crypto: Accelerated Chacha20/Poly1305 implementation
Subsystems (3)
v2 cover crypto: Accelerated Chacha20/Poly1305 implementation
This patch series provide an accelerated/optimized Chacha20 and Poly1305 implementation for Power10 or later CPU (ppc64le). This module implements algorithm specified in RFC7539. The implementation provides 3.5X better performance than the baseline for Chacha20 and Poly1305 individually and 1.5X improvement for Chacha20/Poly1305 operation.
This patch has been tested with the kernel crypto module tcrypt.ko and has passed the selftest. The patch is also tested with CONFIG_CRYPTO_MANAGER_EXTRA_TESTS enabled.
Danny Tsen (5):
An optimized Chacha20 implementation with 8-way unrolling for ppc64le. Glue code for optmized Chacha20 implementation for ppc64le. An optimized Poly1305 implementation with 4-way unrolling for ppc64le. Glue code for optmized Poly1305 implementation for ppc64le. Update Kconfig and Makefile.
arch/powerpc/crypto/Kconfig | 26 + arch/powerpc/crypto/Makefile | 4 + arch/powerpc/crypto/chacha-p10-glue.c | 221 +++++ arch/powerpc/crypto/chacha-p10le-8x.S | 842 ++++++++++++++++++ arch/powerpc/crypto/poly1305-p10-glue.c | 186 ++++ arch/powerpc/crypto/poly1305-p10le_64.S | 1075 +++++++++++++++++++++++ 6 files changed, 2354 insertions(+) create mode 100644 arch/powerpc/crypto/chacha-p10-glue.c create mode 100644 arch/powerpc/crypto/chacha-p10le-8x.S create mode 100644 arch/powerpc/crypto/poly1305-p10-glue.c create mode 100644 arch/powerpc/crypto/poly1305-p10le_64.S
Signature · 1 line
2.31.1
PATCH v2 3/5 An optimized Poly1305 implementation with 4-way unrolling for ppc64le.
Improve overall performance of Poly1305 for Power10 or later CPU.
--- arch/powerpc/crypto/poly1305-p10le_64.S | 1075 +++++++++++++++++++++++ 1 file changed, 1075 insertions(+) create mode 100644 arch/powerpc/crypto/poly1305-p10le_64.S
arch/powerpc/crypto/poly1305-p10le_64.S new file +1075 -0
@@ -0,0 +1,1075 @@ +/* SPDX-License-Identifier: GPL-2.0-or-later */ +# +# Accelerated poly1305 implementation for ppc64le. +# +# Copyright 2023- IBM Corp. All rights reserved +# +#=================================================================================== +# Written by Danny Tsen <[email protected]> +# +# Poly1305 - this version mainly using vector/VSX/Scalar +# - 26 bits limbs +# - Handle multiple 64 byte blcok. +# +# Block size 16 bytes +# key = (r, s) +# clamp r &= 0x0FFFFFFC0FFFFFFC 0x0FFFFFFC0FFFFFFF +# p = 2^130 - 5 +# a += m +# a = (r + a) % p +# a += s +# +# Improve performance by breaking down polynominal to the sum of products with +# h4 = m1 * r⁴ + m2 * r³ + m3 * r² + m4 * r +# +# 07/22/21 - this revison based on the above sum of products. Setup r^4, r^3, r^2, r and s3, s2, s1, s0 +# to 9 vectors for multiplications. +# +# setup r^4, r^3, r^2, r vectors +# vs [r^1, r^3, r^2, r^4] +# vs0 = [r0,.....] +# vs1 = [r1,.....] +# vs2 = [r2,.....] +# vs3 = [r3,.....] +# vs4 = [r4,.....] +# vs5 = [r1*5,...] +# vs6 = [r2*5,...] +# vs7 = [r2*5,...] +# vs8 = [r4*5,...] +# +# Each word in a vector consists a member of a "r/s" in [a * r/s]. +# +# r0, r4*5, r3*5, r2*5, r1*5; +# r1, r0, r4*5, r3*5, r2*5; +# r2, r1, r0, r4*5, r3*5; +# r3, r2, r1, r0, r4*5; +# r4, r3, r2, r1, r0 ; +# +# +# poly1305_p10le_4blocks( uint8_t *k, uint32_t mlen, uint8_t *m) +# k = 32 bytes key +# r3 = k (r, s) +# r4 = mlen +# r5 = m +# +#include <asm/ppc_asm.h> +#include <asm/asm-offsets.h> +#include <asm/asm-compat.h> +#include <linux/linkage.h> + +.machine "any" + +.text + +.macro SAVE_GPR GPR OFFSET FRAME + std \GPR,\OFFSET(\FRAME) +.endm + +.macro SAVE_VRS VRS OFFSET FRAME + li 16, \OFFSET + stvx \VRS, 16, \FRAME +.endm + +.macro SAVE_VSX VSX OFFSET FRAME + li 16, \OFFSET + stxvx \VSX, 16, \FRAME +.endm + +.macro RESTORE_GPR GPR OFFSET FRAME + ld \GPR,\OFFSET(\FRAME) +.endm + +.macro RESTORE_VRS VRS OFFSET FRAME + li 16, \OFFSET + lvx \VRS, 16, \FRAME +.endm + +.macro RESTORE_VSX VSX OFFSET FRAME + li 16, \OFFSET + lxvx \VSX, 16, \FRAME +.endm + +.macro SAVE_REGS + mflr 0 + std 0, 16(1) + stdu 1,-752(1) + + SAVE_GPR 14, 112, 1 + SAVE_GPR 15, 120, 1 + SAVE_GPR 16, 128, 1 + SAVE_GPR 17, 136, 1 + SAVE_GPR 18, 144, 1 + SAVE_GPR 19, 152, 1 + SAVE_GPR 20, 160, 1 + SAVE_GPR 21, 168, 1 + SAVE_GPR 22, 176, 1 + SAVE_GPR 23, 184, 1 + SAVE_GPR 24, 192, 1 + SAVE_GPR 25, 200, 1 + SAVE_GPR 26, 208, 1 + SAVE_GPR 27, 216, 1 + SAVE_GPR 28, 224, 1 + SAVE_GPR 29, 232, 1 + SAVE_GPR 30, 240, 1 + SAVE_GPR 31, 248, 1 + + addi 9, 1, 256 + SAVE_VRS 20, 0, 9 + SAVE_VRS 21, 16, 9 + SAVE_VRS 22, 32, 9 + SAVE_VRS 23, 48, 9 + SAVE_VRS 24, 64, 9 + SAVE_VRS 25, 80, 9 + SAVE_VRS 26, 96, 9 + SAVE_VRS 27, 112, 9 + SAVE_VRS 28, 128, 9 + SAVE_VRS 29, 144, 9 + SAVE_VRS 30, 160, 9 + SAVE_VRS 31, 176, 9 + + SAVE_VSX 14, 192, 9 + SAVE_VSX 15, 208, 9 + SAVE_VSX 16, 224, 9 + SAVE_VSX 17, 240, 9 + SAVE_VSX 18, 256, 9 + SAVE_VSX 19, 272, 9 + SAVE_VSX 20, 288, 9 + SAVE_VSX 21, 304, 9 + SAVE_VSX 22, 320, 9 + SAVE_VSX 23, 336, 9 + SAVE_VSX 24, 352, 9 + SAVE_VSX 25, 368, 9 + SAVE_VSX 26, 384, 9 + SAVE_VSX 27, 400, 9 + SAVE_VSX 28, 416, 9 + SAVE_VSX 29, 432, 9 + SAVE_VSX 30, 448, 9 + SAVE_VSX 31, 464, 9 +.endm # SAVE_REGS + +.macro RESTORE_REGS + addi 9, 1, 256 + RESTORE_VRS 20, 0, 9 + RESTORE_VRS 21, 16, 9 + RESTORE_VRS 22, 32, 9 + RESTORE_VRS 23, 48, 9 + RESTORE_VRS 24, 64, 9 + RESTORE_VRS 25, 80, 9 + RESTORE_VRS 26, 96, 9 + RESTORE_VRS 27, 112, 9 + RESTORE_VRS 28, 128, 9 + RESTORE_VRS 29, 144, 9 + RESTORE_VRS 30, 160, 9 + RESTORE_VRS 31, 176, 9 + + RESTORE_VSX 14, 192, 9 + RESTORE_VSX 15, 208, 9 + RESTORE_VSX 16, 224, 9 + RESTORE_VSX 17, 240, 9 + RESTORE_VSX 18, 256, 9 + RESTORE_VSX 19, 272, 9 + RESTORE_VSX 20, 288, 9 + RESTORE_VSX 21, 304, 9 + RESTORE_VSX 22, 320, 9 + RESTORE_VSX 23, 336, 9 + RESTORE_VSX 24, 352, 9 + RESTORE_VSX 25, 368, 9 + RESTORE_VSX 26, 384, 9 + RESTORE_VSX 27, 400, 9 + RESTORE_VSX 28, 416, 9 + RESTORE_VSX 29, 432, 9 + RESTORE_VSX 30, 448, 9 + RESTORE_VSX 31, 464, 9 + + RESTORE_GPR 14, 112, 1 + RESTORE_GPR 15, 120, 1 + RESTORE_GPR 16, 128, 1 + RESTORE_GPR 17, 136, 1 + RESTORE_GPR 18, 144, 1 + RESTORE_GPR 19, 152, 1 + RESTORE_GPR 20, 160, 1 + RESTORE_GPR 21, 168, 1 + RESTORE_GPR 22, 176, 1 + RESTORE_GPR 23, 184, 1 + RESTORE_GPR 24, 192, 1 + RESTORE_GPR 25, 200, 1 + RESTORE_GPR 26, 208, 1 + RESTORE_GPR 27, 216, 1 + RESTORE_GPR 28, 224, 1 + RESTORE_GPR 29, 232, 1 + RESTORE_GPR 30, 240, 1 + RESTORE_GPR 31, 248, 1 + + addi 1, 1, 752 + ld 0, 16(1) + mtlr 0 +.endm # RESTORE_REGS + +# +# p[0] = a0*r0 + a1*r4*5 + a2*r3*5 + a3*r2*5 + a4*r1*5; +# p[1] = a0*r1 + a1*r0 + a2*r4*5 + a3*r3*5 + a4*r2*5; +# p[2] = a0*r2 + a1*r1 + a2*r0 + a3*r4*5 + a4*r3*5; +# p[3] = a0*r3 + a1*r2 + a2*r1 + a3*r0 + a4*r4*5; +# p[4] = a0*r4 + a1*r3 + a2*r2 + a3*r1 + a4*r0 ; +# +# [r^2, r^3, r^1, r^4] +# [m3, m2, m4, m1] +# +# multiply odd and even words +.macro mul_odd + vmulouw 14, 4, 26 + vmulouw 10, 5, 3 + vmulouw 11, 6, 2 + vmulouw 12, 7, 1 + vmulouw 13, 8, 0 + vmulouw 15, 4, 27 + vaddudm 14, 14, 10 + vaddudm 14, 14, 11 + vmulouw 10, 5, 26 + vmulouw 11, 6, 3 + vaddudm 14, 14, 12 + vaddudm 14, 14, 13 # x0 + vaddudm 15, 15, 10 + vaddudm 15, 15, 11 + vmulouw 12, 7, 2 + vmulouw 13, 8, 1 + vaddudm 15, 15, 12 + vaddudm 15, 15, 13 # x1 + vmulouw 16, 4, 28 + vmulouw 10, 5, 27 + vmulouw 11, 6, 26 + vaddudm 16, 16, 10 + vaddudm 16, 16, 11 + vmulouw 12, 7, 3 + vmulouw 13, 8, 2 + vaddudm 16, 16, 12 + vaddudm 16, 16, 13 # x2 + vmulouw 17, 4, 29 + vmulouw 10, 5, 28 + vmulouw 11, 6, 27 + vaddudm 17, 17, 10 + vaddudm 17, 17, 11 + vmulouw 12, 7, 26 + vmulouw 13, 8, 3 + vaddudm 17, 17, 12 + vaddudm 17, 17, 13 # x3 + vmulouw 18, 4, 30 + vmulouw 10, 5, 29 + vmulouw 11, 6, 28 + vaddudm 18, 18, 10 + vaddudm 18, 18, 11 + vmulouw 12, 7, 27 + vmulouw 13, 8, 26 + vaddudm 18, 18, 12 + vaddudm 18, 18, 13 # x4 +.endm + +.macro mul_even + vmuleuw 9, 4, 26 + vmuleuw 10, 5, 3 + vmuleuw 11, 6, 2 + vmuleuw 12, 7, 1 + vmuleuw 13, 8, 0 + vaddudm 14, 14, 9 + vaddudm 14, 14, 10 + vaddudm 14, 14, 11 + vaddudm 14, 14, 12 + vaddudm 14, 14, 13 # x0 + + vmuleuw 9, 4, 27 + vmuleuw 10, 5, 26 + vmuleuw 11, 6, 3 + vmuleuw 12, 7, 2 + vmuleuw 13, 8, 1 + vaddudm 15, 15, 9 + vaddudm 15, 15, 10 + vaddudm 15, 15, 11 + vaddudm 15, 15, 12 + vaddudm 15, 15, 13 # x1 + + vmuleuw 9, 4, 28 + vmuleuw 10, 5, 27 + vmuleuw 11, 6, 26 + vmuleuw 12, 7, 3 + vmuleuw 13, 8, 2 + vaddudm 16, 16, 9 + vaddudm 16, 16, 10 + vaddudm 16, 16, 11 + vaddudm 16, 16, 12 + vaddudm 16, 16, 13 # x2 + + vmuleuw 9, 4, 29 + vmuleuw 10, 5, 28 + vmuleuw 11, 6, 27 + vmuleuw 12, 7, 26 + vmuleuw 13, 8, 3 + vaddudm 17, 17, 9 + vaddudm 17, 17, 10 + vaddudm 17, 17, 11 + vaddudm 17, 17, 12 + vaddudm 17, 17, 13 # x3 + + vmuleuw 9, 4, 30 + vmuleuw 10, 5, 29 + vmuleuw 11, 6, 28 + vmuleuw 12, 7, 27 + vmuleuw 13, 8, 26 + vaddudm 18, 18, 9 + vaddudm 18, 18, 10 + vaddudm 18, 18, 11 + vaddudm 18, 18, 12 + vaddudm 18, 18, 13 # x4 +.endm + +# +# poly1305_setup_r +# +# setup r^4, r^3, r^2, r vectors +# [r, r^3, r^2, r^4] +# vs0 = [r0,...] +# vs1 = [r1,...] +# vs2 = [r2,...] +# vs3 = [r3,...] +# vs4 = [r4,...] +# vs5 = [r4*5,...] +# vs6 = [r3*5,...] +# vs7 = [r2*5,...] +# vs8 = [r1*5,...] +# +# r0, r4*5, r3*5, r2*5, r1*5; +# r1, r0, r4*5, r3*5, r2*5; +# r2, r1, r0, r4*5, r3*5; +# r3, r2, r1, r0, r4*5; +# r4, r3, r2, r1, r0 ; +# +.macro poly1305_setup_r + + # save r + xxlor 26, 58, 58 + xxlor 27, 59, 59 + xxlor 28, 60, 60 + xxlor 29, 61, 61 + xxlor 30, 62, 62 + + xxlxor 31, 31, 31 + +# [r, r^3, r^2, r^4] + # compute r^2 + vmr 4, 26 + vmr 5, 27 + vmr 6, 28 + vmr 7, 29 + vmr 8, 30 + bl do_mul # r^2 r^1 + xxpermdi 58, 58, 36, 0x3 # r0 + xxpermdi 59, 59, 37, 0x3 # r1 + xxpermdi 60, 60, 38, 0x3 # r2 + xxpermdi 61, 61, 39, 0x3 # r3 + xxpermdi 62, 62, 40, 0x3 # r4 + xxpermdi 36, 36, 36, 0x3 + xxpermdi 37, 37, 37, 0x3 + xxpermdi 38, 38, 38, 0x3 + xxpermdi 39, 39, 39, 0x3 + xxpermdi 40, 40, 40, 0x3 + vspltisb 13, 2 + vsld 9, 27, 13 + vsld 10, 28, 13 + vsld 11, 29, 13 + vsld 12, 30, 13 + vaddudm 0, 9, 27 + vaddudm 1, 10, 28 + vaddudm 2, 11, 29 + vaddudm 3, 12, 30 + + bl do_mul # r^4 r^3 + vmrgow 26, 26, 4 + vmrgow 27, 27, 5 + vmrgow 28, 28, 6 + vmrgow 29, 29, 7 + vmrgow 30, 30, 8 + vspltisb 13, 2 + vsld 9, 27, 13 + vsld 10, 28, 13 + vsld 11, 29, 13 + vsld 12, 30, 13 + vaddudm 0, 9, 27 + vaddudm 1, 10, 28 + vaddudm 2, 11, 29 + vaddudm 3, 12, 30 + + # r^2 r^4 + xxlor 0, 58, 58 + xxlor 1, 59, 59 + xxlor 2, 60, 60 + xxlor 3, 61, 61 + xxlor 4, 62, 62 + xxlor 5, 32, 32 + xxlor 6, 33, 33 + xxlor 7, 34, 34 + xxlor 8, 35, 35 + + vspltw 9, 26, 3 + vspltw 10, 26, 2 + vmrgow 26, 10, 9 + vspltw 9, 27, 3 + vspltw 10, 27, 2 + vmrgow 27, 10, 9 + vspltw 9, 28, 3 + vspltw 10, 28, 2 + vmrgow 28, 10, 9 + vspltw 9, 29, 3 + vspltw 10, 29, 2 + vmrgow 29, 10, 9 + vspltw 9, 30, 3 + vspltw 10, 30, 2 + vmrgow 30, 10, 9 + + vsld 9, 27, 13 + vsld 10, 28, 13 + vsld 11, 29, 13 + vsld 12, 30, 13 + vaddudm 0, 9, 27 + vaddudm 1, 10, 28 + vaddudm 2, 11, 29 + vaddudm 3, 12, 30 +.endm + +SYM_FUNC_START_LOCAL(do_mul) + mul_odd + + # do reduction ( h %= p ) + # carry reduction + vspltisb 9, 2 + vsrd 10, 14, 31 + vsrd 11, 17, 31 + vand 7, 17, 25 + vand 4, 14, 25 + vaddudm 18, 18, 11 + vsrd 12, 18, 31 + vaddudm 15, 15, 10 + + vsrd 11, 15, 31 + vand 8, 18, 25 + vand 5, 15, 25 + vaddudm 4, 4, 12 + vsld 10, 12, 9 + vaddudm 6, 16, 11 + + vsrd 13, 6, 31 + vand 6, 6, 25 + vaddudm 4, 4, 10 + vsrd 10, 4, 31 + vaddudm 7, 7, 13 + + vsrd 11, 7, 31 + vand 7, 7, 25 + vand 4, 4, 25 + vaddudm 5, 5, 10 + vaddudm 8, 8, 11 + blr +SYM_FUNC_END(do_mul) + +# +# init key +# +.macro do_poly1305_init + addis 10, 2, rmask@toc@ha + addi 10, 10, rmask@toc@l + + ld 11, 0(10) + ld 12, 8(10) + + li 14, 16 + li 15, 32 + addis 10, 2, cnum@toc@ha + addi 10, 10, cnum@toc@l + lvx 25, 0, 10 # v25 - mask + lvx 31, 14, 10 # v31 = 1a + lvx 19, 15, 10 # v19 = 1 << 24 + lxv 24, 48(10) # vs24 + lxv 25, 64(10) # vs25 + + # initialize + # load key from r3 to vectors + ld 9, 24(3) + ld 10, 32(3) + and. 9, 9, 11 + and. 10, 10, 12 + + # break 26 bits + extrdi 14, 9, 26, 38 + extrdi 15, 9, 26, 12 + extrdi 16, 9, 12, 0 + mtvsrdd 58, 0, 14 + insrdi 16, 10, 14, 38 + mtvsrdd 59, 0, 15 + extrdi 17, 10, 26, 24 + mtvsrdd 60, 0, 16 + extrdi 18, 10, 24, 0 + mtvsrdd 61, 0, 17 + mtvsrdd 62, 0, 18 + + # r1 = r1 * 5, r2 = r2 * 5, r3 = r3 * 5, r4 = r4 * 5 + li 9, 5 + mtvsrdd 36, 0, 9 + vmulouw 0, 27, 4 # v0 = rr0 + vmulouw 1, 28, 4 # v1 = rr1 + vmulouw 2, 29, 4 # v2 = rr2 + vmulouw 3, 30, 4 # v3 = rr3 +.endm + +# +# poly1305_p10le_4blocks( uint8_t *k, uint32_t mlen, uint8_t *m) +# k = 32 bytes key +# r3 = k (r, s) +# r4 = mlen +# r5 = m +# +SYM_FUNC_START(poly1305_p10le_4blocks) +.align 5 + cmpdi 5, 64 + blt Out_no_poly1305 + + SAVE_REGS + + do_poly1305_init + + li 21, 0 # counter to message + + poly1305_setup_r + + # load previous H state + # break/convert r6 to 26 bits + ld 9, 0(3) + ld 10, 8(3) + ld 19, 16(3) + sldi 19, 19, 24 + mtvsrdd 41, 0, 19 + extrdi 14, 9, 26, 38 + extrdi 15, 9, 26, 12 + extrdi 16, 9, 12, 0 + mtvsrdd 36, 0, 14 + insrdi 16, 10, 14, 38 + mtvsrdd 37, 0, 15 + extrdi 17, 10, 26, 24 + mtvsrdd 38, 0, 16 + extrdi 18, 10, 24, 0 + mtvsrdd 39, 0, 17 + mtvsrdd 40, 0, 18 + vor 8, 8, 9 + + # input m1 m2 + add 20, 4, 21 + xxlor 49, 24, 24 + xxlor 50, 25, 25 + lxvw4x 43, 0, 20 + addi 17, 20, 16 + lxvw4x 44, 0, 17 + vperm 14, 11, 12, 17 + vperm 15, 11, 12, 18 + vand 9, 14, 25 # a0 + vsrd 10, 14, 31 # >> 26 + vsrd 11, 10, 31 # 12 bits left + vand 10, 10, 25 # a1 + vspltisb 13, 12 + vand 16, 15, 25 + vsld 12, 16, 13 + vor 11, 11, 12 + vand 11, 11, 25 # a2 + vspltisb 13, 14 + vsrd 12, 15, 13 # >> 14 + vsrd 13, 12, 31 # >> 26, a4 + vand 12, 12, 25 # a3 + + vaddudm 20, 4, 9 + vaddudm 21, 5, 10 + vaddudm 22, 6, 11 + vaddudm 23, 7, 12 + vaddudm 24, 8, 13 + + # m3 m4 + addi 17, 17, 16 + lxvw4x 43, 0, 17 + addi 17, 17, 16 + lxvw4x 44, 0, 17 + vperm 14, 11, 12, 17 + vperm 15, 11, 12, 18 + vand 9, 14, 25 # a0 + vsrd 10, 14, 31 # >> 26 + vsrd 11, 10, 31 # 12 bits left + vand 10, 10, 25 # a1 + vspltisb 13, 12 + vand 16, 15, 25 + vsld 12, 16, 13 + vspltisb 13, 14 + vor 11, 11, 12 + vand 11, 11, 25 # a2 + vsrd 12, 15, 13 # >> 14 + vsrd 13, 12, 31 # >> 26, a4 + vand 12, 12, 25 # a3 + + # Smash 4 message blocks into 5 vectors of [m4, m2, m3, m1] + vmrgow 4, 9, 20 + vmrgow 5, 10, 21 + vmrgow 6, 11, 22 + vmrgow 7, 12, 23 + vmrgow 8, 13, 24 + vaddudm 8, 8, 19 + + addi 5, 5, -64 # len -= 64 + addi 21, 21, 64 # offset += 64 + + li 9, 64 + divdu 31, 5, 9 + + cmpdi 31, 0 + ble Skip_block_loop + + mtctr 31 + +# h4 = m1 * r⁴ + m2 * r³ + m3 * r² + m4 * r +# Rewrite the polynominal sum of product as follows, +# h1 = (h0 + m1) * r^2, h2 = (h0 + m2) * r^2 +# h3 = (h1 + m3) * r^2, h4 = (h2 + m4) * r^2 --> (h0 + m1) r*4 + (h3 + m3) r^2, (h0 + m2) r^4 + (h0 + m4) r^2 +# .... Repeat +# h5 = (h3 + m5) * r^2, h6 = (h4 + m6) * r^2 --> +# h7 = (h5 + m7) * r^2, h8 = (h6 + m8) * r^1 --> m5 * r^4 + m6 * r^3 + m7 * r^2 + m8 * r +# +loop_4blocks: + + # Multiply odd words and even words + mul_odd + mul_even + # carry reduction + vspltisb 9, 2 + vsrd 10, 14, 31 + vsrd 11, 17, 31 + vand 7, 17, 25 + vand 4, 14, 25 + vaddudm 18, 18, 11 + vsrd 12, 18, 31 + vaddudm 15, 15, 10 + + vsrd 11, 15, 31 + vand 8, 18, 25 + vand 5, 15, 25 + vaddudm 4, 4, 12 + vsld 10, 12, 9 + vaddudm 6, 16, 11 + + vsrd 13, 6, 31 + vand 6, 6, 25 + vaddudm 4, 4, 10 + vsrd 10, 4, 31 + vaddudm 7, 7, 13 + + vsrd 11, 7, 31 + vand 7, 7, 25 + vand 4, 4, 25 + vaddudm 5, 5, 10 + vaddudm 8, 8, 11 + + # input m1 m2 m3 m4 + add 20, 4, 21 + xxlor 49, 24, 24 + xxlor 50, 25, 25 + lxvw4x 43, 0, 20 + addi 17, 20, 16 + lxvw4x 44, 0, 17 + vperm 14, 11, 12, 17 + vperm 15, 11, 12, 18 + addi 17, 17, 16 + lxvw4x 43, 0, 17 + addi 17, 17, 16 + lxvw4x 44, 0, 17 + vperm 17, 11, 12, 17 + vperm 18, 11, 12, 18 + + vand 20, 14, 25 # a0 + vand 9, 17, 25 # a0 + vsrd 21, 14, 31 # >> 26 + vsrd 22, 21, 31 # 12 bits left + vsrd 10, 17, 31 # >> 26 + vsrd 11, 10, 31 # 12 bits left + + vand 21, 21, 25 # a1 + vand 10, 10, 25 # a1 + + vspltisb 13, 12 + vand 16, 15, 25 + vsld 23, 16, 13 + vor 22, 22, 23 + vand 22, 22, 25 # a2 + vand 16, 18, 25 + vsld 12, 16, 13 + vor 11, 11, 12 + vand 11, 11, 25 # a2 + vspltisb 13, 14 + vsrd 23, 15, 13 # >> 14 + vsrd 24, 23, 31 # >> 26, a4 + vand 23, 23, 25 # a3 + vsrd 12, 18, 13 # >> 14 + vsrd 13, 12, 31 # >> 26, a4 + vand 12, 12, 25 # a3 + + vaddudm 4, 4, 20 + vaddudm 5, 5, 21 + vaddudm 6, 6, 22 + vaddudm 7, 7, 23 + vaddudm 8, 8, 24 + + # Smash 4 message blocks into 5 vectors of [m4, m2, m3, m1] + vmrgow 4, 9, 4 + vmrgow 5, 10, 5 + vmrgow 6, 11, 6 + vmrgow 7, 12, 7 + vmrgow 8, 13, 8 + vaddudm 8, 8, 19 + + addi 5, 5, -64 # len -= 64 + addi 21, 21, 64 # offset += 64 + + bdnz loop_4blocks + +Skip_block_loop: + xxlor 58, 0, 0 + xxlor 59, 1, 1 + xxlor 60, 2, 2 + xxlor 61, 3, 3 + xxlor 62, 4, 4 + xxlor 32, 5, 5 + xxlor 33, 6, 6 + xxlor 34, 7, 7 + xxlor 35, 8, 8 + + # Multiply odd words and even words + mul_odd + mul_even + + # Sum the products. + xxpermdi 41, 31, 46, 0 + xxpermdi 42, 31, 47, 0 + vaddudm 4, 14, 9 + xxpermdi 36, 31, 36, 3 + vaddudm 5, 15, 10 + xxpermdi 37, 31, 37, 3 + xxpermdi 43, 31, 48, 0 + vaddudm 6, 16, 11 + xxpermdi 38, 31, 38, 3 + xxpermdi 44, 31, 49, 0 + vaddudm 7, 17, 12 + xxpermdi 39, 31, 39, 3 + xxpermdi 45, 31, 50, 0 + vaddudm 8, 18, 13 + xxpermdi 40, 31, 40, 3 + + # carry reduction + vspltisb 9, 2 + vsrd 10, 4, 31 + vsrd 11, 7, 31 + vand 7, 7, 25 + vand 4, 4, 25 + vaddudm 8, 8, 11 + vsrd 12, 8, 31 + vaddudm 5, 5, 10 + + vsrd 11, 5, 31 + vand 8, 8, 25 + vand 5, 5, 25 + vaddudm 4, 4, 12 + vsld 10, 12, 9 + vaddudm 6, 6, 11 + + vsrd 13, 6, 31 + vand 6, 6, 25 + vaddudm 4, 4, 10 + vsrd 10, 4, 31 + vaddudm 7, 7, 13 + + vsrd 11, 7, 31 + vand 7, 7, 25 + vand 4, 4, 25 + vaddudm 5, 5, 10 + vsrd 10, 5, 31 + vand 5, 5, 25 + vaddudm 6, 6, 10 + vaddudm 8, 8, 11 + + b do_final_update + +do_final_update: + # combine 26 bit limbs + # v4, v5, v6, v7 and v8 are 26 bit vectors + vsld 5, 5, 31 + vor 20, 4, 5 + vspltisb 11, 12 + vsrd 12, 6, 11 + vsld 6, 6, 31 + vsld 6, 6, 31 + vor 20, 20, 6 + vspltisb 11, 14 + vsld 7, 7, 11 + vor 21, 7, 12 + mfvsrld 16, 40 # save last 2 bytes + vsld 8, 8, 11 + vsld 8, 8, 31 + vor 21, 21, 8 + mfvsrld 17, 52 + mfvsrld 19, 53 + srdi 16, 16, 24 + + std 17, 0(3) + std 19, 8(3) + stw 16, 16(3) + +Out_loop: + li 3, 0 + + RESTORE_REGS + + blr + +Out_no_poly1305: + li 3, 0 + blr +SYM_FUNC_END(poly1305_p10le_4blocks) + +# +# ======================================================================= +# The following functions implement 64 x 64 bits multiplication poly1305. +# +SYM_FUNC_START_LOCAL(Poly1305_init_64) + # mask 0x0FFFFFFC0FFFFFFC + # mask 0x0FFFFFFC0FFFFFFF + addis 10, 2, rmask@toc@ha + addi 10, 10, rmask@toc@l + ld 11, 0(10) + ld 12, 8(10) + + # initialize + # load key from r3 + ld 9, 24(3) + ld 10, 32(3) + and. 9, 9, 11 # cramp mask r0 + and. 10, 10, 12 # cramp mask r1 + + srdi 21, 10, 2 + add 19, 21, 10 # s1: r19 - (r1 >> 2) *5 + + # setup r and s + li 25, 0 + mtvsrdd 32+0, 9, 19 # r0, s1 + mtvsrdd 32+1, 10, 9 # r1, r0 + mtvsrdd 32+2, 19, 25 # s1 + mtvsrdd 32+3, 9, 25 # r0 + + blr +SYM_FUNC_END(Poly1305_init_64) + +# Poly1305_mult +# v6 = (h0, h1), v8 = h2 +# v0 = (r0, s1), v1 = (r1, r0), v2 = s1, v3 = r0 +# +# Output: v7, v10, v11 +# +SYM_FUNC_START_LOCAL(Poly1305_mult) + # + # d0 = h0 * r0 + h1 * s1 + vmsumudm 7, 6, 0, 9 # h0 * r0, h1 * s1 + + # d1 = h0 * r1 + h1 * r0 + h2 * s1 + vmsumudm 11, 6, 1, 9 # h0 * r1, h1 * r0 + vmsumudm 10, 8, 2, 11 # d1 += h2 * s1 + + # d2 = r0 + vmsumudm 11, 8, 3, 9 # d2 = h2 * r0 + blr +SYM_FUNC_END(Poly1305_mult) + +# +# carry reduction +# h %=p +# +# Input: v7, v10, v11 +# Output: r27, r28, r29 +# +SYM_FUNC_START_LOCAL(Carry_reduction) + mfvsrld 27, 32+7 + mfvsrld 28, 32+10 + mfvsrld 29, 32+11 + mfvsrd 20, 32+7 # h0.h + mfvsrd 21, 32+10 # h1.h + + addc 28, 28, 20 + adde 29, 29, 21 + srdi 22, 29, 0x2 + sldi 23, 22, 0x2 + add 23, 23, 22 # (h2 & 3) * 5 + addc 27, 27, 23 # h0 + addze 28, 28 # h1 + andi. 29, 29, 0x3 # h2 + blr +SYM_FUNC_END(Carry_reduction) + +# +# poly1305 multiplication +# h *= r, h %= p +# d0 = h0 * r0 + h1 * s1 +# d1 = h0 * r1 + h1 * r0 + h2 * s1 +# d2 = h0 * r0 +# +# +# unsigned int poly1305_test_64s(unisgned char *state, const byte *src, size_t len, highbit) +# - no highbit if final leftover block (highbit = 0) +# +SYM_FUNC_START(poly1305_64s) + cmpdi 5, 0 + ble Out_no_poly1305_64 + + mflr 0 + std 0, 16(1) + stdu 1,-400(1) + + SAVE_GPR 14, 112, 1 + SAVE_GPR 15, 120, 1 + SAVE_GPR 16, 128, 1 + SAVE_GPR 17, 136, 1 + SAVE_GPR 18, 144, 1 + SAVE_GPR 19, 152, 1 + SAVE_GPR 20, 160, 1 + SAVE_GPR 21, 168, 1 + SAVE_GPR 22, 176, 1 + SAVE_GPR 23, 184, 1 + SAVE_GPR 24, 192, 1 + SAVE_GPR 25, 200, 1 + SAVE_GPR 26, 208, 1 + SAVE_GPR 27, 216, 1 + SAVE_GPR 28, 224, 1 + SAVE_GPR 29, 232, 1 + SAVE_GPR 30, 240, 1 + SAVE_GPR 31, 248, 1 + + # Init poly1305 + bl Poly1305_init_64 + + li 25, 0 # offset to inp and outp + + add 11, 25, 4 + + # load h + # h0, h1, h2? + ld 27, 0(3) + ld 28, 8(3) + lwz 29, 16(3) + + li 30, 16 + divdu 31, 5, 30 + + mtctr 31 + + mr 24, 6 # highbit + +Loop_block_64: + vxor 9, 9, 9 + + ld 20, 0(11) + ld 21, 8(11) + addi 11, 11, 16 + + addc 27, 27, 20 + adde 28, 28, 21 + adde 29, 29, 24 + + li 22, 0 + mtvsrdd 32+6, 27, 28 # h0, h1 + mtvsrdd 32+8, 29, 22 # h2 + + bl Poly1305_mult + + bl Carry_reduction + + bdnz Loop_block_64 + + std 27, 0(3) + std 28, 8(3) + stw 29, 16(3) + + li 3, 0 + + RESTORE_GPR 14, 112, 1 + RESTORE_GPR 15, 120, 1 + RESTORE_GPR 16, 128, 1 + RESTORE_GPR 17, 136, 1 + RESTORE_GPR 18, 144, 1 + RESTORE_GPR 19, 152, 1 + RESTORE_GPR 20, 160, 1 + RESTORE_GPR 21, 168, 1 + RESTORE_GPR 22, 176, 1 + RESTORE_GPR 23, 184, 1 + RESTORE_GPR 24, 192, 1 + RESTORE_GPR 25, 200, 1 + RESTORE_GPR 26, 208, 1 + RESTORE_GPR 27, 216, 1 + RESTORE_GPR 28, 224, 1 + RESTORE_GPR 29, 232, 1 + RESTORE_GPR 30, 240, 1 + RESTORE_GPR 31, 248, 1 + + addi 1, 1, 400 + ld 0, 16(1) + mtlr 0 + + blr + +Out_no_poly1305_64: + li 3, 0 + blr +SYM_FUNC_END(poly1305_64s) + +# +# Input: r3 = h, r4 = s, r5 = mac +# mac = h + s +# +SYM_FUNC_START(poly1305_emit_64) + ld 10, 0(3) + ld 11, 8(3) + ld 12, 16(3) + + # compare modulus + # h + 5 + (-p) + mr 6, 10 + mr 7, 11 + mr 8, 12 + addic. 6, 6, 5 + addze 7, 7 + addze 8, 8 + srdi 9, 8, 2 # overflow? + cmpdi 9, 0 + beq Skip_h64 + mr 10, 6 + mr 11, 7 + mr 12, 8 + +Skip_h64: + ld 6, 0(4) + ld 7, 8(4) + addc 10, 10, 6 + adde 11, 11, 7 + addze 12, 12 + + std 10, 0(5) + std 11, 8(5) + blr +SYM_FUNC_END(poly1305_emit_64) + +SYM_DATA_START_LOCAL(RMASK) +.align 5 +rmask: +.byte 0xff, 0xff, 0xff, 0x0f, 0xfc, 0xff, 0xff, 0x0f, 0xfc, 0xff, 0xff, 0x0f, 0xfc, 0xff, 0xff, 0x0f +cnum: +.long 0x03ffffff, 0x00000000, 0x03ffffff, 0x00000000 +.long 0x1a, 0x00, 0x1a, 0x00 +.long 0x01000000, 0x01000000, 0x01000000, 0x01000000 +.long 0x00010203, 0x04050607, 0x10111213, 0x14151617 +.long 0x08090a0b, 0x0c0d0e0f, 0x18191a1b, 0x1c1d1e1f +SYM_DATA_END(RMASK)
Signature · 1 line
2.31.1
PATCH v2 4/5 Glue code for optmized Poly1305 implementation for ppc64le.
In reply to Danny Tsen--- arch/powerpc/crypto/poly1305-p10-glue.c | 186 ++++++++++++++++++++++++ 1 file changed, 186 insertions(+) create mode 100644 arch/powerpc/crypto/poly1305-p10-glue.c
arch/powerpc/crypto/poly1305-p10-glue.c new file +186 -0
@@ -0,0 +1,186 @@ +// SPDX-License-Identifier: GPL-2.0 +/* +* Poly1305 authenticator algorithm, RFC7539. +* +* Copyright 2023- IBM Corp. All rights reserved. +*/ + +#include <crypto/algapi.h> +#include <linux/crypto.h> +#include <linux/kernel.h> +#include <linux/module.h> +#include <linux/jump_label.h> +#include <crypto/internal/hash.h> +#include <crypto/internal/poly1305.h> +#include <crypto/internal/simd.h> +#include <linux/cpufeature.h> +#include <asm/unaligned.h> +#include <asm/simd.h> +#include <asm/switch_to.h> + +asmlinkage void poly1305_p10le_4blocks(void *h, const u8 *m, u32 mlen); +asmlinkage void poly1305_64s(void *h, const u8 *m, u32 mlen, int highbit); +asmlinkage void poly1305_emit_64(void *h, void *s, u8 *dst); + +static void vsx_begin (void ) +{ +preempt_disable (); +enable_kernel_vsx (); +} + +static void vsx_end (void ) +{ +disable_kernel_vsx (); +preempt_enable (); +} + +static int crypto_poly1305_p10_init (struct shash_desc *desc) +{ +struct poly1305_desc_ctx *dctx =shash_desc_ctx (desc); + +poly1305_core_init (&dctx->h); + dctx->buflen =0 ; + dctx->rset =0 ; + dctx->sset = false; + +return 0 ; +} + +static unsigned int crypto_poly1305_setdctxkey (struct poly1305_desc_ctx *dctx, +const u8 *inp,unsigned int len) +{ +unsigned int acc =0 ; + +if (unlikely (!dctx->sset)) { +if (!dctx->rset && len >=POLY1305_BLOCK_SIZE ) { +struct poly1305_core_key *key = &dctx->core_r; + + key->key.r64[0 ] =get_unaligned_le64 (&inp[0 ]); + key->key.r64[1 ] =get_unaligned_le64 (&inp[8 ]); + inp +=POLY1305_BLOCK_SIZE ; + len -=POLY1305_BLOCK_SIZE ; + acc +=POLY1305_BLOCK_SIZE ; + dctx->rset =1 ; + } +if (len >=POLY1305_BLOCK_SIZE ) { + dctx->s[0 ] =get_unaligned_le32 (&inp[0 ]); + dctx->s[1 ] =get_unaligned_le32 (&inp[4 ]); + dctx->s[2 ] =get_unaligned_le32 (&inp[8 ]); + dctx->s[3 ] =get_unaligned_le32 (&inp[12 ]); + acc +=POLY1305_BLOCK_SIZE ; + dctx->sset = true; + } + } +return acc; +} + +static int crypto_poly1305_p10_update (struct shash_desc *desc, +const u8 *src,unsigned int srclen) +{ +struct poly1305_desc_ctx *dctx =shash_desc_ctx (desc); +unsigned int bytes, used; + +if (unlikely (dctx->buflen)) { + bytes =min (srclen,POLY1305_BLOCK_SIZE - dctx->buflen); +memcpy (dctx->buf + dctx->buflen, src, bytes); + src += bytes; + srclen -= bytes; + dctx->buflen += bytes; + +if (dctx->buflen ==POLY1305_BLOCK_SIZE ) { +if (likely (!crypto_poly1305_setdctxkey (dctx, dctx->buf, +POLY1305_BLOCK_SIZE ))) { +vsx_begin (); +poly1305_64s (&dctx->h, dctx->buf, +POLY1305_BLOCK_SIZE ,1 ); +vsx_end (); + } + dctx->buflen =0 ; + } + } + +if (likely (srclen >=POLY1305_BLOCK_SIZE )) { + bytes =round_down (srclen,POLY1305_BLOCK_SIZE ); + used =crypto_poly1305_setdctxkey (dctx, src, bytes); +if (likely (used)) { + srclen -= used; + src += used; + } +if (crypto_simd_usable () && (srclen >=POLY1305_BLOCK_SIZE *4 )) { +vsx_begin (); +poly1305_p10le_4blocks (&dctx->h, src, srclen); +vsx_end (); + src += srclen - (srclen % (POLY1305_BLOCK_SIZE *4 )); + srclen %=POLY1305_BLOCK_SIZE *4 ; + } +while (srclen >=POLY1305_BLOCK_SIZE ) { +vsx_begin (); +poly1305_64s (&dctx->h, src,POLY1305_BLOCK_SIZE ,1 ); +vsx_end (); + srclen -=POLY1305_BLOCK_SIZE ; + src +=POLY1305_BLOCK_SIZE ; + } + } + +if (unlikely (srclen)) { + dctx->buflen = srclen; +memcpy (dctx->buf, src, srclen); + } + +return 0 ; +} + +static int crypto_poly1305_p10_final (struct shash_desc *desc,u8 *dst) +{ +struct poly1305_desc_ctx *dctx =shash_desc_ctx (desc); + +if (unlikely (!dctx->sset)) +return -ENOKEY ; + +if ((dctx->buflen)) { + dctx->buf[dctx->buflen++] =1 ; +memset (dctx->buf + dctx->buflen,0 , +POLY1305_BLOCK_SIZE - dctx->buflen); +vsx_begin (); +poly1305_64s (&dctx->h, dctx->buf,POLY1305_BLOCK_SIZE ,0 ); +vsx_end (); + dctx->buflen =0 ; + } + +poly1305_emit_64 (&dctx->h, &dctx->s, dst); +return 0 ; +} + +static struct shash_alg poly1305_alg = { + .digestsize =POLY1305_DIGEST_SIZE , + .init = crypto_poly1305_p10_init, + .update = crypto_poly1305_p10_update, + .final = crypto_poly1305_p10_final, + .descsize =sizeof (struct poly1305_desc_ctx ), + .base = { + .cra_name ="poly1305" , + .cra_driver_name ="poly1305-p10" , + .cra_priority =300 , + .cra_blocksize =POLY1305_BLOCK_SIZE , + .cra_module =THIS_MODULE , + }, +}; + +static int __init poly1305_p10_init (void ) +{ +return crypto_register_shash (&poly1305_alg); +} + +static void __exit poly1305_p10_exit(void) +{ +crypto_unregister_shash (&poly1305_alg); +} + +module_cpu_feature_match (PPC_MODULE_FEATURE_P10 , poly1305_p10_init); +module_exit (poly1305_p10_exit); + +MODULE_LICENSE ("GPL" ); +MODULE_AUTHOR ("Danny Tsen <[email protected]>" ); +MODULE_DESCRIPTION ("Optimized Poly1305 for P10" ); +MODULE_ALIAS_CRYPTO ("poly1305" ); +MODULE_ALIAS_CRYPTO ("poly1305-p10" );
Signature · 1 line
2.31.1
PATCH v2 1/5 An optimized Chacha20 implementation with 8-way unrolling for ppc64le.
In reply to Danny TsenImprove overall performance of chacha20 encrypt and decrypt operations for Power10 or later CPU.
--- arch/powerpc/crypto/chacha-p10le-8x.S | 842 ++++++++++++++++++++++++++ 1 file changed, 842 insertions(+) create mode 100644 arch/powerpc/crypto/chacha-p10le-8x.S
arch/powerpc/crypto/chacha-p10le-8x.S new file +842 -0
@@ -0,0 +1,842 @@ +/* SPDX-License-Identifier: GPL-2.0-or-later */ +# +# Accelerated chacha20 implementation for ppc64le. +# +# Copyright 2023- IBM Corp. All rights reserved +# +#=================================================================================== +# Written by Danny Tsen <[email protected]> +# +# chacha_p10le_8x(u32 *state, byte *dst, const byte *src, +# size_t len, int nrounds); +# +# do rounds, 8 quarter rounds +# 1. a += b; d ^= a; d <<<= 16; +# 2. c += d; b ^= c; b <<<= 12; +# 3. a += b; d ^= a; d <<<= 8; +# 4. c += d; b ^= c; b <<<= 7 +# +# row1 = (row1 + row2), row4 = row1 xor row4, row4 rotate each word by 16 +# row3 = (row3 + row4), row2 = row3 xor row2, row2 rotate each word by 12 +# row1 = (row1 + row2), row4 = row1 xor row4, row4 rotate each word by 8 +# row3 = (row3 + row4), row2 = row3 xor row2, row2 rotate each word by 7 +# +# 4 blocks (a b c d) +# +# a0 b0 c0 d0 +# a1 b1 c1 d1 +# ... +# a4 b4 c4 d4 +# ... +# a8 b8 c8 d8 +# ... +# a12 b12 c12 d12 +# a13 ... +# a14 ... +# a15 b15 c15 d15 +# +# Column round (v0, v4, v8, v12, v1, v5, v9, v13, v2, v6, v10, v14, v3, v7, v11, v15) +# Diagnal round (v0, v5, v10, v15, v1, v6, v11, v12, v2, v7, v8, v13, v3, v4, v9, v14) +# + +#include <asm/ppc_asm.h> +#include <asm/asm-offsets.h> +#include <asm/asm-compat.h> +#include <linux/linkage.h> + +.machine "any" +.text + +.macro SAVE_GPR GPR OFFSET FRAME + std \GPR,\OFFSET(\FRAME) +.endm + +.macro SAVE_VRS VRS OFFSET FRAME + li 16, \OFFSET + stvx \VRS, 16, \FRAME +.endm + +.macro SAVE_VSX VSX OFFSET FRAME + li 16, \OFFSET + stxvx \VSX, 16, \FRAME +.endm + +.macro RESTORE_GPR GPR OFFSET FRAME + ld \GPR,\OFFSET(\FRAME) +.endm + +.macro RESTORE_VRS VRS OFFSET FRAME + li 16, \OFFSET + lvx \VRS, 16, \FRAME +.endm + +.macro RESTORE_VSX VSX OFFSET FRAME + li 16, \OFFSET + lxvx \VSX, 16, \FRAME +.endm + +.macro SAVE_REGS + mflr 0 + std 0, 16(1) + stdu 1,-752(1) + + SAVE_GPR 14, 112, 1 + SAVE_GPR 15, 120, 1 + SAVE_GPR 16, 128, 1 + SAVE_GPR 17, 136, 1 + SAVE_GPR 18, 144, 1 + SAVE_GPR 19, 152, 1 + SAVE_GPR 20, 160, 1 + SAVE_GPR 21, 168, 1 + SAVE_GPR 22, 176, 1 + SAVE_GPR 23, 184, 1 + SAVE_GPR 24, 192, 1 + SAVE_GPR 25, 200, 1 + SAVE_GPR 26, 208, 1 + SAVE_GPR 27, 216, 1 + SAVE_GPR 28, 224, 1 + SAVE_GPR 29, 232, 1 + SAVE_GPR 30, 240, 1 + SAVE_GPR 31, 248, 1 + + addi 9, 1, 256 + SAVE_VRS 20, 0, 9 + SAVE_VRS 21, 16, 9 + SAVE_VRS 22, 32, 9 + SAVE_VRS 23, 48, 9 + SAVE_VRS 24, 64, 9 + SAVE_VRS 25, 80, 9 + SAVE_VRS 26, 96, 9 + SAVE_VRS 27, 112, 9 + SAVE_VRS 28, 128, 9 + SAVE_VRS 29, 144, 9 + SAVE_VRS 30, 160, 9 + SAVE_VRS 31, 176, 9 + + SAVE_VSX 14, 192, 9 + SAVE_VSX 15, 208, 9 + SAVE_VSX 16, 224, 9 + SAVE_VSX 17, 240, 9 + SAVE_VSX 18, 256, 9 + SAVE_VSX 19, 272, 9 + SAVE_VSX 20, 288, 9 + SAVE_VSX 21, 304, 9 + SAVE_VSX 22, 320, 9 + SAVE_VSX 23, 336, 9 + SAVE_VSX 24, 352, 9 + SAVE_VSX 25, 368, 9 + SAVE_VSX 26, 384, 9 + SAVE_VSX 27, 400, 9 + SAVE_VSX 28, 416, 9 + SAVE_VSX 29, 432, 9 + SAVE_VSX 30, 448, 9 + SAVE_VSX 31, 464, 9 +.endm # SAVE_REGS + +.macro RESTORE_REGS + addi 9, 1, 256 + RESTORE_VRS 20, 0, 9 + RESTORE_VRS 21, 16, 9 + RESTORE_VRS 22, 32, 9 + RESTORE_VRS 23, 48, 9 + RESTORE_VRS 24, 64, 9 + RESTORE_VRS 25, 80, 9 + RESTORE_VRS 26, 96, 9 + RESTORE_VRS 27, 112, 9 + RESTORE_VRS 28, 128, 9 + RESTORE_VRS 29, 144, 9 + RESTORE_VRS 30, 160, 9 + RESTORE_VRS 31, 176, 9 + + RESTORE_VSX 14, 192, 9 + RESTORE_VSX 15, 208, 9 + RESTORE_VSX 16, 224, 9 + RESTORE_VSX 17, 240, 9 + RESTORE_VSX 18, 256, 9 + RESTORE_VSX 19, 272, 9 + RESTORE_VSX 20, 288, 9 + RESTORE_VSX 21, 304, 9 + RESTORE_VSX 22, 320, 9 + RESTORE_VSX 23, 336, 9 + RESTORE_VSX 24, 352, 9 + RESTORE_VSX 25, 368, 9 + RESTORE_VSX 26, 384, 9 + RESTORE_VSX 27, 400, 9 + RESTORE_VSX 28, 416, 9 + RESTORE_VSX 29, 432, 9 + RESTORE_VSX 30, 448, 9 + RESTORE_VSX 31, 464, 9 + + RESTORE_GPR 14, 112, 1 + RESTORE_GPR 15, 120, 1 + RESTORE_GPR 16, 128, 1 + RESTORE_GPR 17, 136, 1 + RESTORE_GPR 18, 144, 1 + RESTORE_GPR 19, 152, 1 + RESTORE_GPR 20, 160, 1 + RESTORE_GPR 21, 168, 1 + RESTORE_GPR 22, 176, 1 + RESTORE_GPR 23, 184, 1 + RESTORE_GPR 24, 192, 1 + RESTORE_GPR 25, 200, 1 + RESTORE_GPR 26, 208, 1 + RESTORE_GPR 27, 216, 1 + RESTORE_GPR 28, 224, 1 + RESTORE_GPR 29, 232, 1 + RESTORE_GPR 30, 240, 1 + RESTORE_GPR 31, 248, 1 + + addi 1, 1, 752 + ld 0, 16(1) + mtlr 0 +.endm # RESTORE_REGS + +.macro QT_loop_8x + # QR(v0, v4, v8, v12, v1, v5, v9, v13, v2, v6, v10, v14, v3, v7, v11, v15) + xxlor 0, 32+25, 32+25 + xxlor 32+25, 20, 20 + vadduwm 0, 0, 4 + vadduwm 1, 1, 5 + vadduwm 2, 2, 6 + vadduwm 3, 3, 7 + vadduwm 16, 16, 20 + vadduwm 17, 17, 21 + vadduwm 18, 18, 22 + vadduwm 19, 19, 23 + + vpermxor 12, 12, 0, 25 + vpermxor 13, 13, 1, 25 + vpermxor 14, 14, 2, 25 + vpermxor 15, 15, 3, 25 + vpermxor 28, 28, 16, 25 + vpermxor 29, 29, 17, 25 + vpermxor 30, 30, 18, 25 + vpermxor 31, 31, 19, 25 + xxlor 32+25, 0, 0 + vadduwm 8, 8, 12 + vadduwm 9, 9, 13 + vadduwm 10, 10, 14 + vadduwm 11, 11, 15 + vadduwm 24, 24, 28 + vadduwm 25, 25, 29 + vadduwm 26, 26, 30 + vadduwm 27, 27, 31 + vxor 4, 4, 8 + vxor 5, 5, 9 + vxor 6, 6, 10 + vxor 7, 7, 11 + vxor 20, 20, 24 + vxor 21, 21, 25 + vxor 22, 22, 26 + vxor 23, 23, 27 + + xxlor 0, 32+25, 32+25 + xxlor 32+25, 21, 21 + vrlw 4, 4, 25 # + vrlw 5, 5, 25 + vrlw 6, 6, 25 + vrlw 7, 7, 25 + vrlw 20, 20, 25 # + vrlw 21, 21, 25 + vrlw 22, 22, 25 + vrlw 23, 23, 25 + xxlor 32+25, 0, 0 + vadduwm 0, 0, 4 + vadduwm 1, 1, 5 + vadduwm 2, 2, 6 + vadduwm 3, 3, 7 + vadduwm 16, 16, 20 + vadduwm 17, 17, 21 + vadduwm 18, 18, 22 + vadduwm 19, 19, 23 + + xxlor 0, 32+25, 32+25 + xxlor 32+25, 22, 22 + vpermxor 12, 12, 0, 25 + vpermxor 13, 13, 1, 25 + vpermxor 14, 14, 2, 25 + vpermxor 15, 15, 3, 25 + vpermxor 28, 28, 16, 25 + vpermxor 29, 29, 17, 25 + vpermxor 30, 30, 18, 25 + vpermxor 31, 31, 19, 25 + xxlor 32+25, 0, 0 + vadduwm 8, 8, 12 + vadduwm 9, 9, 13 + vadduwm 10, 10, 14 + vadduwm 11, 11, 15 + vadduwm 24, 24, 28 + vadduwm 25, 25, 29 + vadduwm 26, 26, 30 + vadduwm 27, 27, 31 + xxlor 0, 32+28, 32+28 + xxlor 32+28, 23, 23 + vxor 4, 4, 8 + vxor 5, 5, 9 + vxor 6, 6, 10 + vxor 7, 7, 11 + vxor 20, 20, 24 + vxor 21, 21, 25 + vxor 22, 22, 26 + vxor 23, 23, 27 + vrlw 4, 4, 28 # + vrlw 5, 5, 28 + vrlw 6, 6, 28 + vrlw 7, 7, 28 + vrlw 20, 20, 28 # + vrlw 21, 21, 28 + vrlw 22, 22, 28 + vrlw 23, 23, 28 + xxlor 32+28, 0, 0 + + # QR(v0, v5, v10, v15, v1, v6, v11, v12, v2, v7, v8, v13, v3, v4, v9, v14) + xxlor 0, 32+25, 32+25 + xxlor 32+25, 20, 20 + vadduwm 0, 0, 5 + vadduwm 1, 1, 6 + vadduwm 2, 2, 7 + vadduwm 3, 3, 4 + vadduwm 16, 16, 21 + vadduwm 17, 17, 22 + vadduwm 18, 18, 23 + vadduwm 19, 19, 20 + + vpermxor 15, 15, 0, 25 + vpermxor 12, 12, 1, 25 + vpermxor 13, 13, 2, 25 + vpermxor 14, 14, 3, 25 + vpermxor 31, 31, 16, 25 + vpermxor 28, 28, 17, 25 + vpermxor 29, 29, 18, 25 + vpermxor 30, 30, 19, 25 + + xxlor 32+25, 0, 0 + vadduwm 10, 10, 15 + vadduwm 11, 11, 12 + vadduwm 8, 8, 13 + vadduwm 9, 9, 14 + vadduwm 26, 26, 31 + vadduwm 27, 27, 28 + vadduwm 24, 24, 29 + vadduwm 25, 25, 30 + vxor 5, 5, 10 + vxor 6, 6, 11 + vxor 7, 7, 8 + vxor 4, 4, 9 + vxor 21, 21, 26 + vxor 22, 22, 27 + vxor 23, 23, 24 + vxor 20, 20, 25 + + xxlor 0, 32+25, 32+25 + xxlor 32+25, 21, 21 + vrlw 5, 5, 25 + vrlw 6, 6, 25 + vrlw 7, 7, 25 + vrlw 4, 4, 25 + vrlw 21, 21, 25 + vrlw 22, 22, 25 + vrlw 23, 23, 25 + vrlw 20, 20, 25 + xxlor 32+25, 0, 0 + + vadduwm 0, 0, 5 + vadduwm 1, 1, 6 + vadduwm 2, 2, 7 + vadduwm 3, 3, 4 + vadduwm 16, 16, 21 + vadduwm 17, 17, 22 + vadduwm 18, 18, 23 + vadduwm 19, 19, 20 + + xxlor 0, 32+25, 32+25 + xxlor 32+25, 22, 22 + vpermxor 15, 15, 0, 25 + vpermxor 12, 12, 1, 25 + vpermxor 13, 13, 2, 25 + vpermxor 14, 14, 3, 25 + vpermxor 31, 31, 16, 25 + vpermxor 28, 28, 17, 25 + vpermxor 29, 29, 18, 25 + vpermxor 30, 30, 19, 25 + xxlor 32+25, 0, 0 + + vadduwm 10, 10, 15 + vadduwm 11, 11, 12 + vadduwm 8, 8, 13 + vadduwm 9, 9, 14 + vadduwm 26, 26, 31 + vadduwm 27, 27, 28 + vadduwm 24, 24, 29 + vadduwm 25, 25, 30 + + xxlor 0, 32+28, 32+28 + xxlor 32+28, 23, 23 + vxor 5, 5, 10 + vxor 6, 6, 11 + vxor 7, 7, 8 + vxor 4, 4, 9 + vxor 21, 21, 26 + vxor 22, 22, 27 + vxor 23, 23, 24 + vxor 20, 20, 25 + vrlw 5, 5, 28 + vrlw 6, 6, 28 + vrlw 7, 7, 28 + vrlw 4, 4, 28 + vrlw 21, 21, 28 + vrlw 22, 22, 28 + vrlw 23, 23, 28 + vrlw 20, 20, 28 + xxlor 32+28, 0, 0 +.endm + +.macro QT_loop_4x + # QR(v0, v4, v8, v12, v1, v5, v9, v13, v2, v6, v10, v14, v3, v7, v11, v15) + vadduwm 0, 0, 4 + vadduwm 1, 1, 5 + vadduwm 2, 2, 6 + vadduwm 3, 3, 7 + vpermxor 12, 12, 0, 20 + vpermxor 13, 13, 1, 20 + vpermxor 14, 14, 2, 20 + vpermxor 15, 15, 3, 20 + vadduwm 8, 8, 12 + vadduwm 9, 9, 13 + vadduwm 10, 10, 14 + vadduwm 11, 11, 15 + vxor 4, 4, 8 + vxor 5, 5, 9 + vxor 6, 6, 10 + vxor 7, 7, 11 + vrlw 4, 4, 21 + vrlw 5, 5, 21 + vrlw 6, 6, 21 + vrlw 7, 7, 21 + vadduwm 0, 0, 4 + vadduwm 1, 1, 5 + vadduwm 2, 2, 6 + vadduwm 3, 3, 7 + vpermxor 12, 12, 0, 22 + vpermxor 13, 13, 1, 22 + vpermxor 14, 14, 2, 22 + vpermxor 15, 15, 3, 22 + vadduwm 8, 8, 12 + vadduwm 9, 9, 13 + vadduwm 10, 10, 14 + vadduwm 11, 11, 15 + vxor 4, 4, 8 + vxor 5, 5, 9 + vxor 6, 6, 10 + vxor 7, 7, 11 + vrlw 4, 4, 23 + vrlw 5, 5, 23 + vrlw 6, 6, 23 + vrlw 7, 7, 23 + + # QR(v0, v5, v10, v15, v1, v6, v11, v12, v2, v7, v8, v13, v3, v4, v9, v14) + vadduwm 0, 0, 5 + vadduwm 1, 1, 6 + vadduwm 2, 2, 7 + vadduwm 3, 3, 4 + vpermxor 15, 15, 0, 20 + vpermxor 12, 12, 1, 20 + vpermxor 13, 13, 2, 20 + vpermxor 14, 14, 3, 20 + vadduwm 10, 10, 15 + vadduwm 11, 11, 12 + vadduwm 8, 8, 13 + vadduwm 9, 9, 14 + vxor 5, 5, 10 + vxor 6, 6, 11 + vxor 7, 7, 8 + vxor 4, 4, 9 + vrlw 5, 5, 21 + vrlw 6, 6, 21 + vrlw 7, 7, 21 + vrlw 4, 4, 21 + vadduwm 0, 0, 5 + vadduwm 1, 1, 6 + vadduwm 2, 2, 7 + vadduwm 3, 3, 4 + vpermxor 15, 15, 0, 22 + vpermxor 12, 12, 1, 22 + vpermxor 13, 13, 2, 22 + vpermxor 14, 14, 3, 22 + vadduwm 10, 10, 15 + vadduwm 11, 11, 12 + vadduwm 8, 8, 13 + vadduwm 9, 9, 14 + vxor 5, 5, 10 + vxor 6, 6, 11 + vxor 7, 7, 8 + vxor 4, 4, 9 + vrlw 5, 5, 23 + vrlw 6, 6, 23 + vrlw 7, 7, 23 + vrlw 4, 4, 23 +.endm + +# Transpose +.macro TP_4x a0 a1 a2 a3 + xxmrghw 10, 32+\a0, 32+\a1 # a0, a1, b0, b1 + xxmrghw 11, 32+\a2, 32+\a3 # a2, a3, b2, b3 + xxmrglw 12, 32+\a0, 32+\a1 # c0, c1, d0, d1 + xxmrglw 13, 32+\a2, 32+\a3 # c2, c3, d2, d3 + xxpermdi 32+\a0, 10, 11, 0 # a0, a1, a2, a3 + xxpermdi 32+\a1, 10, 11, 3 # b0, b1, b2, b3 + xxpermdi 32+\a2, 12, 13, 0 # c0, c1, c2, c3 + xxpermdi 32+\a3, 12, 13, 3 # d0, d1, d2, d3 +.endm + +# key stream = working state + state +.macro Add_state S + vadduwm \S+0, \S+0, 16-\S + vadduwm \S+4, \S+4, 17-\S + vadduwm \S+8, \S+8, 18-\S + vadduwm \S+12, \S+12, 19-\S + + vadduwm \S+1, \S+1, 16-\S + vadduwm \S+5, \S+5, 17-\S + vadduwm \S+9, \S+9, 18-\S + vadduwm \S+13, \S+13, 19-\S + + vadduwm \S+2, \S+2, 16-\S + vadduwm \S+6, \S+6, 17-\S + vadduwm \S+10, \S+10, 18-\S + vadduwm \S+14, \S+14, 19-\S + + vadduwm \S+3, \S+3, 16-\S + vadduwm \S+7, \S+7, 17-\S + vadduwm \S+11, \S+11, 18-\S + vadduwm \S+15, \S+15, 19-\S +.endm + +# +# write 256 bytes +# +.macro Write_256 S + add 9, 14, 5 + add 16, 14, 4 + lxvw4x 0, 0, 9 + lxvw4x 1, 17, 9 + lxvw4x 2, 18, 9 + lxvw4x 3, 19, 9 + lxvw4x 4, 20, 9 + lxvw4x 5, 21, 9 + lxvw4x 6, 22, 9 + lxvw4x 7, 23, 9 + lxvw4x 8, 24, 9 + lxvw4x 9, 25, 9 + lxvw4x 10, 26, 9 + lxvw4x 11, 27, 9 + lxvw4x 12, 28, 9 + lxvw4x 13, 29, 9 + lxvw4x 14, 30, 9 + lxvw4x 15, 31, 9 + + xxlxor \S+32, \S+32, 0 + xxlxor \S+36, \S+36, 1 + xxlxor \S+40, \S+40, 2 + xxlxor \S+44, \S+44, 3 + xxlxor \S+33, \S+33, 4 + xxlxor \S+37, \S+37, 5 + xxlxor \S+41, \S+41, 6 + xxlxor \S+45, \S+45, 7 + xxlxor \S+34, \S+34, 8 + xxlxor \S+38, \S+38, 9 + xxlxor \S+42, \S+42, 10 + xxlxor \S+46, \S+46, 11 + xxlxor \S+35, \S+35, 12 + xxlxor \S+39, \S+39, 13 + xxlxor \S+43, \S+43, 14 + xxlxor \S+47, \S+47, 15 + + stxvw4x \S+32, 0, 16 + stxvw4x \S+36, 17, 16 + stxvw4x \S+40, 18, 16 + stxvw4x \S+44, 19, 16 + + stxvw4x \S+33, 20, 16 + stxvw4x \S+37, 21, 16 + stxvw4x \S+41, 22, 16 + stxvw4x \S+45, 23, 16 + + stxvw4x \S+34, 24, 16 + stxvw4x \S+38, 25, 16 + stxvw4x \S+42, 26, 16 + stxvw4x \S+46, 27, 16 + + stxvw4x \S+35, 28, 16 + stxvw4x \S+39, 29, 16 + stxvw4x \S+43, 30, 16 + stxvw4x \S+47, 31, 16 + +.endm + +# +# chacha20_p10le_8x(u32 *state, byte *dst, const byte *src, size_t len, int nrounds); +# +SYM_FUNC_START(chacha_p10le_8x) +.align 5 + cmpdi 6, 0 + ble Out_no_chacha + + SAVE_REGS + + # r17 - r31 mainly for Write_256 macro. + li 17, 16 + li 18, 32 + li 19, 48 + li 20, 64 + li 21, 80 + li 22, 96 + li 23, 112 + li 24, 128 + li 25, 144 + li 26, 160 + li 27, 176 + li 28, 192 + li 29, 208 + li 30, 224 + li 31, 240 + + mr 15, 6 # len + li 14, 0 # offset to inp and outp + + lxvw4x 48, 0, 3 # vr16, constants + lxvw4x 49, 17, 3 # vr17, key 1 + lxvw4x 50, 18, 3 # vr18, key 2 + lxvw4x 51, 19, 3 # vr19, counter, nonce + + # create (0, 1, 2, 3) counters + vspltisw 0, 0 + vspltisw 1, 1 + vspltisw 2, 2 + vspltisw 3, 3 + vmrghw 4, 0, 1 + vmrglw 5, 2, 3 + vsldoi 30, 4, 5, 8 # vr30 counter, 4 (0, 1, 2, 3) + + vspltisw 21, 12 + vspltisw 23, 7 + + addis 11, 2, permx@toc@ha + addi 11, 11, permx@toc@l + lxvw4x 32+20, 0, 11 + lxvw4x 32+22, 17, 11 + + sradi 8, 7, 1 + + mtctr 8 + + # save constants to vsx + xxlor 16, 48, 48 + xxlor 17, 49, 49 + xxlor 18, 50, 50 + xxlor 19, 51, 51 + + vspltisw 25, 4 + vspltisw 26, 8 + + xxlor 25, 32+26, 32+26 + xxlor 24, 32+25, 32+25 + + vadduwm 31, 30, 25 # counter = (0, 1, 2, 3) + (4, 4, 4, 4) + xxlor 30, 32+30, 32+30 + xxlor 31, 32+31, 32+31 + + xxlor 20, 32+20, 32+20 + xxlor 21, 32+21, 32+21 + xxlor 22, 32+22, 32+22 + xxlor 23, 32+23, 32+23 + + cmpdi 6, 512 + blt Loop_last + +Loop_8x: + xxspltw 32+0, 16, 0 + xxspltw 32+1, 16, 1 + xxspltw 32+2, 16, 2 + xxspltw 32+3, 16, 3 + + xxspltw 32+4, 17, 0 + xxspltw 32+5, 17, 1 + xxspltw 32+6, 17, 2 + xxspltw 32+7, 17, 3 + xxspltw 32+8, 18, 0 + xxspltw 32+9, 18, 1 + xxspltw 32+10, 18, 2 + xxspltw 32+11, 18, 3 + xxspltw 32+12, 19, 0 + xxspltw 32+13, 19, 1 + xxspltw 32+14, 19, 2 + xxspltw 32+15, 19, 3 + vadduwm 12, 12, 30 # increase counter + + xxspltw 32+16, 16, 0 + xxspltw 32+17, 16, 1 + xxspltw 32+18, 16, 2 + xxspltw 32+19, 16, 3 + + xxspltw 32+20, 17, 0 + xxspltw 32+21, 17, 1 + xxspltw 32+22, 17, 2 + xxspltw 32+23, 17, 3 + xxspltw 32+24, 18, 0 + xxspltw 32+25, 18, 1 + xxspltw 32+26, 18, 2 + xxspltw 32+27, 18, 3 + xxspltw 32+28, 19, 0 + xxspltw 32+29, 19, 1 + vadduwm 28, 28, 31 # increase counter + xxspltw 32+30, 19, 2 + xxspltw 32+31, 19, 3 + +.align 5 +quarter_loop_8x: + QT_loop_8x + + bdnz quarter_loop_8x + + xxlor 0, 32+30, 32+30 + xxlor 32+30, 30, 30 + vadduwm 12, 12, 30 + xxlor 32+30, 0, 0 + TP_4x 0, 1, 2, 3 + TP_4x 4, 5, 6, 7 + TP_4x 8, 9, 10, 11 + TP_4x 12, 13, 14, 15 + + xxlor 0, 48, 48 + xxlor 1, 49, 49 + xxlor 2, 50, 50 + xxlor 3, 51, 51 + xxlor 48, 16, 16 + xxlor 49, 17, 17 + xxlor 50, 18, 18 + xxlor 51, 19, 19 + Add_state 0 + xxlor 48, 0, 0 + xxlor 49, 1, 1 + xxlor 50, 2, 2 + xxlor 51, 3, 3 + Write_256 0 + addi 14, 14, 256 # offset +=256 + addi 15, 15, -256 # len -=256 + + xxlor 5, 32+31, 32+31 + xxlor 32+31, 31, 31 + vadduwm 28, 28, 31 + xxlor 32+31, 5, 5 + TP_4x 16+0, 16+1, 16+2, 16+3 + TP_4x 16+4, 16+5, 16+6, 16+7 + TP_4x 16+8, 16+9, 16+10, 16+11 + TP_4x 16+12, 16+13, 16+14, 16+15 + + xxlor 32, 16, 16 + xxlor 33, 17, 17 + xxlor 34, 18, 18 + xxlor 35, 19, 19 + Add_state 16 + Write_256 16 + addi 14, 14, 256 # offset +=256 + addi 15, 15, -256 # len +=256 + + xxlor 32+24, 24, 24 + xxlor 32+25, 25, 25 + xxlor 32+30, 30, 30 + vadduwm 30, 30, 25 + vadduwm 31, 30, 24 + xxlor 30, 32+30, 32+30 + xxlor 31, 32+31, 32+31 + + cmpdi 15, 0 + beq Out_loop + + cmpdi 15, 512 + blt Loop_last + + mtctr 8 + b Loop_8x + +Loop_last: + lxvw4x 48, 0, 3 # vr16, constants + lxvw4x 49, 17, 3 # vr17, key 1 + lxvw4x 50, 18, 3 # vr18, key 2 + lxvw4x 51, 19, 3 # vr19, counter, nonce + + vspltisw 21, 12 + vspltisw 23, 7 + addis 11, 2, permx@toc@ha + addi 11, 11, permx@toc@l + lxvw4x 32+20, 0, 11 + lxvw4x 32+22, 17, 11 + + sradi 8, 7, 1 + mtctr 8 + +Loop_4x: + vspltw 0, 16, 0 + vspltw 1, 16, 1 + vspltw 2, 16, 2 + vspltw 3, 16, 3 + + vspltw 4, 17, 0 + vspltw 5, 17, 1 + vspltw 6, 17, 2 + vspltw 7, 17, 3 + vspltw 8, 18, 0 + vspltw 9, 18, 1 + vspltw 10, 18, 2 + vspltw 11, 18, 3 + vspltw 12, 19, 0 + vadduwm 12, 12, 30 # increase counter + vspltw 13, 19, 1 + vspltw 14, 19, 2 + vspltw 15, 19, 3 + +.align 5 +quarter_loop: + QT_loop_4x + + bdnz quarter_loop + + vadduwm 12, 12, 30 + TP_4x 0, 1, 2, 3 + TP_4x 4, 5, 6, 7 + TP_4x 8, 9, 10, 11 + TP_4x 12, 13, 14, 15 + + Add_state 0 + Write_256 0 + addi 14, 14, 256 # offset += 256 + addi 15, 15, -256 # len += 256 + + # Update state counter + vspltisw 25, 4 + vadduwm 30, 30, 25 + + cmpdi 15, 0 + beq Out_loop + cmpdi 15, 256 + blt Out_loop + + mtctr 8 + b Loop_4x + +Out_loop: + RESTORE_REGS + blr + +Out_no_chacha: + li 3, 0 + blr +SYM_FUNC_END(chacha_p10le_8x) + +SYM_DATA_START_LOCAL(PERMX) +.align 5 +permx: +.long 0x22330011, 0x66774455, 0xaabb8899, 0xeeffccdd +.long 0x11223300, 0x55667744, 0x99aabb88, 0xddeeffcc +SYM_DATA_END(PERMX)
Signature · 1 line
2.31.1
PATCH v2 5/5 Update Kconfig and Makefile.
In reply to Danny TsenDefined CRYPTO_CHACHA20_P10 and CRYPTO POLY1305_P10 in Kconfig to support optimized implementation for Power10 and later CPU.
Added new module driver chacha-p10-crypto and poly1305-p10-crypto.
--- arch/powerpc/crypto/Kconfig | 26 ++++++++++++++++++++++++++ arch/powerpc/crypto/Makefile | 4 ++++ 2 files changed, 30 insertions(+)
arch/powerpc/crypto/Kconfig +26 -0
@@ -111,4 +111,30 @@ config CRYPTO_AES_GCM_P10 Support for cryptographic acceleration instructions on Power10 or later CPU. This module supports stitched acceleration for AES/GCM. +config CRYPTO_CHACHA20_P10 + tristate "Ciphers: ChaCha20, XChacha20, XChacha12 (P10 or later)" + depends on PPC64 && CPU_LITTLE_ENDIAN + select CRYPTO_SKCIPHER + select CRYPTO_LIB_CHACHA_GENERIC + select CRYPTO_ARCH_HAVE_LIB_CHACHA + help + Length-preserving ciphers: ChaCha20, XChaCha20, and XChaCha12 + stream cipher algorithms + + Architecture: PowerPC64 + - Power10 or later + - Little-endian + +config CRYPTO_POLY1305_P10 + tristate "Hash functions: Poly1305 (P10 or later)" + depends on PPC64 && CPU_LITTLE_ENDIAN + select CRYPTO_HASH + select CRYPTO_LIB_POLY1305_GENERIC + help + Poly1305 authenticator algorithm (RFC7539) + + Architecture: PowerPC64 + - Power10 or later + - Little-endian + endmenu
arch/powerpc/crypto/Makefile +4 -0
@@ -14,6 +14,8 @@ obj-$(CONFIG_CRYPTO_CRC32C_VPMSUM) += crc32c-vpmsum.o obj-$(CONFIG_CRYPTO_CRCT10DIF_VPMSUM) += crct10dif-vpmsum.o obj-$(CONFIG_CRYPTO_VPMSUM_TESTER) += crc-vpmsum_test.o obj-$(CONFIG_CRYPTO_AES_GCM_P10) += aes-gcm-p10-crypto.o +obj-$(CONFIG_CRYPTO_CHACHA20_P10) += chacha-p10-crypto.o +obj-$(CONFIG_CRYPTO_POLY1305_P10) += poly1305-p10-crypto.o aes-ppc-spe-y := aes-spe-core.o aes-spe-keys.o aes-tab-4k.o aes-spe-modes.o aes-spe-glue.o md5-ppc-y := md5-asm.o md5-glue.o
@@ -23,6 +25,8 @@ sha256-ppc-spe-y := sha256-spe-asm.o sha256-spe-glue.o crc32c-vpmsum-y := crc32c-vpmsum_asm.o crc32c-vpmsum_glue.o crct10dif-vpmsum-y := crct10dif-vpmsum_asm.o crct10dif-vpmsum_glue.o aes-gcm-p10-crypto-y := aes-gcm-p10-glue.o aes-gcm-p10.o ghashp8-ppc.o aesp8-ppc.o +chacha-p10-crypto-y := chacha-p10-glue.o chacha-p10le-8x.o +poly1305-p10-crypto-y := poly1305-p10-glue.o poly1305-p10le_64.o quiet_cmd_perl = PERL $@ cmd_perl = $(PERL) $< $(if $(CONFIG_CPU_LITTLE_ENDIAN), linux-ppc64le, linux-ppc64) > $@
Signature · 1 line
2.31.1
PATCH v2 2/5 Glue code for optmized Chacha20 implementation for ppc64le.
In reply to Danny Tsen--- arch/powerpc/crypto/chacha-p10-glue.c | 221 ++++++++++++++++++++++++++ 1 file changed, 221 insertions(+) create mode 100644 arch/powerpc/crypto/chacha-p10-glue.c
arch/powerpc/crypto/chacha-p10-glue.c new file +221 -0
@@ -0,0 +1,221 @@ +// SPDX-License-Identifier: GPL-2.0-or-later +/* +* PowerPC P10 (ppc64le) accelerated ChaCha and XChaCha stream ciphers, +* including ChaCha20 (RFC7539) +* +* Copyright 2023- IBM Corp. All rights reserved. +*/ + +#include <crypto/algapi.h> +#include <crypto/internal/chacha.h> +#include <crypto/internal/simd.h> +#include <crypto/internal/skcipher.h> +#include <linux/kernel.h> +#include <linux/module.h> +#include <linux/cpufeature.h> +#include <linux/sizes.h> +#include <asm/simd.h> +#include <asm/switch_to.h> + +asmlinkage void chacha_p10le_8x(u32 *state, u8 *dst, const u8 *src, +unsigned int len,int nrounds); + +static __ro_after_init DEFINE_STATIC_KEY_FALSE (have_p10); + +static void vsx_begin (void ) +{ +preempt_disable (); +enable_kernel_vsx (); +} + +static void vsx_end (void ) +{ +disable_kernel_vsx (); +preempt_enable (); +} + +static void chacha_p10_do_8x (u32 *state,u8 *dst,const u8 *src, +unsigned int bytes,int nrounds) +{ +unsigned int l = bytes & ~0x0FF ; + +if (l >0 ) { +chacha_p10le_8x (state, dst, src, l, nrounds); + bytes -= l; + src += l; + dst += l; + state[12 ] += l /CHACHA_BLOCK_SIZE ; + } + +if (bytes >0 ) +chacha_crypt_generic (state, dst, src, bytes, nrounds); +} + +void hchacha_block_arch (const u32 *state,u32 *stream,int nrounds) +{ +hchacha_block_generic (state, stream, nrounds); +} +EXPORT_SYMBOL (hchacha_block_arch); + +void chacha_init_arch (u32 *state,const u32 *key,const u8 *iv) +{ +chacha_init_generic (state, key, iv); +} +EXPORT_SYMBOL (chacha_init_arch); + +void chacha_crypt_arch (u32 *state,u8 *dst,const u8 *src,unsigned int bytes, +int nrounds) +{ +if (!static_branch_likely (&have_p10) || bytes <=CHACHA_BLOCK_SIZE || + !crypto_simd_usable ()) +return chacha_crypt_generic (state, dst, src, bytes, nrounds); + +do { + unsigned int todo = min_t(unsigned int, bytes, SZ_4K); + +vsx_begin (); +chacha_p10_do_8x (state, dst, src, todo, nrounds); +vsx_end (); + + bytes -= todo; + src += todo; + dst += todo; + }while (bytes); +} +EXPORT_SYMBOL (chacha_crypt_arch); + +static int chacha_p10_stream_xor (struct skcipher_request *req, +const struct chacha_ctx *ctx,const u8 *iv) +{ +struct skcipher_walk walk; +u32 state[16 ]; +int err; + + err =skcipher_walk_virt (&walk, req, false); +if (err) +return err; + +chacha_init_generic (state, ctx->key, iv); + +while (walk.nbytes >0 ) { +unsigned int nbytes = walk.nbytes; + +if (nbytes < walk.total) + nbytes =rounddown (nbytes, walk.stride); + +if (!crypto_simd_usable ()) { +chacha_crypt_generic (state, walk.dst.virt.addr, + walk.src.virt.addr, nbytes, + ctx->nrounds); + }else { +vsx_begin (); +chacha_p10_do_8x (state, walk.dst.virt.addr, + walk.src.virt.addr, nbytes, ctx->nrounds); +vsx_end (); + } + err =skcipher_walk_done (&walk, walk.nbytes - nbytes); +if (err) +break ; + } + +return err; +} + +static int chacha_p10 (struct skcipher_request *req) +{ +struct crypto_skcipher *tfm =crypto_skcipher_reqtfm (req); +struct chacha_ctx *ctx =crypto_skcipher_ctx (tfm); + +return chacha_p10_stream_xor (req, ctx, req->iv); +} + +static int xchacha_p10 (struct skcipher_request *req) +{ +struct crypto_skcipher *tfm =crypto_skcipher_reqtfm (req); +struct chacha_ctx *ctx =crypto_skcipher_ctx (tfm); +struct chacha_ctx subctx; +u32 state[16 ]; +u8 real_iv[16 ]; + +chacha_init_generic (state, ctx->key, req->iv); +hchacha_block_arch (state, subctx.key, ctx->nrounds); + subctx.nrounds = ctx->nrounds; + +memcpy (&real_iv[0 ], req->iv +24 ,8 ); +memcpy (&real_iv[8 ], req->iv +16 ,8 ); +return chacha_p10_stream_xor (req, &subctx, real_iv); +} + +static struct skcipher_alg algs[] = { + { + .base.cra_name ="chacha20" , + .base.cra_driver_name ="chacha20-p10" , + .base.cra_priority =300 , + .base.cra_blocksize =1 , + .base.cra_ctxsize =sizeof (struct chacha_ctx ), + .base.cra_module =THIS_MODULE , + + .min_keysize =CHACHA_KEY_SIZE , + .max_keysize =CHACHA_KEY_SIZE , + .ivsize =CHACHA_IV_SIZE , + .chunksize =CHACHA_BLOCK_SIZE , + .setkey = chacha20_setkey, + .encrypt = chacha_p10, + .decrypt = chacha_p10, + }, { + .base.cra_name ="xchacha20" , + .base.cra_driver_name ="xchacha20-p10" , + .base.cra_priority =300 , + .base.cra_blocksize =1 , + .base.cra_ctxsize =sizeof (struct chacha_ctx ), + .base.cra_module =THIS_MODULE , + + .min_keysize =CHACHA_KEY_SIZE , + .max_keysize =CHACHA_KEY_SIZE , + .ivsize =XCHACHA_IV_SIZE , + .chunksize =CHACHA_BLOCK_SIZE , + .setkey = chacha20_setkey, + .encrypt = xchacha_p10, + .decrypt = xchacha_p10, + }, { + .base.cra_name ="xchacha12" , + .base.cra_driver_name ="xchacha12-p10" , + .base.cra_priority =300 , + .base.cra_blocksize =1 , + .base.cra_ctxsize =sizeof (struct chacha_ctx ), + .base.cra_module =THIS_MODULE , + + .min_keysize =CHACHA_KEY_SIZE , + .max_keysize =CHACHA_KEY_SIZE , + .ivsize =XCHACHA_IV_SIZE , + .chunksize =CHACHA_BLOCK_SIZE , + .setkey = chacha12_setkey, + .encrypt = xchacha_p10, + .decrypt = xchacha_p10, + } +}; + +static int __init chacha_p10_init (void ) +{ +static_branch_enable (&have_p10); + +return crypto_register_skciphers (algs,ARRAY_SIZE (algs)); +} + +static void __exit chacha_p10_exit(void) +{ +crypto_unregister_skciphers (algs,ARRAY_SIZE (algs)); +} + +module_cpu_feature_match (PPC_MODULE_FEATURE_P10 , chacha_p10_init); +module_exit (chacha_p10_exit); + +MODULE_DESCRIPTION ("ChaCha and XChaCha stream ciphers (P10 accelerated)" ); +MODULE_AUTHOR ("Danny Tsen <[email protected]>" ); +MODULE_LICENSE ("GPL v2" ); +MODULE_ALIAS_CRYPTO ("chacha20" ); +MODULE_ALIAS_CRYPTO ("chacha20-p10" ); +MODULE_ALIAS_CRYPTO ("xchacha20" ); +MODULE_ALIAS_CRYPTO ("xchacha20-p10" ); +MODULE_ALIAS_CRYPTO ("xchacha12" ); +MODULE_ALIAS_CRYPTO ("xchacha12-p10" );
Signature · 1 line
2.31.1
Herbert Xu
maintainerQuoted 19 lines · Danny Tsen
This patch series provide an accelerated/optimized Chacha20 and Poly1305 implementation for Power10 or later CPU (ppc64le). This module implements algorithm specified in RFC7539. The implementation provides 3.5X better performance than the baseline for Chacha20 and Poly1305 individually and 1.5X improvement for Chacha20/Poly1305 operation.
This patch has been tested with the kernel crypto module tcrypt.ko and has passed the selftest. The patch is also tested with
CONFIG_CRYPTO_MANAGER_EXTRA_TESTSenabled.Danny Tsen (5):
An optimized Chacha20 implementation with 8-way unrolling for ppc64le. Glue code for optmized Chacha20 implementation for ppc64le. An optimized Poly1305 implementation with 4-way unrolling for ppc64le. Glue code for optmized Poly1305 implementation for ppc64le. UpdateKconfigandMakefile.
Quoted 14 lines
arch/powerpc/crypto/Kconfig | 26 + arch/powerpc/crypto/Makefile | 4 + arch/powerpc/crypto/chacha-p10-glue.c | 221 +++++ arch/powerpc/crypto/chacha-p10le-8x.S | 842 ++++++++++++++++++ arch/powerpc/crypto/poly1305-p10-glue.c | 186 ++++ arch/powerpc/crypto/poly1305-p10le_64.S | 1075 +++++++++++++++++++++++ 6 files changed, 2354 insertions(+) create mode 100644 arch/powerpc/crypto/chacha-p10-glue.c create mode 100644 arch/powerpc/crypto/chacha-p10le-8x.S create mode 100644 arch/powerpc/crypto/poly1305-p10-glue.c create mode 100644 arch/powerpc/crypto/poly1305-p10le_64.SSignature · 1 line
2.31.1
All applied. Thanks.
Signature · 3 lines
Email: Herbert Xu <[email protected]> Home Page: http://gondor.apana.org.au/~herbert/ PGP Key: http://gondor.apana.org.au/~herbert/pubkey.txt
Thanks.
-Danny
________________________________
From: Herbert Xu <[email protected]>
Sent: Friday, July 14, 2023 4:49 PM
To: Danny Tsen <[email protected]>
Cc: [email protected] <[email protected]>; [email protected] <[email protected]>; [email protected] <[email protected]>; [email protected] <[email protected]>; [email protected] <[email protected]>; [email protected] <[email protected]>; [email protected] <[email protected]>; [email protected] <[email protected]>; Danny Tsen <[email protected]>
Subject: [EXTERNAL] Re: [PATCH v2 0/5] crypto: Accelerated Chacha20/Poly1305 implementation
Quoted 31 lines · Danny Tsen
This patch series provide an accelerated/optimized Chacha20 and Poly1305 implementation for Power10 or later CPU (ppc64le). This module implements algorithm specified in RFC7539. The implementation provides 3.5X better performance than the baseline for Chacha20 and Poly1305 individually and 1.5X improvement for Chacha20/Poly1305 operation.
This patch has been tested with the kernel crypto module tcrypt.ko and has passed the selftest. The patch is also tested with
CONFIG_CRYPTO_MANAGER_EXTRA_TESTSenabled.Danny Tsen (5):
An optimized Chacha20 implementation with 8-way unrolling for ppc64le. Glue code for optmized Chacha20 implementation for ppc64le. An optimized Poly1305 implementation with 4-way unrolling for ppc64le. Glue code for optmized Poly1305 implementation for ppc64le. UpdateKconfigandMakefile.arch/powerpc/crypto/Kconfig | 26 + arch/powerpc/crypto/Makefile | 4 + arch/powerpc/crypto/chacha-p10-glue.c | 221 +++++ arch/powerpc/crypto/chacha-p10le-8x.S | 842 ++++++++++++++++++ arch/powerpc/crypto/poly1305-p10-glue.c | 186 ++++ arch/powerpc/crypto/poly1305-p10le_64.S | 1075 +++++++++++++++++++++++ 6 files changed, 2354 insertions(+) create mode 100644 arch/powerpc/crypto/chacha-p10-glue.c create mode 100644 arch/powerpc/crypto/chacha-p10le-8x.S create mode 100644 arch/powerpc/crypto/poly1305-p10-glue.c create mode 100644 arch/powerpc/crypto/poly1305-p10le_64.S
Quoted 2 lines
--
2.31.1
All applied. Thanks.
--
Email: Herbert Xu <[email protected]> Home Page: http://gondor.apana.org.au/~herbert/
PGP Key: http://gondor.apana.org.au/~herbert/pubkey.txt