/*-
 * Copyright (c) 2023 The FreeBSD Foundation
 *
 * This software was developed by Robert Clausecker <fuz@FreeBSD.org>
 * under sponsorship from the FreeBSD Foundation.
 *
 * Redistribution and use in source and binary forms, with or without
 * modification, are permitted provided that the following conditions
 * are met:
 * 1. Redistributions of source code must retain the above copyright
 *    notice, this list of conditions and the following disclaimer.
 * 2. Redistributions in binary form must reproduce the above copyright
 *    notice, this list of conditions and the following disclaimer in the
 *    documentation and/or other materials provided with the distribution.
 *
 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ''AS IS'' AND
 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
 * SUCH DAMAGE
 */

#include <machine/asm.h>

#include "amd64_archlevel.h"

#define ALIGN_TEXT      .p2align 4,0x90 /* 16-byte alignment, nop filled */

ARCHFUNCS(timingsafe_bcmp)
        ARCHFUNC(timingsafe_bcmp, scalar)
        ARCHFUNC(timingsafe_bcmp, baseline)
ENDARCHFUNCS(timingsafe_bcmp)

ARCHENTRY(timingsafe_bcmp, scalar)
        cmp     $16, %rdx               # at least 17 bytes to process?
        ja      .Lgt16

        cmp     $8, %edx                # at least 9 bytes to process?
        ja      .L0916

        cmp     $4, %edx                # at least 5 bytes to process?
        ja      .L0508

        cmp     $2, %edx                # at least 3 bytes to process?
        ja      .L0304

        test    %edx, %edx              # buffer empty?
        jnz     .L0102

        xor     %eax, %eax              # empty buffer always matches
        ret

.L0102: movzbl  (%rdi), %eax            # load 1--2 bytes from first buffer
        movzbl  -1(%rdi, %rdx, 1), %ecx
        xor     (%rsi), %al             # xor in second buffer
        xor     -1(%rsi, %rdx, 1), %cl
        or      %ecx, %eax              # mismatch in any of the two?
        ret

.L0304: movzwl  (%rdi), %eax
        movzwl  -2(%rdi, %rdx, 1), %ecx
        xor     (%rsi), %ax
        xor     -2(%rsi, %rdx, 1), %cx
        or      %ecx, %eax
        ret

.L0508: mov     (%rdi), %eax
        mov     -4(%rdi, %rdx, 1), %ecx
        xor     (%rsi), %eax
        xor     -4(%rsi, %rdx, 1), %ecx
        or      %ecx, %eax
        ret

.L0916: mov     (%rdi), %rax
        mov     -8(%rdi, %rdx, 1), %rcx
        xor     (%rsi), %rax
        xor     -8(%rsi, %rdx, 1), %rcx
        or      %rcx, %rax
        setnz   %al                     # ensure EAX nonzero even if only
        ret                             # high bits of RAX were set

        /* more than 16 bytes: process buffer in a loop */
.Lgt16: mov     (%rdi), %rax            # process first 16 bytes
        mov     8(%rdi), %r9
        mov     $32, %ecx
        xor     (%rsi), %rax
        xor     8(%rsi), %r9
        or      %r9, %rax

        cmp     %rdx, %rcx              # enough left for a full iteration?
        jae     .Ltail

        /* main loop processing 16 bytes per iteration */
        ALIGN_TEXT
0:      mov     -16(%rdi, %rcx, 1), %r8
        mov     -8(%rdi, %rcx, 1), %r9
        xor     -16(%rsi, %rcx, 1), %r8
        xor     -8(%rsi, %rcx, 1), %r9
        add     $16, %rcx
        or      %r9, %r8
        or      %r8, %rax

        cmp     %rdx, %rcx
        jb      0b

        /* process last 16 bytes */
.Ltail: mov     -16(%rdi, %rdx, 1), %r8
        mov     -8(%rdi, %rdx, 1), %r9
        xor     -16(%rsi, %rdx, 1), %r8
        xor     -8(%rsi, %rdx, 1), %r9
        or      %r9, %r8
        or      %r8, %rax
        setnz   %al
        ret
ARCHEND(timingsafe_bcmp, scalar)

ARCHENTRY(timingsafe_bcmp, baseline)
        cmp     $32, %rdx               # at least 33 bytes to process?
        ja      .Lgt32b

        cmp     $16, %edx               # at least 17 bytes to process?
        ja      .L1732b

        cmp     $8, %edx                # at least 9 bytes to process?
        ja      .L0916b

        cmp     $4, %edx                # at least 5 bytes to process?
        ja      .L0508b

        cmp     $2, %edx                # at least 3 bytes to process?
        ja      .L0304b

        test    %edx, %edx              # buffer empty?
        jnz     .L0102b

        xor     %eax, %eax              # empty buffer always matches
        ret

.L0102b:
        movzbl  (%rdi), %eax            # load 1--2 bytes from first buffer
        movzbl  -1(%rdi, %rdx, 1), %ecx
        xor     (%rsi), %al             # xor in second buffer
        xor     -1(%rsi, %rdx, 1), %cl
        or      %ecx, %eax              # mismatch in any of the two?
        ret

.L0304b:
        movzwl  (%rdi), %eax
        movzwl  -2(%rdi, %rdx, 1), %ecx
        xor     (%rsi), %ax
        xor     -2(%rsi, %rdx, 1), %cx
        or      %ecx, %eax
        ret

.L0508b:
        mov     (%rdi), %eax
        mov     -4(%rdi, %rdx, 1), %ecx
        xor     (%rsi), %eax
        xor     -4(%rsi, %rdx, 1), %ecx
        or      %ecx, %eax
        ret

.L0916b:
        mov     (%rdi), %rax
        mov     -8(%rdi, %rdx, 1), %rcx
        xor     (%rsi), %rax
        xor     -8(%rsi, %rdx, 1), %rcx
        or      %rcx, %rax
        setnz   %al                     # ensure EAX nonzero even if only
        ret                             # high bits of RAX were set

.L1732b:
        movdqu          (%rdi), %xmm0
        movdqu          (%rsi), %xmm2
        movdqu          -16(%rdi, %rdx, 1), %xmm1
        movdqu          -16(%rsi, %rdx, 1), %xmm3
        pcmpeqb         %xmm2, %xmm0
        pcmpeqb         %xmm3, %xmm1
        pand            %xmm1, %xmm0
        pmovmskb        %xmm0, %eax     # 1 where equal
        xor             $0xffff, %eax   # 1 where not equal
        ret

        /* more than 32 bytes: process buffer in a loop */
.Lgt32b:
        movdqu          (%rdi), %xmm4
        movdqu          (%rsi), %xmm2
        movdqu          16(%rdi), %xmm1
        movdqu          16(%rsi), %xmm3
        mov             $64, %ecx
        pcmpeqb         %xmm2, %xmm4
        pcmpeqb         %xmm3, %xmm1
        pand            %xmm1, %xmm4
        cmp             %rdx, %rcx      # enough left for a full iteration?
        jae             .Ltailb

        /* main loop processing 32 bytes per iteration */
        ALIGN_TEXT
0:      movdqu          -32(%rdi, %rcx, 1), %xmm0
        movdqu          -32(%rsi, %rcx, 1), %xmm2
        movdqu          -16(%rdi, %rcx, 1), %xmm1
        movdqu          -16(%rsi, %rcx, 1), %xmm3
        add             $32, %rcx
        pcmpeqb         %xmm2, %xmm0
        pcmpeqb         %xmm3, %xmm1
        pand            %xmm1, %xmm0
        pand            %xmm0, %xmm4
        cmp             %rdx, %rcx
        jb              0b

        /* process last 32 bytes */
.Ltailb:
        movdqu          -32(%rdi, %rdx, 1), %xmm0
        movdqu          -32(%rsi, %rdx, 1), %xmm2
        movdqu          -16(%rdi, %rdx, 1), %xmm1
        movdqu          -16(%rsi, %rdx, 1), %xmm3
        pcmpeqb         %xmm2, %xmm0
        pcmpeqb         %xmm3, %xmm1
        pand            %xmm1, %xmm0
        pand            %xmm4, %xmm0
        pmovmskb        %xmm0, %eax
        xor             $0xffff, %eax
        ret
ARCHEND(timingsafe_bcmp, baseline)

        .section .note.GNU-stack,"",%progbits