const u8 key[POLY1305_BLOCK_SIZE])
{
poly1305_init_x86_64(ctx, key);
}
static void poly1305_simd_blocks(void *ctx, const u8 *inp, size_t len,
const u32 padbit)
{
struct poly1305_arch_internal *state = ctx;
struct poly1305_arch_internal *ctx =
container_of(&state->h.h, struct poly1305_arch_internal, h);
/* SIMD disables preemption, so relax after processing each page. */
BUILD_BUG_ON(SZ_4K < POLY1305_BLOCK_SIZE ||
SZ_4K % POLY1305_BLOCK_SIZE);
if (!static_branch_likely(&poly1305_use_avx) ||
(len < (POLY1305_BLOCK_SIZE * 18) && !state->is_base2_26) ||
!crypto_simd_usable()) {
convert_to_base2_64(ctx);
if (!static_branch_likely(&poly1305_use_avx)) {
poly1305_blocks_x86_64(ctx, inp, len, padbit);
return;
}
do {
const size_t bytes = min_t(size_t, len, SZ_4K);
const unsigned int bytes = min(len, SZ_4K);
kernel_fpu_begin();
if (static_branch_likely(&poly1305_use_avx512))
@ -117,24 +84,26 @@ static void poly1305_simd_blocks(void *ctx, const u8 *inp, size_t len,
inp += bytes;
} while (len);
}
EXPORT_SYMBOL_GPL(poly1305_blocks_arch);
static void poly1305_simd_emit(void *ctx, u8 mac[POLY1305_DIGEST_SIZE],
const u32 nonce[4])
void poly1305_emit_arch(const struct poly1305_state *ctx,
u8 mac[POLY1305_DIGEST_SIZE], const u32 nonce[4])
{
if (!static_branch_likely(&poly1305_use_avx))
poly1305_emit_x86_64(ctx, mac, nonce);
else
poly1305_emit_avx(ctx, mac, nonce);
}
EXPORT_SYMBOL_GPL(poly1305_emit_arch);
void poly1305_init_arch(struct poly1305_desc_ctx *dctx, const u8 key[POLY1305_KEY_SIZE])
{
poly1305_simd_init(&dctx->h, key);
dctx->s[0] = get_unaligned_le32(&key[16]);
dctx->s[1] = get_unaligned_le32(&key[20]);
dctx->s[2] = get_unaligned_le32(&key[24]);
dctx->s[3] = get_unaligned_le32(&key[28]);
dctx->buflen = 0;
poly1305_block_init_arch(&dctx->state, key);
}
EXPORT_SYMBOL(poly1305_init_arch);
@ -151,14 +120,15 @@ void poly1305_update_arch(struct poly1305_desc_ctx *dctx, const u8 *src,
dctx->buflen += bytes;
if (dctx->buflen == POLY1305_BLOCK_SIZE) {
poly1305_simd_blocks(&dctx->h, dctx->buf, POLY1305_BLOCK_SIZE, 1);
poly1305_blocks_arch(&dctx->state, dctx->buf,
POLY1305_BLOCK_SIZE, 1);
dctx->buflen = 0;
}
}
if (likely(srclen >= POLY1305_BLOCK_SIZE)) {
bytes = round_down(srclen, POLY1305_BLOCK_SIZE);
poly1305_simd_blocks(&dctx->h, src, bytes, 1);
poly1305_blocks_arch(&dctx->state, src, bytes, 1);
src += bytes;
srclen -= bytes;
}
@ -176,10 +146,11 @@ void poly1305_final_arch(struct poly1305_desc_ctx *dctx, u8 *dst)
dctx->buf[dctx->buflen++] = 1;
memset(dctx->buf + dctx->buflen, 0,
POLY1305_BLOCK_SIZE - dctx->buflen);
poly1305_simd_blocks(&dctx->h, dctx->buf, POLY1305_BLOCK_SIZE, 0);
poly1305_blocks_arch(&dctx->state, dctx->buf,
POLY1305_BLOCK_SIZE, 0);
}
poly1305_simd_emit(&dctx->h, dst, dctx->s);
poly1305_emit_arch(&dctx->h, dst, dctx->s);
memzero_explicit(dctx, sizeof(*dctx));
}
EXPORT_SYMBOL(poly1305_final_arch);