[PATCH net-next v9 05/19] zinc: ChaCha20 ARM and ARM64 implementations

@@ -4,7 +4,9 @@ ccflags-$(CONFIG_ZINC_DEBUG) += -DDEBUG
 
 zinc_chacha20-y := chacha20/chacha20.o
 zinc_chacha20-$(CONFIG_ZINC_ARCH_X86_64) += chacha20/chacha20-x86_64.o
-perlasm-targets += chacha20/chacha20-x86_64.S
+zinc_chacha20-$(CONFIG_ZINC_ARCH_ARM) += chacha20/chacha20-arm.o chacha20/chacha20-unrolled-arm.o
+zinc_chacha20-$(CONFIG_ZINC_ARCH_ARM64) += chacha20/chacha20-arm64.o
+perlasm-targets += chacha20/chacha20-x86_64.S chacha20/chacha20-arm.S chacha20/chacha20-arm64.S
 obj-$(CONFIG_ZINC_CHACHA20) += zinc_chacha20.o
 
 quiet_cmd_perlasm = PERLASM $@

@@ -0,0 +1,98 @@
+// SPDX-License-Identifier: GPL-2.0 OR MIT
+/*
+ * Copyright (C) 2015-2019 Jason A. Donenfeld <Jason@zx2c4.com>. All Rights Reserved.
+ */
+
+#include <asm/hwcap.h>
+#include <asm/neon.h>
+#if defined(CONFIG_ZINC_ARCH_ARM)
+#include <asm/system_info.h>
+#include <asm/cputype.h>
+#endif
+
+asmlinkage void chacha20_arm(u8 *out, const u8 *in, const size_t len,
+			     const u32 key[8], const u32 counter[4]);
+asmlinkage void hchacha20_arm(const u32 state[16], u32 out[8]);
+asmlinkage void chacha20_neon(u8 *out, const u8 *in, const size_t len,
+			      const u32 key[8], const u32 counter[4]);
+
+static bool chacha20_use_neon __ro_after_init;
+static bool *const chacha20_nobs[] __initconst = { &chacha20_use_neon };
+static void __init chacha20_fpu_init(void)
+{
+#if defined(CONFIG_ZINC_ARCH_ARM64)
+	chacha20_use_neon = elf_hwcap & HWCAP_ASIMD;
+#elif defined(CONFIG_ZINC_ARCH_ARM)
+	switch (read_cpuid_part()) {
+	case ARM_CPU_PART_CORTEX_A7:
+	case ARM_CPU_PART_CORTEX_A5:
+		/* The Cortex-A7 and Cortex-A5 do not perform well with the NEON
+		 * implementation but do incredibly with the scalar one and use
+		 * less power.
+		 */
+		break;
+	default:
+		chacha20_use_neon = elf_hwcap & HWCAP_NEON;
+	}
+#endif
+}
+
+static inline bool chacha20_arch(struct chacha20_ctx *ctx, u8 *dst,
+				 const u8 *src, size_t len,
+				 simd_context_t *simd_context)
+{
+	/* SIMD disables preemption, so relax after processing each page. */
+	BUILD_BUG_ON(PAGE_SIZE < CHACHA20_BLOCK_SIZE ||
+		     PAGE_SIZE % CHACHA20_BLOCK_SIZE);
+
+	for (;;) {
+		if (IS_ENABLED(CONFIG_KERNEL_MODE_NEON) && chacha20_use_neon &&
+		    len >= CHACHA20_BLOCK_SIZE * 3 && simd_use(simd_context)) {
+			const size_t bytes = min_t(size_t, len, PAGE_SIZE);
+
+			chacha20_neon(dst, src, bytes, ctx->key, ctx->counter);
+			ctx->counter[0] += (bytes + 63) / 64;
+			len -= bytes;
+			if (!len)
+				break;
+			dst += bytes;
+			src += bytes;
+			simd_relax(simd_context);
+		} else {
+			chacha20_arm(dst, src, len, ctx->key, ctx->counter);
+			ctx->counter[0] += (len + 63) / 64;
+			break;
+		}
+	}
+
+	return true;
+}
+
+static inline bool hchacha20_arch(u32 derived_key[CHACHA20_KEY_WORDS],
+				  const u8 nonce[HCHACHA20_NONCE_SIZE],
+				  const u8 key[HCHACHA20_KEY_SIZE],
+				  simd_context_t *simd_context)
+{
+	if (IS_ENABLED(CONFIG_ZINC_ARCH_ARM)) {
+		u32 x[] = { CHACHA20_CONSTANT_EXPA,
+			    CHACHA20_CONSTANT_ND_3,
+			    CHACHA20_CONSTANT_2_BY,
+			    CHACHA20_CONSTANT_TE_K,
+			    get_unaligned_le32(key + 0),
+			    get_unaligned_le32(key + 4),
+			    get_unaligned_le32(key + 8),
+			    get_unaligned_le32(key + 12),
+			    get_unaligned_le32(key + 16),
+			    get_unaligned_le32(key + 20),
+			    get_unaligned_le32(key + 24),
+			    get_unaligned_le32(key + 28),
+			    get_unaligned_le32(nonce + 0),
+			    get_unaligned_le32(nonce + 4),
+			    get_unaligned_le32(nonce + 8),
+			    get_unaligned_le32(nonce + 12)
+			  };
+		hchacha20_arm(x, derived_key);
+		return true;
+	}
+	return false;
+}

@@ -0,0 +1,1227 @@
+#!/usr/bin/env perl
+# SPDX-License-Identifier: GPL-2.0 OR BSD-3-Clause
+#
+# This code is taken from the OpenSSL project but the author, Andy Polyakov,
+# has relicensed it under the licenses specified in the SPDX header above.
+# The original headers, including the original license headers, are
+# included below for completeness.
+#
+# ====================================================================
+# Written by Andy Polyakov <appro@openssl.org> for the OpenSSL
+# project. The module is, however, dual licensed under OpenSSL and
+# CRYPTOGAMS licenses depending on where you obtain it. For further
+# details see http://www.openssl.org/~appro/cryptogams/.
+# ====================================================================
+#
+# December 2014
+#
+# ChaCha20 for ARMv4.
+#
+# September 2018
+#
+# Improve scalar performance per Eric Biggers' suggestion to eliminate
+# separate rotates. This requires b[0..3] and d[0..3] to be maintained
+# pre-rotated, hence odd twists prior inner loop and when accumulating
+# key material. Since amount of instructions is reduced as result, even
+# NEON performance is improved somewhat, most notably by ~9% on low-end
+# Cortex-A5/A7. Full unroll was shown to provide even better scalar
+# performance on Cortex-A5/A7, naturally at the cost of manyfold size
+# increase. We let it be. Oversized code works in benchmarks, but is not
+# necessarily optimal in real life, when it's likely to be out-of-cache
+# upon entry and evict significant part of cache upon completion.
+#
+# Performance in cycles per byte out of large buffer.
+#
+#			IALU/gcc-4.4    1xNEON      3xNEON+1xIALU
+#
+# Cortex-A5		14.2(*)/+160%   21.8        12.9(**)
+# Cortex-A8		10.2(*)/+190%   13.9        6.10
+# Cortex-A9		10.8(*)/+150%   14.3        6.50
+# Cortex-A15		11.0/+40%       16.0        4.90
+# Snapdragon S4		13.9(***)/+90%  13.6        4.90
+#
+# (*)	most "favourable" result for aligned data on little-endian
+#	processor, result for misaligned data is 10-15% lower;
+# (**)	pure 4xNEON [with "vertical" layout] was shown to provide ~8%
+#	better performance on Cortex-A5/A7, but not on others;
+# (***)	it's 17% slower than original, trade-off is considered
+#	acceptable, because of improvement on others, specifically
+#	+36% on Cortex-A5/A7 and +20% on Cortex-A9;
+
+$flavour = shift;
+if ($flavour=~/\w[\w\-]*\.\w+$/) { $output=$flavour; undef $flavour; }
+else { while (($output=shift) && ($output!~/\w[\w\-]*\.\w+$/)) {} }
+
+if ($flavour && $flavour ne "void") {
+    $0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
+    ( $xlate="${dir}arm-xlate.pl" and -f $xlate ) or
+    ( $xlate="${dir}../../perlasm/arm-xlate.pl" and -f $xlate) or
+    die "can't locate arm-xlate.pl";
+
+    open STDOUT,"| \"$^X\" $xlate $flavour $output";
+} else {
+    open STDOUT,">$output";
+}
+
+sub AUTOLOAD()		# thunk [simplified] x86-style perlasm
+{ my $opcode = $AUTOLOAD; $opcode =~ s/.*:://; $opcode =~ s/_/\./;
+  my $arg = pop;
+    $arg = "#$arg" if ($arg*1 eq $arg);
+    $code .= "\t$opcode\t".join(',',@_,$arg)."\n";
+}
+
+my @x=map("r$_",(0..7,"x","x","x","x",12,"x",14,"x"));
+my @t=map("r$_",(8..11));
+
+sub ROUND {
+my ($a0,$b0,$c0,$d0)=@_;
+my ($a1,$b1,$c1,$d1)=map(($_&~3)+(($_+1)&3),($a0,$b0,$c0,$d0));
+my ($a2,$b2,$c2,$d2)=map(($_&~3)+(($_+1)&3),($a1,$b1,$c1,$d1));
+my ($a3,$b3,$c3,$d3)=map(($_&~3)+(($_+1)&3),($a2,$b2,$c2,$d2));
+my $odd = $d0&1;
+my ($xc,$xc_) = (@t[0..1]);
+my ($xd,$xd_) = $odd ? (@t[2],@x[$d1]) : (@x[$d0],@t[2]);
+my @ret;
+
+	# Consider order in which variables are addressed by their
+	# index:
+	#
+	#       a   b   c   d
+	#
+	#       0   4   8  12 < even round
+	#       1   5   9  13
+	#       2   6  10  14
+	#       3   7  11  15
+	#       0   5  10  15 < odd round
+	#       1   6  11  12
+	#       2   7   8  13
+	#       3   4   9  14
+	#
+	# 'a', 'b' are permanently allocated in registers, @x[0..7],
+	# while 'c's and pair of 'd's are maintained in memory. If
+	# you observe 'c' column, you'll notice that pair of 'c's is
+	# invariant between rounds. This means that we have to reload
+	# them once per round, in the middle. This is why you'll see
+	# bunch of 'c' stores and loads in the middle, but none in
+	# the beginning or end. If you observe 'd' column, you'll
+	# notice that 15 and 13 are reused in next pair of rounds.
+	# This is why these two are chosen for offloading to memory,
+	# to make loads count more.
+							push @ret,(
+	"&add	(@x[$a0],@x[$a0],@x[$b0],'ror#13')",
+	 "&add	(@x[$a1],@x[$a1],@x[$b1],'ror#13')",
+	"&eor	($xd,@x[$a0],$xd,'ror#24')",
+	 "&eor	($xd_,@x[$a1],$xd_,'ror#24')",
+
+	"&add	($xc,$xc,$xd,'ror#16')",
+	 "&add	($xc_,$xc_,$xd_,'ror#16')",
+	"&eor	(@x[$b0],$xc, @x[$b0],'ror#13')",
+	 "&eor	(@x[$b1],$xc_,@x[$b1],'ror#13')",
+
+	"&add	(@x[$a0],@x[$a0],@x[$b0],'ror#20')",
+	 "&add	(@x[$a1],@x[$a1],@x[$b1],'ror#20')",
+	"&eor	($xd,@x[$a0],$xd,'ror#16')",
+	 "&eor	($xd_,@x[$a1],$xd_,'ror#16')"		);
+							push @ret,(
+	"&str	($xd,'[sp,#4*(16+$d0)]')"		) if ($odd);
+							push @ret,(
+	"&add	($xc,$xc,$xd,'ror#24')"			);
+							push @ret,(
+	"&ldr	($xd,'[sp,#4*(16+$d2)]')"		) if ($odd);
+							push @ret,(
+	 "&str	($xd_,'[sp,#4*(16+$d1)]')"		) if (!$odd);
+							push @ret,(
+	 "&add	($xc_,$xc_,$xd_,'ror#24')"		);
+							push @ret,(
+	 "&ldr	($xd_,'[sp,#4*(16+$d3)]')"		) if (!$odd);
+							push @ret,(
+	"&str	($xc,'[sp,#4*(16+$c0)]')",
+	"&eor	(@x[$b0],@x[$b0],$xc,'ror#12')",
+	 "&str	($xc_,'[sp,#4*(16+$c1)]')",
+	 "&eor	(@x[$b1],@x[$b1],$xc_,'ror#12')"	);
+
+	$xd=@x[$d2]					if (!$odd);
+	$xd_=@x[$d3]					if ($odd);
+							push @ret,(
+	"&ldr	($xc,'[sp,#4*(16+$c2)]')",
+	"&add	(@x[$a2],@x[$a2],@x[$b2],'ror#13')",
+	 "&ldr	($xc_,'[sp,#4*(16+$c3)]')",
+	 "&add	(@x[$a3],@x[$a3],@x[$b3],'ror#13')",
+	"&eor	($xd,@x[$a2],$xd,'ror#24')",
+	 "&eor	($xd_,@x[$a3],$xd_,'ror#24')",
+
+	"&add	($xc,$xc,$xd,'ror#16')",
+	 "&add	($xc_,$xc_,$xd_,'ror#16')",
+	"&eor	(@x[$b2],$xc, @x[$b2],'ror#13')",
+	 "&eor	(@x[$b3],$xc_,@x[$b3],'ror#13')",
+
+	"&add	(@x[$a2],@x[$a2],@x[$b2],'ror#20')",
+	 "&add	(@x[$a3],@x[$a3],@x[$b3],'ror#20')",
+	"&eor	($xd,@x[$a2],$xd,'ror#16')",
+	 "&eor	($xd_,@x[$a3],$xd_,'ror#16')",
+
+	"&add	($xc,$xc,$xd,'ror#24')",
+	 "&add	($xc_,$xc_,$xd_,'ror#24')",
+	"&eor	(@x[$b2],@x[$b2],$xc,'ror#12')",
+	 "&eor	(@x[$b3],@x[$b3],$xc_,'ror#12')"	);
+
+	@ret;
+}
+
+$code.=<<___;
+#ifndef __KERNEL__
+# include "arm_arch.h"
+#else
+# define __ARM_ARCH__ __LINUX_ARM_ARCH__
+# define __ARM_MAX_ARCH__ __LINUX_ARM_ARCH__
+# define ChaCha20_ctr32 chacha20_arm_cryptogams
+# define ChaCha20_neon  chacha20_neon
+#endif
+
+.text
+#if defined(__thumb2__) || defined(__clang__)
+.syntax	unified
+# define ldrhsb	ldrbhs
+#endif
+#if defined(__thumb2__)
+.thumb
+#else
+.code	32
+#endif
+
+.align	5
+.Lsigma:
+.long	0x61707865,0x3320646e,0x79622d32,0x6b206574	@ endian-neutral
+.Lone:
+.long	1,0,0,0
+.Lrot8:
+.long	0x02010003,0x06050407
+#if __ARM_MAX_ARCH__>=7 && !defined(__KERNEL__)
+.LOPENSSL_armcap:
+.word   OPENSSL_armcap_P-.LChaCha20_ctr32
+#else
+.word	-1
+#endif
+
+.globl	ChaCha20_ctr32
+.type	ChaCha20_ctr32,%function
+.align	5
+ChaCha20_ctr32:
+.LChaCha20_ctr32:
+	ldr	r12,[sp,#0]		@ pull pointer to counter and nonce
+	stmdb	sp!,{r0-r2,r4-r11,lr}
+#if __ARM_ARCH__<7 && !defined(__thumb2__)
+	sub	r14,pc,#16		@ ChaCha20_ctr32
+#else
+	adr	r14,.LChaCha20_ctr32
+#endif
+	cmp	r2,#0			@ len==0?
+#ifdef	__thumb2__
+	itt	eq
+#endif
+	addeq	sp,sp,#4*3
+	beq	.Lno_data
+#if __ARM_MAX_ARCH__>=7 && !defined(__KERNEL__)
+	cmp	r2,#192			@ test len
+	bls	.Lshort
+	ldr	r4,[r14,#-24]
+	ldr	r4,[r14,r4]
+# ifdef	__APPLE__
+	ldr	r4,[r4]
+# endif
+	tst	r4,#ARMV7_NEON
+	bne	.LChaCha20_neon
+.Lshort:
+#endif
+	ldmia	r12,{r4-r7}		@ load counter and nonce
+	sub	sp,sp,#4*(16)		@ off-load area
+	sub	r14,r14,#64		@ .Lsigma
+	stmdb	sp!,{r4-r7}		@ copy counter and nonce
+	ldmia	r3,{r4-r11}		@ load key
+	ldmia	r14,{r0-r3}		@ load sigma
+	stmdb	sp!,{r4-r11}		@ copy key
+	stmdb	sp!,{r0-r3}		@ copy sigma
+	str	r10,[sp,#4*(16+10)]	@ off-load "@x[10]"
+	str	r11,[sp,#4*(16+11)]	@ off-load "@x[11]"
+	b	.Loop_outer_enter
+
+.align	4
+.Loop_outer:
+	ldmia	sp,{r0-r9}		@ load key material
+	str	@t[3],[sp,#4*(32+2)]	@ save len
+	str	r12,  [sp,#4*(32+1)]	@ save inp
+	str	r14,  [sp,#4*(32+0)]	@ save out
+.Loop_outer_enter:
+	ldr	@t[3], [sp,#4*(15)]
+	 mov	@x[4],@x[4],ror#19	@ twist b[0..3]
+	ldr	@x[12],[sp,#4*(12)]	@ modulo-scheduled load
+	 mov	@x[5],@x[5],ror#19
+	ldr	@t[2], [sp,#4*(13)]
+	 mov	@x[6],@x[6],ror#19
+	ldr	@x[14],[sp,#4*(14)]
+	 mov	@x[7],@x[7],ror#19
+	mov	@t[3],@t[3],ror#8	@ twist d[0..3]
+	mov	@x[12],@x[12],ror#8
+	mov	@t[2],@t[2],ror#8
+	mov	@x[14],@x[14],ror#8
+	str	@t[3], [sp,#4*(16+15)]
+	mov	@t[3],#10
+	b	.Loop
+
+.align	4
+.Loop:
+	subs	@t[3],@t[3],#1
+___
+	foreach (&ROUND(0, 4, 8,12)) { eval; }
+	foreach (&ROUND(0, 5,10,15)) { eval; }
+$code.=<<___;
+	bne	.Loop
+
+	ldr	@t[3],[sp,#4*(32+2)]	@ load len
+
+	str	@t[0], [sp,#4*(16+8)]	@ modulo-scheduled store
+	str	@t[1], [sp,#4*(16+9)]
+	str	@x[12],[sp,#4*(16+12)]
+	str	@t[2], [sp,#4*(16+13)]
+	str	@x[14],[sp,#4*(16+14)]
+
+	@ at this point we have first half of 512-bit result in
+	@ @x[0-7] and second half at sp+4*(16+8)
+
+	cmp	@t[3],#64		@ done yet?
+#ifdef	__thumb2__
+	itete	lo
+#endif
+	addlo	r12,sp,#4*(0)		@ shortcut or ...
+	ldrhs	r12,[sp,#4*(32+1)]	@ ... load inp
+	addlo	r14,sp,#4*(0)		@ shortcut or ...
+	ldrhs	r14,[sp,#4*(32+0)]	@ ... load out
+
+	ldr	@t[0],[sp,#4*(0)]	@ load key material
+	ldr	@t[1],[sp,#4*(1)]
+
+#if __ARM_ARCH__>=6 || !defined(__ARMEB__)
+# if __ARM_ARCH__<7
+	orr	@t[2],r12,r14
+	tst	@t[2],#3		@ are input and output aligned?
+	ldr	@t[2],[sp,#4*(2)]
+	bne	.Lunaligned
+	cmp	@t[3],#64		@ restore flags
+# else
+	ldr	@t[2],[sp,#4*(2)]
+# endif
+	ldr	@t[3],[sp,#4*(3)]
+
+	add	@x[0],@x[0],@t[0]	@ accumulate key material
+	add	@x[1],@x[1],@t[1]
+# ifdef	__thumb2__
+	itt	hs
+# endif
+	ldrhs	@t[0],[r12],#16		@ load input
+	ldrhs	@t[1],[r12,#-12]
+
+	add	@x[2],@x[2],@t[2]
+	add	@x[3],@x[3],@t[3]
+# ifdef	__thumb2__
+	itt	hs
+# endif
+	ldrhs	@t[2],[r12,#-8]
+	ldrhs	@t[3],[r12,#-4]
+# if __ARM_ARCH__>=6 && defined(__ARMEB__)
+	rev	@x[0],@x[0]
+	rev	@x[1],@x[1]
+	rev	@x[2],@x[2]
+	rev	@x[3],@x[3]
+# endif
+# ifdef	__thumb2__
+	itt	hs
+# endif
+	eorhs	@x[0],@x[0],@t[0]	@ xor with input
+	eorhs	@x[1],@x[1],@t[1]
+	 add	@t[0],sp,#4*(4)
+	str	@x[0],[r14],#16		@ store output
+# ifdef	__thumb2__
+	itt	hs
+# endif
+	eorhs	@x[2],@x[2],@t[2]
+	eorhs	@x[3],@x[3],@t[3]
+	 ldmia	@t[0],{@t[0]-@t[3]}	@ load key material
+	str	@x[1],[r14,#-12]
+	str	@x[2],[r14,#-8]
+	str	@x[3],[r14,#-4]
+
+	add	@x[4],@t[0],@x[4],ror#13 @ accumulate key material
+	add	@x[5],@t[1],@x[5],ror#13
+# ifdef	__thumb2__
+	itt	hs
+# endif
+	ldrhs	@t[0],[r12],#16		@ load input
+	ldrhs	@t[1],[r12,#-12]
+	add	@x[6],@t[2],@x[6],ror#13
+	add	@x[7],@t[3],@x[7],ror#13
+# ifdef	__thumb2__
+	itt	hs
+# endif
+	ldrhs	@t[2],[r12,#-8]
+	ldrhs	@t[3],[r12,#-4]
+# if __ARM_ARCH__>=6 && defined(__ARMEB__)
+	rev	@x[4],@x[4]
+	rev	@x[5],@x[5]
+	rev	@x[6],@x[6]
+	rev	@x[7],@x[7]
+# endif
+# ifdef	__thumb2__
+	itt	hs
+# endif
+	eorhs	@x[4],@x[4],@t[0]
+	eorhs	@x[5],@x[5],@t[1]
+	 add	@t[0],sp,#4*(8)
+	str	@x[4],[r14],#16		@ store output
+# ifdef	__thumb2__
+	itt	hs
+# endif
+	eorhs	@x[6],@x[6],@t[2]
+	eorhs	@x[7],@x[7],@t[3]
+	str	@x[5],[r14,#-12]
+	 ldmia	@t[0],{@t[0]-@t[3]}	@ load key material
+	str	@x[6],[r14,#-8]
+	 add	@x[0],sp,#4*(16+8)
+	str	@x[7],[r14,#-4]
+
+	ldmia	@x[0],{@x[0]-@x[7]}	@ load second half
+
+	add	@x[0],@x[0],@t[0]	@ accumulate key material
+	add	@x[1],@x[1],@t[1]
+# ifdef	__thumb2__
+	itt	hs
+# endif
+	ldrhs	@t[0],[r12],#16		@ load input
+	ldrhs	@t[1],[r12,#-12]
+# ifdef	__thumb2__
+	itt	hi
+# endif
+	 strhi	@t[2],[sp,#4*(16+10)]	@ copy "@x[10]" while at it
+	 strhi	@t[3],[sp,#4*(16+11)]	@ copy "@x[11]" while at it
+	add	@x[2],@x[2],@t[2]
+	add	@x[3],@x[3],@t[3]
+# ifdef	__thumb2__
+	itt	hs
+# endif
+	ldrhs	@t[2],[r12,#-8]
+	ldrhs	@t[3],[r12,#-4]
+# if __ARM_ARCH__>=6 && defined(__ARMEB__)
+	rev	@x[0],@x[0]
+	rev	@x[1],@x[1]
+	rev	@x[2],@x[2]
+	rev	@x[3],@x[3]
+# endif
+# ifdef	__thumb2__
+	itt	hs
+# endif
+	eorhs	@x[0],@x[0],@t[0]
+	eorhs	@x[1],@x[1],@t[1]
+	 add	@t[0],sp,#4*(12)
+	str	@x[0],[r14],#16		@ store output
+# ifdef	__thumb2__
+	itt	hs
+# endif
+	eorhs	@x[2],@x[2],@t[2]
+	eorhs	@x[3],@x[3],@t[3]
+	str	@x[1],[r14,#-12]
+	 ldmia	@t[0],{@t[0]-@t[3]}	@ load key material
+	str	@x[2],[r14,#-8]
+	str	@x[3],[r14,#-4]
+
+	add	@x[4],@t[0],@x[4],ror#24 @ accumulate key material
+	add	@x[5],@t[1],@x[5],ror#24
+# ifdef	__thumb2__
+	itt	hi
+# endif
+	 addhi	@t[0],@t[0],#1		@ next counter value
+	 strhi	@t[0],[sp,#4*(12)]	@ save next counter value
+# ifdef	__thumb2__
+	itt	hs
+# endif
+	ldrhs	@t[0],[r12],#16		@ load input
+	ldrhs	@t[1],[r12,#-12]
+	add	@x[6],@t[2],@x[6],ror#24
+	add	@x[7],@t[3],@x[7],ror#24
+# ifdef	__thumb2__
+	itt	hs
+# endif
+	ldrhs	@t[2],[r12,#-8]
+	ldrhs	@t[3],[r12,#-4]
+# if __ARM_ARCH__>=6 && defined(__ARMEB__)
+	rev	@x[4],@x[4]
+	rev	@x[5],@x[5]
+	rev	@x[6],@x[6]
+	rev	@x[7],@x[7]
+# endif
+# ifdef	__thumb2__
+	itt	hs
+# endif
+	eorhs	@x[4],@x[4],@t[0]
+	eorhs	@x[5],@x[5],@t[1]
+# ifdef	__thumb2__
+	 it	ne
+# endif
+	 ldrne	@t[0],[sp,#4*(32+2)]	@ re-load len
+# ifdef	__thumb2__
+	itt	hs
+# endif
+	eorhs	@x[6],@x[6],@t[2]
+	eorhs	@x[7],@x[7],@t[3]
+	str	@x[4],[r14],#16		@ store output
+	str	@x[5],[r14,#-12]
+# ifdef	__thumb2__
+	it	hs
+# endif
+	 subhs	@t[3],@t[0],#64		@ len-=64
+	str	@x[6],[r14,#-8]
+	str	@x[7],[r14,#-4]
+	bhi	.Loop_outer
+
+	beq	.Ldone
+# if __ARM_ARCH__<7
+	b	.Ltail
+
+.align	4
+.Lunaligned:				@ unaligned endian-neutral path
+	cmp	@t[3],#64		@ restore flags
+# endif
+#endif
+#if __ARM_ARCH__<7
+	ldr	@t[3],[sp,#4*(3)]
+___
+for ($i=0;$i<16;$i+=4) {
+my $j=$i&0x7;
+my $twist="";
+if ($i==4)     { $twist = ",ror#13"; }
+elsif ($i==12) { $twist = ",ror#24"; }
+
+$code.=<<___	if ($i==4);
+	add	@x[0],sp,#4*(16+8)
+___
+$code.=<<___	if ($i==8);
+	ldmia	@x[0],{@x[0]-@x[7]}		@ load second half
+# ifdef	__thumb2__
+	itt	hi
+# endif
+	strhi	@t[2],[sp,#4*(16+10)]		@ copy "@x[10]"
+	strhi	@t[3],[sp,#4*(16+11)]		@ copy "@x[11]"
+___
+$code.=<<___;
+	add	@x[$j+0],@t[0],@x[$j+0]$twist	@ accumulate key material
+___
+$code.=<<___	if ($i==12);
+# ifdef	__thumb2__
+	itt	hi
+# endif
+	addhi	@t[0],@t[0],#1			@ next counter value
+	strhi	@t[0],[sp,#4*(12)]		@ save next counter value
+___
+$code.=<<___;
+	add	@x[$j+1],@t[1],@x[$j+1]$twist
+	add	@x[$j+2],@t[2],@x[$j+2]$twist
+# ifdef	__thumb2__
+	itete	lo
+# endif
+	eorlo	@t[0],@t[0],@t[0]		@ zero or ...
+	ldrhsb	@t[0],[r12],#16			@ ... load input
+	eorlo	@t[1],@t[1],@t[1]
+	ldrhsb	@t[1],[r12,#-12]
+
+	add	@x[$j+3],@t[3],@x[$j+3]$twist
+# ifdef	__thumb2__
+	itete	lo
+# endif
+	eorlo	@t[2],@t[2],@t[2]
+	ldrhsb	@t[2],[r12,#-8]
+	eorlo	@t[3],@t[3],@t[3]
+	ldrhsb	@t[3],[r12,#-4]
+
+	eor	@x[$j+0],@t[0],@x[$j+0]		@ xor with input (or zero)
+	eor	@x[$j+1],@t[1],@x[$j+1]
+# ifdef	__thumb2__
+	itt	hs
+# endif
+	ldrhsb	@t[0],[r12,#-15]		@ load more input
+	ldrhsb	@t[1],[r12,#-11]
+	eor	@x[$j+2],@t[2],@x[$j+2]
+	 strb	@x[$j+0],[r14],#16		@ store output
+	eor	@x[$j+3],@t[3],@x[$j+3]
+# ifdef	__thumb2__
+	itt	hs
+# endif
+	ldrhsb	@t[2],[r12,#-7]
+	ldrhsb	@t[3],[r12,#-3]
+	 strb	@x[$j+1],[r14,#-12]
+	eor	@x[$j+0],@t[0],@x[$j+0],lsr#8
+	 strb	@x[$j+2],[r14,#-8]
+	eor	@x[$j+1],@t[1],@x[$j+1],lsr#8
+# ifdef	__thumb2__
+	itt	hs
+# endif
+	ldrhsb	@t[0],[r12,#-14]		@ load more input
+	ldrhsb	@t[1],[r12,#-10]
+	 strb	@x[$j+3],[r14,#-4]
+	eor	@x[$j+2],@t[2],@x[$j+2],lsr#8
+	 strb	@x[$j+0],[r14,#-15]
+	eor	@x[$j+3],@t[3],@x[$j+3],lsr#8
+# ifdef	__thumb2__
+	itt	hs
+# endif
+	ldrhsb	@t[2],[r12,#-6]
+	ldrhsb	@t[3],[r12,#-2]
+	 strb	@x[$j+1],[r14,#-11]
+	eor	@x[$j+0],@t[0],@x[$j+0],lsr#8
+	 strb	@x[$j+2],[r14,#-7]
+	eor	@x[$j+1],@t[1],@x[$j+1],lsr#8
+# ifdef	__thumb2__
+	itt	hs
+# endif
+	ldrhsb	@t[0],[r12,#-13]		@ load more input
+	ldrhsb	@t[1],[r12,#-9]
+	 strb	@x[$j+3],[r14,#-3]
+	eor	@x[$j+2],@t[2],@x[$j+2],lsr#8
+	 strb	@x[$j+0],[r14,#-14]
+	eor	@x[$j+3],@t[3],@x[$j+3],lsr#8
+# ifdef	__thumb2__
+	itt	hs
+# endif
+	ldrhsb	@t[2],[r12,#-5]
+	ldrhsb	@t[3],[r12,#-1]
+	 strb	@x[$j+1],[r14,#-10]
+	 strb	@x[$j+2],[r14,#-6]
+	eor	@x[$j+0],@t[0],@x[$j+0],lsr#8
+	 strb	@x[$j+3],[r14,#-2]
+	eor	@x[$j+1],@t[1],@x[$j+1],lsr#8
+	 strb	@x[$j+0],[r14,#-13]
+	eor	@x[$j+2],@t[2],@x[$j+2],lsr#8
+	 strb	@x[$j+1],[r14,#-9]
+	eor	@x[$j+3],@t[3],@x[$j+3],lsr#8
+	 strb	@x[$j+2],[r14,#-5]
+	 strb	@x[$j+3],[r14,#-1]
+___
+$code.=<<___	if ($i<12);
+	add	@t[0],sp,#4*(4+$i)
+	ldmia	@t[0],{@t[0]-@t[3]}		@ load key material
+___
+}
+$code.=<<___;
+# ifdef	__thumb2__
+	it	ne
+# endif
+	ldrne	@t[0],[sp,#4*(32+2)]		@ re-load len
+# ifdef	__thumb2__
+	it	hs
+# endif
+	subhs	@t[3],@t[0],#64			@ len-=64
+	bhi	.Loop_outer
+
+	beq	.Ldone
+#endif
+
+.Ltail:
+	ldr	r12,[sp,#4*(32+1)]	@ load inp
+	add	@t[1],sp,#4*(0)
+	ldr	r14,[sp,#4*(32+0)]	@ load out
+
+.Loop_tail:
+	ldrb	@t[2],[@t[1]],#1	@ read buffer on stack
+	ldrb	@t[3],[r12],#1		@ read input
+	subs	@t[0],@t[0],#1
+	eor	@t[3],@t[3],@t[2]
+	strb	@t[3],[r14],#1		@ store output
+	bne	.Loop_tail
+
+.Ldone:
+	add	sp,sp,#4*(32+3)
+.Lno_data:
+#if __ARM_ARCH__>=5
+	ldmia	sp!,{r4-r11,pc}
+#else
+	ldmia	sp!,{r4-r12,lr}
+	tst	lr,#1
+	moveq	pc,lr			@ be binary compatible with V4, yet
+	.long	0xe12fff1e		@ interoperable with Thumb ISA:-)
+#endif
+.size	ChaCha20_ctr32,.-ChaCha20_ctr32
+___
+
+{{{
+my ($a0,$b0,$c0,$d0,$a1,$b1,$c1,$d1,$a2,$b2,$c2,$d2,$t0,$t1,$t2,$t3) =
+    map("q$_",(0..15));
+
+# This can replace vshr-by-24+vsli-by-8. It gives ~3% improvement on
+# Cortex-A5/A7, but hurts Cortex-A9 by 5% and Snapdragon S4 by 14%!
+sub vperm()
+{ my ($dst,$src,$tbl) = @_;
+    $code .= "	vtbl.8	$dst#lo,{$src#lo},$tbl#lo\n";
+    $code .= "	vtbl.8	$dst#hi,{$src#hi},$tbl#lo\n";
+}
+
+sub NEONROUND {
+my $odd = pop;
+my ($a,$b,$c,$d,$t)=@_;
+
+	(
+	"&vadd_i32	($a,$a,$b)",
+	"&veor		($d,$d,$a)",
+	"&vrev32_16	($d,$d)",	# vrot ($d,16)
+
+	"&vadd_i32	($c,$c,$d)",
+	"&veor		($t,$b,$c)",
+	"&vshr_u32	($b,$t,20)",
+	"&vsli_32	($b,$t,12)",
+
+	"&vadd_i32	($a,$a,$b)",
+	"&veor		($t,$d,$a)",
+	"&vshr_u32	($d,$t,24)",
+	"&vsli_32	($d,$t,8)",
+	#"&vperm	($d,$t,$t3)",
+
+	"&vadd_i32	($c,$c,$d)",
+	"&veor		($t,$b,$c)",
+	"&vshr_u32	($b,$t,25)",
+	"&vsli_32	($b,$t,7)",
+
+	"&vext_8	($c,$c,$c,8)",
+	"&vext_8	($b,$b,$b,$odd?12:4)",
+	"&vext_8	($d,$d,$d,$odd?4:12)"
+	);
+}
+
+$code.=<<___;
+#if (defined(__KERNEL__) && defined(CONFIG_KERNEL_MODE_NEON)) || (!defined(__KERNEL__) && __ARM_MAX_ARCH__>=7)
+.arch	armv7-a
+.fpu	neon
+
+# ifdef __KERNEL__
+.globl	ChaCha20_neon
+@ For optimal performance it's appropriate for caller to enforce
+@ minimum input length, 193 bytes is suggested.
+# endif
+.type	ChaCha20_neon,%function
+.align	5
+ChaCha20_neon:
+	ldr		r12,[sp,#0]		@ pull pointer to counter and nonce
+	stmdb		sp!,{r0-r2,r4-r11,lr}
+.LChaCha20_neon:
+	adr		r14,.Lsigma
+	vstmdb		sp!,{d8-d15}		@ ABI spec says so
+	stmdb		sp!,{r0-r3}
+
+	vld1.32		{$b0-$c0},[r3]		@ load key
+	ldmia		r3,{r4-r11}		@ load key
+
+	sub		sp,sp,#4*(16+16)
+	vld1.32		{$d0},[r12]		@ load counter and nonce
+	add		r12,sp,#4*8
+	ldmia		r14,{r0-r3}		@ load sigma
+	vld1.32		{$a0},[r14]!		@ load sigma
+	vld1.32		{$t0},[r14]!		@ one
+	@ vld1.32	{$t3#lo},[r14]		@ rot8
+	vst1.32		{$c0-$d0},[r12]		@ copy 1/2key|counter|nonce
+	vst1.32		{$a0-$b0},[sp]		@ copy sigma|1/2key
+
+	str		r10,[sp,#4*(16+10)]	@ off-load "@x[10]"
+	str		r11,[sp,#4*(16+11)]	@ off-load "@x[11]"
+	vshl.i32	$t1#lo,$t0#lo,#1	@ two
+	vstr		$t0#lo,[sp,#4*(16+0)]
+	vshl.i32	$t2#lo,$t0#lo,#2	@ four
+	vstr		$t1#lo,[sp,#4*(16+2)]
+	vmov		$a1,$a0
+	vstr		$t2#lo,[sp,#4*(16+4)]
+	vmov		$a2,$a0
+	@ vstr		$t3#lo,[sp,#4*(16+6)]
+	vmov		$b1,$b0
+	vmov		$b2,$b0
+	b		.Loop_neon_enter
+
+.align	4
+.Loop_neon_outer:
+	ldmia		sp,{r0-r9}		@ load key material
+	cmp		@t[3],#64*2		@ if len<=64*2
+	bls		.Lbreak_neon		@ switch to integer-only
+	@ vldr		$t3#lo,[sp,#4*(16+6)]	@ rot8
+	vmov		$a1,$a0
+	str		@t[3],[sp,#4*(32+2)]	@ save len
+	vmov		$a2,$a0
+	str		r12,  [sp,#4*(32+1)]	@ save inp
+	vmov		$b1,$b0
+	str		r14,  [sp,#4*(32+0)]	@ save out
+	vmov		$b2,$b0
+.Loop_neon_enter:
+	ldr		@t[3], [sp,#4*(15)]
+	 mov		@x[4],@x[4],ror#19	@ twist b[0..3]
+	vadd.i32	$d1,$d0,$t0		@ counter+1
+	ldr		@x[12],[sp,#4*(12)]	@ modulo-scheduled load
+	 mov		@x[5],@x[5],ror#19
+	vmov		$c1,$c0
+	ldr		@t[2], [sp,#4*(13)]
+	 mov		@x[6],@x[6],ror#19
+	vmov		$c2,$c0
+	ldr		@x[14],[sp,#4*(14)]
+	 mov		@x[7],@x[7],ror#19
+	vadd.i32	$d2,$d1,$t0		@ counter+2
+	add		@x[12],@x[12],#3	@ counter+3
+	mov		@t[3],@t[3],ror#8	@ twist d[0..3]
+	mov		@x[12],@x[12],ror#8
+	mov		@t[2],@t[2],ror#8
+	mov		@x[14],@x[14],ror#8
+	str		@t[3], [sp,#4*(16+15)]
+	mov		@t[3],#10
+	b		.Loop_neon
+
+.align	4
+.Loop_neon:
+	subs		@t[3],@t[3],#1
+___
+	my @thread0=&NEONROUND($a0,$b0,$c0,$d0,$t0,0);
+	my @thread1=&NEONROUND($a1,$b1,$c1,$d1,$t1,0);
+	my @thread2=&NEONROUND($a2,$b2,$c2,$d2,$t2,0);
+	my @thread3=&ROUND(0,4,8,12);
+
+	foreach (@thread0) {
+		eval;			eval(shift(@thread3));
+		eval(shift(@thread1));	eval(shift(@thread3));
+		eval(shift(@thread2));	eval(shift(@thread3));
+	}
+
+	@thread0=&NEONROUND($a0,$b0,$c0,$d0,$t0,1);
+	@thread1=&NEONROUND($a1,$b1,$c1,$d1,$t1,1);
+	@thread2=&NEONROUND($a2,$b2,$c2,$d2,$t2,1);
+	@thread3=&ROUND(0,5,10,15);
+
+	foreach (@thread0) {
+		eval;			eval(shift(@thread3));
+		eval(shift(@thread1));	eval(shift(@thread3));
+		eval(shift(@thread2));	eval(shift(@thread3));
+	}
+$code.=<<___;
+	bne		.Loop_neon
+
+	add		@t[3],sp,#32
+	vld1.32		{$t0-$t1},[sp]		@ load key material
+	vld1.32		{$t2-$t3},[@t[3]]
+
+	ldr		@t[3],[sp,#4*(32+2)]	@ load len
+
+	str		@t[0], [sp,#4*(16+8)]	@ modulo-scheduled store
+	str		@t[1], [sp,#4*(16+9)]
+	str		@x[12],[sp,#4*(16+12)]
+	str		@t[2], [sp,#4*(16+13)]
+	str		@x[14],[sp,#4*(16+14)]
+
+	@ at this point we have first half of 512-bit result in
+	@ @x[0-7] and second half at sp+4*(16+8)
+
+	ldr		r12,[sp,#4*(32+1)]	@ load inp
+	ldr		r14,[sp,#4*(32+0)]	@ load out
+
+	vadd.i32	$a0,$a0,$t0		@ accumulate key material
+	vadd.i32	$a1,$a1,$t0
+	vadd.i32	$a2,$a2,$t0
+	vldr		$t0#lo,[sp,#4*(16+0)]	@ one
+
+	vadd.i32	$b0,$b0,$t1
+	vadd.i32	$b1,$b1,$t1
+	vadd.i32	$b2,$b2,$t1
+	vldr		$t1#lo,[sp,#4*(16+2)]	@ two
+
+	vadd.i32	$c0,$c0,$t2
+	vadd.i32	$c1,$c1,$t2
+	vadd.i32	$c2,$c2,$t2
+	vadd.i32	$d1#lo,$d1#lo,$t0#lo	@ counter+1
+	vadd.i32	$d2#lo,$d2#lo,$t1#lo	@ counter+2
+
+	vadd.i32	$d0,$d0,$t3
+	vadd.i32	$d1,$d1,$t3
+	vadd.i32	$d2,$d2,$t3
+
+	cmp		@t[3],#64*4
+	blo		.Ltail_neon
+
+	vld1.8		{$t0-$t1},[r12]!	@ load input
+	 mov		@t[3],sp
+	vld1.8		{$t2-$t3},[r12]!
+	veor		$a0,$a0,$t0		@ xor with input
+	veor		$b0,$b0,$t1
+	vld1.8		{$t0-$t1},[r12]!
+	veor		$c0,$c0,$t2
+	veor		$d0,$d0,$t3
+	vld1.8		{$t2-$t3},[r12]!
+
+	veor		$a1,$a1,$t0
+	 vst1.8		{$a0-$b0},[r14]!	@ store output
+	veor		$b1,$b1,$t1
+	vld1.8		{$t0-$t1},[r12]!
+	veor		$c1,$c1,$t2
+	 vst1.8		{$c0-$d0},[r14]!
+	veor		$d1,$d1,$t3
+	vld1.8		{$t2-$t3},[r12]!
+
+	veor		$a2,$a2,$t0
+	 vld1.32	{$a0-$b0},[@t[3]]!	@ load for next iteration
+	 veor		$t0#hi,$t0#hi,$t0#hi
+	 vldr		$t0#lo,[sp,#4*(16+4)]	@ four
+	veor		$b2,$b2,$t1
+	 vld1.32	{$c0-$d0},[@t[3]]
+	veor		$c2,$c2,$t2
+	 vst1.8		{$a1-$b1},[r14]!
+	veor		$d2,$d2,$t3
+	 vst1.8		{$c1-$d1},[r14]!
+
+	vadd.i32	$d0#lo,$d0#lo,$t0#lo	@ next counter value
+	vldr		$t0#lo,[sp,#4*(16+0)]	@ one
+
+	ldmia		sp,{@t[0]-@t[3]}	@ load key material
+	add		@x[0],@x[0],@t[0]	@ accumulate key material
+	ldr		@t[0],[r12],#16		@ load input
+	 vst1.8		{$a2-$b2},[r14]!
+	add		@x[1],@x[1],@t[1]
+	ldr		@t[1],[r12,#-12]
+	 vst1.8		{$c2-$d2},[r14]!
+	add		@x[2],@x[2],@t[2]
+	ldr		@t[2],[r12,#-8]
+	add		@x[3],@x[3],@t[3]
+	ldr		@t[3],[r12,#-4]
+# ifdef	__ARMEB__
+	rev		@x[0],@x[0]
+	rev		@x[1],@x[1]
+	rev		@x[2],@x[2]
+	rev		@x[3],@x[3]
+# endif
+	eor		@x[0],@x[0],@t[0]	@ xor with input
+	 add		@t[0],sp,#4*(4)
+	eor		@x[1],@x[1],@t[1]
+	str		@x[0],[r14],#16		@ store output
+	eor		@x[2],@x[2],@t[2]
+	str		@x[1],[r14,#-12]
+	eor		@x[3],@x[3],@t[3]
+	 ldmia		@t[0],{@t[0]-@t[3]}	@ load key material
+	str		@x[2],[r14,#-8]
+	str		@x[3],[r14,#-4]
+
+	add		@x[4],@t[0],@x[4],ror#13 @ accumulate key material
+	ldr		@t[0],[r12],#16		@ load input
+	add		@x[5],@t[1],@x[5],ror#13
+	ldr		@t[1],[r12,#-12]
+	add		@x[6],@t[2],@x[6],ror#13
+	ldr		@t[2],[r12,#-8]
+	add		@x[7],@t[3],@x[7],ror#13
+	ldr		@t[3],[r12,#-4]
+# ifdef	__ARMEB__
+	rev		@x[4],@x[4]
+	rev		@x[5],@x[5]
+	rev		@x[6],@x[6]
+	rev		@x[7],@x[7]
+# endif
+	eor		@x[4],@x[4],@t[0]
+	 add		@t[0],sp,#4*(8)
+	eor		@x[5],@x[5],@t[1]
+	str		@x[4],[r14],#16		@ store output
+	eor		@x[6],@x[6],@t[2]
+	str		@x[5],[r14,#-12]
+	eor		@x[7],@x[7],@t[3]
+	 ldmia		@t[0],{@t[0]-@t[3]}	@ load key material
+	str		@x[6],[r14,#-8]
+	 add		@x[0],sp,#4*(16+8)
+	str		@x[7],[r14,#-4]
+
+	ldmia		@x[0],{@x[0]-@x[7]}	@ load second half
+
+	add		@x[0],@x[0],@t[0]	@ accumulate key material
+	ldr		@t[0],[r12],#16		@ load input
+	add		@x[1],@x[1],@t[1]
+	ldr		@t[1],[r12,#-12]
+# ifdef	__thumb2__
+	it	hi
+# endif
+	 strhi		@t[2],[sp,#4*(16+10)]	@ copy "@x[10]" while at it
+	add		@x[2],@x[2],@t[2]
+	ldr		@t[2],[r12,#-8]
+# ifdef	__thumb2__
+	it	hi
+# endif
+	 strhi		@t[3],[sp,#4*(16+11)]	@ copy "@x[11]" while at it
+	add		@x[3],@x[3],@t[3]
+	ldr		@t[3],[r12,#-4]
+# ifdef	__ARMEB__
+	rev		@x[0],@x[0]
+	rev		@x[1],@x[1]
+	rev		@x[2],@x[2]
+	rev		@x[3],@x[3]
+# endif
+	eor		@x[0],@x[0],@t[0]
+	 add		@t[0],sp,#4*(12)
+	eor		@x[1],@x[1],@t[1]
+	str		@x[0],[r14],#16		@ store output
+	eor		@x[2],@x[2],@t[2]
+	str		@x[1],[r14,#-12]
+	eor		@x[3],@x[3],@t[3]
+	 ldmia		@t[0],{@t[0]-@t[3]}	@ load key material
+	str		@x[2],[r14,#-8]
+	str		@x[3],[r14,#-4]
+
+	add		@x[4],@t[0],@x[4],ror#24 @ accumulate key material
+	 add		@t[0],@t[0],#4		@ next counter value
+	add		@x[5],@t[1],@x[5],ror#24
+	 str		@t[0],[sp,#4*(12)]	@ save next counter value
+	ldr		@t[0],[r12],#16		@ load input
+	add		@x[6],@t[2],@x[6],ror#24
+	 add		@x[4],@x[4],#3		@ counter+3
+	ldr		@t[1],[r12,#-12]
+	add		@x[7],@t[3],@x[7],ror#24
+	ldr		@t[2],[r12,#-8]
+	ldr		@t[3],[r12,#-4]
+# ifdef	__ARMEB__
+	rev		@x[4],@x[4]
+	rev		@x[5],@x[5]
+	rev		@x[6],@x[6]
+	rev		@x[7],@x[7]
+# endif
+	eor		@x[4],@x[4],@t[0]
+# ifdef	__thumb2__
+	it	hi
+# endif
+	 ldrhi		@t[0],[sp,#4*(32+2)]	@ re-load len
+	eor		@x[5],@x[5],@t[1]
+	eor		@x[6],@x[6],@t[2]
+	str		@x[4],[r14],#16		@ store output
+	eor		@x[7],@x[7],@t[3]
+	str		@x[5],[r14,#-12]
+	 sub		@t[3],@t[0],#64*4	@ len-=64*4
+	str		@x[6],[r14,#-8]
+	str		@x[7],[r14,#-4]
+	bhi		.Loop_neon_outer
+
+	b		.Ldone_neon
+
+.align	4
+.Lbreak_neon:
+	@ harmonize NEON and integer-only stack frames: load data
+	@ from NEON frame, but save to integer-only one; distance
+	@ between the two is 4*(32+4+16-32)=4*(20).
+
+	str		@t[3], [sp,#4*(20+32+2)]	@ save len
+	 add		@t[3],sp,#4*(32+4)
+	str		r12,   [sp,#4*(20+32+1)]	@ save inp
+	str		r14,   [sp,#4*(20+32+0)]	@ save out
+
+	ldr		@x[12],[sp,#4*(16+10)]
+	ldr		@x[14],[sp,#4*(16+11)]
+	 vldmia		@t[3],{d8-d15}			@ fulfill ABI requirement
+	str		@x[12],[sp,#4*(20+16+10)]	@ copy "@x[10]"
+	str		@x[14],[sp,#4*(20+16+11)]	@ copy "@x[11]"
+
+	ldr		@t[3], [sp,#4*(15)]
+	 mov		@x[4],@x[4],ror#19		@ twist b[0..3]
+	ldr		@x[12],[sp,#4*(12)]		@ modulo-scheduled load
+	 mov		@x[5],@x[5],ror#19
+	ldr		@t[2], [sp,#4*(13)]
+	 mov		@x[6],@x[6],ror#19
+	ldr		@x[14],[sp,#4*(14)]
+	 mov		@x[7],@x[7],ror#19
+	mov		@t[3],@t[3],ror#8		@ twist d[0..3]
+	mov		@x[12],@x[12],ror#8
+	mov		@t[2],@t[2],ror#8
+	mov		@x[14],@x[14],ror#8
+	str		@t[3], [sp,#4*(20+16+15)]
+	add		@t[3],sp,#4*(20)
+	vst1.32		{$a0-$b0},[@t[3]]!		@ copy key
+	add		sp,sp,#4*(20)			@ switch frame
+	vst1.32		{$c0-$d0},[@t[3]]
+	mov		@t[3],#10
+	b		.Loop				@ go integer-only
+
+.align	4
+.Ltail_neon:
+	cmp		@t[3],#64*3
+	bhs		.L192_or_more_neon
+	cmp		@t[3],#64*2
+	bhs		.L128_or_more_neon
+	cmp		@t[3],#64*1
+	bhs		.L64_or_more_neon
+
+	add		@t[0],sp,#4*(8)
+	vst1.8		{$a0-$b0},[sp]
+	add		@t[2],sp,#4*(0)
+	vst1.8		{$c0-$d0},[@t[0]]
+	b		.Loop_tail_neon
+
+.align	4
+.L64_or_more_neon:
+	vld1.8		{$t0-$t1},[r12]!
+	vld1.8		{$t2-$t3},[r12]!
+	veor		$a0,$a0,$t0
+	veor		$b0,$b0,$t1
+	veor		$c0,$c0,$t2
+	veor		$d0,$d0,$t3
+	vst1.8		{$a0-$b0},[r14]!
+	vst1.8		{$c0-$d0},[r14]!
+
+	beq		.Ldone_neon
+
+	add		@t[0],sp,#4*(8)
+	vst1.8		{$a1-$b1},[sp]
+	add		@t[2],sp,#4*(0)
+	vst1.8		{$c1-$d1},[@t[0]]
+	sub		@t[3],@t[3],#64*1	@ len-=64*1
+	b		.Loop_tail_neon
+
+.align	4
+.L128_or_more_neon:
+	vld1.8		{$t0-$t1},[r12]!
+	vld1.8		{$t2-$t3},[r12]!
+	veor		$a0,$a0,$t0
+	veor		$b0,$b0,$t1
+	vld1.8		{$t0-$t1},[r12]!
+	veor		$c0,$c0,$t2
+	veor		$d0,$d0,$t3
+	vld1.8		{$t2-$t3},[r12]!
+
+	veor		$a1,$a1,$t0
+	veor		$b1,$b1,$t1
+	 vst1.8		{$a0-$b0},[r14]!
+	veor		$c1,$c1,$t2
+	 vst1.8		{$c0-$d0},[r14]!
+	veor		$d1,$d1,$t3
+	vst1.8		{$a1-$b1},[r14]!
+	vst1.8		{$c1-$d1},[r14]!
+
+	beq		.Ldone_neon
+
+	add		@t[0],sp,#4*(8)
+	vst1.8		{$a2-$b2},[sp]
+	add		@t[2],sp,#4*(0)
+	vst1.8		{$c2-$d2},[@t[0]]
+	sub		@t[3],@t[3],#64*2	@ len-=64*2
+	b		.Loop_tail_neon
+
+.align	4
+.L192_or_more_neon:
+	vld1.8		{$t0-$t1},[r12]!
+	vld1.8		{$t2-$t3},[r12]!
+	veor		$a0,$a0,$t0
+	veor		$b0,$b0,$t1
+	vld1.8		{$t0-$t1},[r12]!
+	veor		$c0,$c0,$t2
+	veor		$d0,$d0,$t3
+	vld1.8		{$t2-$t3},[r12]!
+
+	veor		$a1,$a1,$t0
+	veor		$b1,$b1,$t1
+	vld1.8		{$t0-$t1},[r12]!
+	veor		$c1,$c1,$t2
+	 vst1.8		{$a0-$b0},[r14]!
+	veor		$d1,$d1,$t3
+	vld1.8		{$t2-$t3},[r12]!
+
+	veor		$a2,$a2,$t0
+	 vst1.8		{$c0-$d0},[r14]!
+	veor		$b2,$b2,$t1
+	 vst1.8		{$a1-$b1},[r14]!
+	veor		$c2,$c2,$t2
+	 vst1.8		{$c1-$d1},[r14]!
+	veor		$d2,$d2,$t3
+	vst1.8		{$a2-$b2},[r14]!
+	vst1.8		{$c2-$d2},[r14]!
+
+	beq		.Ldone_neon
+
+	ldmia		sp,{@t[0]-@t[3]}	@ load key material
+	add		@x[0],@x[0],@t[0]	@ accumulate key material
+	 add		@t[0],sp,#4*(4)
+	add		@x[1],@x[1],@t[1]
+	add		@x[2],@x[2],@t[2]
+	add		@x[3],@x[3],@t[3]
+	 ldmia		@t[0],{@t[0]-@t[3]}	@ load key material
+
+	add		@x[4],@t[0],@x[4],ror#13 @ accumulate key material
+	 add		@t[0],sp,#4*(8)
+	add		@x[5],@t[1],@x[5],ror#13
+	add		@x[6],@t[2],@x[6],ror#13
+	add		@x[7],@t[3],@x[7],ror#13
+	 ldmia		@t[0],{@t[0]-@t[3]}	@ load key material
+# ifdef	__ARMEB__
+	rev		@x[0],@x[0]
+	rev		@x[1],@x[1]
+	rev		@x[2],@x[2]
+	rev		@x[3],@x[3]
+	rev		@x[4],@x[4]
+	rev		@x[5],@x[5]
+	rev		@x[6],@x[6]
+	rev		@x[7],@x[7]
+# endif
+	stmia		sp,{@x[0]-@x[7]}
+	 add		@x[0],sp,#4*(16+8)
+
+	ldmia		@x[0],{@x[0]-@x[7]}	@ load second half
+
+	add		@x[0],@x[0],@t[0]	@ accumulate key material
+	 add		@t[0],sp,#4*(12)
+	add		@x[1],@x[1],@t[1]
+	add		@x[2],@x[2],@t[2]
+	add		@x[3],@x[3],@t[3]
+	 ldmia		@t[0],{@t[0]-@t[3]}	@ load key material
+
+	add		@x[4],@t[0],@x[4],ror#24 @ accumulate key material
+	 add		@t[0],sp,#4*(8)
+	add		@x[5],@t[1],@x[5],ror#24
+	 add		@x[4],@x[4],#3		@ counter+3
+	add		@x[6],@t[2],@x[6],ror#24
+	add		@x[7],@t[3],@x[7],ror#24
+	 ldr		@t[3],[sp,#4*(32+2)]	@ re-load len
+# ifdef	__ARMEB__
+	rev		@x[0],@x[0]
+	rev		@x[1],@x[1]
+	rev		@x[2],@x[2]
+	rev		@x[3],@x[3]
+	rev		@x[4],@x[4]
+	rev		@x[5],@x[5]
+	rev		@x[6],@x[6]
+	rev		@x[7],@x[7]
+# endif
+	stmia		@t[0],{@x[0]-@x[7]}
+	 add		@t[2],sp,#4*(0)
+	 sub		@t[3],@t[3],#64*3	@ len-=64*3
+
+.Loop_tail_neon:
+	ldrb		@t[0],[@t[2]],#1	@ read buffer on stack
+	ldrb		@t[1],[r12],#1		@ read input
+	subs		@t[3],@t[3],#1
+	eor		@t[0],@t[0],@t[1]
+	strb		@t[0],[r14],#1		@ store output
+	bne		.Loop_tail_neon
+
+.Ldone_neon:
+	add		sp,sp,#4*(32+4)
+	vldmia		sp,{d8-d15}
+	add		sp,sp,#4*(16+3)
+	ldmia		sp!,{r4-r11,pc}
+.size	ChaCha20_neon,.-ChaCha20_neon
+# ifndef __KERNEL__
+.comm	OPENSSL_armcap_P,4,4
+# endif
+#endif
+___
+}}}
+
+open SELF,$0;
+while(<SELF>) {
+	next if (/^#!/);
+	last if (!s/^#/@/ and !/^$/);
+	print;
+}
+close SELF;
+
+foreach (split("\n",$code)) {
+	s/\`([^\`]*)\`/eval $1/geo;
+
+	s/\bq([0-9]+)#(lo|hi)/sprintf "d%d",2*$1+($2 eq "hi")/geo;
+
+	print $_,"\n";
+}
+close STDOUT;

@@ -0,0 +1,1163 @@
+#!/usr/bin/env perl
+# SPDX-License-Identifier: GPL-2.0 OR BSD-3-Clause
+#
+# This code is taken from the OpenSSL project but the author, Andy Polyakov,
+# has relicensed it under the licenses specified in the SPDX header above.
+# The original headers, including the original license headers, are
+# included below for completeness.
+#
+# ====================================================================
+# Written by Andy Polyakov <appro@openssl.org> for the OpenSSL
+# project. The module is, however, dual licensed under OpenSSL and
+# CRYPTOGAMS licenses depending on where you obtain it. For further
+# details see http://www.openssl.org/~appro/cryptogams/.
+# ====================================================================
+#
+# June 2015
+#
+# ChaCha20 for ARMv8.
+#
+# Performance in cycles per byte out of large buffer.
+#
+#			IALU/gcc-4.9    3xNEON+1xIALU	6xNEON+2xIALU(*)
+#
+# Apple A7		5.50/+49%       3.33            1.70
+# Cortex-A53		8.40/+80%       4.72		4.72(**)
+# Cortex-A57		8.06/+43%       4.90            4.43(***)
+# Denver		4.50/+82%       2.63		2.67(**)
+# X-Gene		9.50/+46%       8.82		8.89(**)
+# Mongoose		8.00/+44%	3.64		3.25(***)
+# Kryo			8.17/+50%	4.83		4.65(***)
+#
+# (*)	since no non-Apple processor exhibits significantly better
+#	performance, the code path is #ifdef __APPLE__-ed;
+# (**)	it's expected that doubling interleave factor doesn't help
+#	all processors, only those with higher NEON latency and
+#	higher instruction issue rate;
+# (***)	expected improvement was actually higher;
+
+$flavour=shift;
+if ($flavour=~/\w[\w\-]*\.\w+$/) { $output=$flavour; undef $flavour; }
+else { while (($output=shift) && ($output!~/\w[\w\-]*\.\w+$/)) {} }
+
+if ($flavour && $flavour ne "void") {
+    $0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
+    ( $xlate="${dir}arm-xlate.pl" and -f $xlate ) or
+    ( $xlate="${dir}../../perlasm/arm-xlate.pl" and -f $xlate) or
+    die "can't locate arm-xlate.pl";
+
+    open STDOUT,"| \"$^X\" $xlate $flavour $output";
+} else {
+    open STDOUT,">$output";
+}
+
+sub AUTOLOAD()		# thunk [simplified] x86-style perlasm
+{ my $opcode = $AUTOLOAD; $opcode =~ s/.*:://; $opcode =~ s/_/\./;
+  my $arg = pop;
+    $arg = "#$arg" if ($arg*1 eq $arg);
+    $code .= "\t$opcode\t".join(',',@_,$arg)."\n";
+}
+
+my ($out,$inp,$len,$key,$ctr) = map("x$_",(0..4));
+
+my @x=map("x$_",(5..17,19..21));
+my @d=map("x$_",(22..28,30));
+
+sub ROUND {
+my ($a0,$b0,$c0,$d0)=@_;
+my ($a1,$b1,$c1,$d1)=map(($_&~3)+(($_+1)&3),($a0,$b0,$c0,$d0));
+my ($a2,$b2,$c2,$d2)=map(($_&~3)+(($_+1)&3),($a1,$b1,$c1,$d1));
+my ($a3,$b3,$c3,$d3)=map(($_&~3)+(($_+1)&3),($a2,$b2,$c2,$d2));
+
+    (
+	"&add_32	(@x[$a0],@x[$a0],@x[$b0])",
+	 "&add_32	(@x[$a1],@x[$a1],@x[$b1])",
+	  "&add_32	(@x[$a2],@x[$a2],@x[$b2])",
+	   "&add_32	(@x[$a3],@x[$a3],@x[$b3])",
+	"&eor_32	(@x[$d0],@x[$d0],@x[$a0])",
+	 "&eor_32	(@x[$d1],@x[$d1],@x[$a1])",
+	  "&eor_32	(@x[$d2],@x[$d2],@x[$a2])",
+	   "&eor_32	(@x[$d3],@x[$d3],@x[$a3])",
+	"&ror_32	(@x[$d0],@x[$d0],16)",
+	 "&ror_32	(@x[$d1],@x[$d1],16)",
+	  "&ror_32	(@x[$d2],@x[$d2],16)",
+	   "&ror_32	(@x[$d3],@x[$d3],16)",
+
+	"&add_32	(@x[$c0],@x[$c0],@x[$d0])",
+	 "&add_32	(@x[$c1],@x[$c1],@x[$d1])",
+	  "&add_32	(@x[$c2],@x[$c2],@x[$d2])",
+	   "&add_32	(@x[$c3],@x[$c3],@x[$d3])",
+	"&eor_32	(@x[$b0],@x[$b0],@x[$c0])",
+	 "&eor_32	(@x[$b1],@x[$b1],@x[$c1])",
+	  "&eor_32	(@x[$b2],@x[$b2],@x[$c2])",
+	   "&eor_32	(@x[$b3],@x[$b3],@x[$c3])",
+	"&ror_32	(@x[$b0],@x[$b0],20)",
+	 "&ror_32	(@x[$b1],@x[$b1],20)",
+	  "&ror_32	(@x[$b2],@x[$b2],20)",
+	   "&ror_32	(@x[$b3],@x[$b3],20)",
+
+	"&add_32	(@x[$a0],@x[$a0],@x[$b0])",
+	 "&add_32	(@x[$a1],@x[$a1],@x[$b1])",
+	  "&add_32	(@x[$a2],@x[$a2],@x[$b2])",
+	   "&add_32	(@x[$a3],@x[$a3],@x[$b3])",
+	"&eor_32	(@x[$d0],@x[$d0],@x[$a0])",
+	 "&eor_32	(@x[$d1],@x[$d1],@x[$a1])",
+	  "&eor_32	(@x[$d2],@x[$d2],@x[$a2])",
+	   "&eor_32	(@x[$d3],@x[$d3],@x[$a3])",
+	"&ror_32	(@x[$d0],@x[$d0],24)",
+	 "&ror_32	(@x[$d1],@x[$d1],24)",
+	  "&ror_32	(@x[$d2],@x[$d2],24)",
+	   "&ror_32	(@x[$d3],@x[$d3],24)",
+
+	"&add_32	(@x[$c0],@x[$c0],@x[$d0])",
+	 "&add_32	(@x[$c1],@x[$c1],@x[$d1])",
+	  "&add_32	(@x[$c2],@x[$c2],@x[$d2])",
+	   "&add_32	(@x[$c3],@x[$c3],@x[$d3])",
+	"&eor_32	(@x[$b0],@x[$b0],@x[$c0])",
+	 "&eor_32	(@x[$b1],@x[$b1],@x[$c1])",
+	  "&eor_32	(@x[$b2],@x[$b2],@x[$c2])",
+	   "&eor_32	(@x[$b3],@x[$b3],@x[$c3])",
+	"&ror_32	(@x[$b0],@x[$b0],25)",
+	 "&ror_32	(@x[$b1],@x[$b1],25)",
+	  "&ror_32	(@x[$b2],@x[$b2],25)",
+	   "&ror_32	(@x[$b3],@x[$b3],25)"
+    );
+}
+
+$code.=<<___;
+#ifndef	__KERNEL__
+# include "arm_arch.h"
+.extern	OPENSSL_armcap_P
+#else
+# define ChaCha20_ctr32 chacha20_arm
+# define ChaCha20_neon  chacha20_neon
+#endif
+
+.text
+
+.align	5
+.Lsigma:
+.quad	0x3320646e61707865,0x6b20657479622d32		// endian-neutral
+.Lone:
+.long	1,0,0,0
+#ifndef	__KERNEL__
+.LOPENSSL_armcap_P:
+# ifdef	__ILP32__
+.long	OPENSSL_armcap_P-.
+# else
+.quad	OPENSSL_armcap_P-.
+# endif
+#endif
+
+.globl	ChaCha20_ctr32
+.type	ChaCha20_ctr32,%function
+.align	5
+ChaCha20_ctr32:
+	cbz	$len,.Labort
+#ifndef	__KERNEL__
+	adr	@x[0],.LOPENSSL_armcap_P
+	cmp	$len,#192
+	b.lo	.Lshort
+# ifdef	__ILP32__
+	ldrsw	@x[1],[@x[0]]
+# else
+	ldr	@x[1],[@x[0]]
+# endif
+	ldr	w17,[@x[1],@x[0]]
+	tst	w17,#ARMV7_NEON
+	b.ne	ChaCha20_neon
+
+.Lshort:
+#endif
+	stp	x29,x30,[sp,#-96]!
+	add	x29,sp,#0
+
+	adr	@x[0],.Lsigma
+	stp	x19,x20,[sp,#16]
+	stp	x21,x22,[sp,#32]
+	stp	x23,x24,[sp,#48]
+	stp	x25,x26,[sp,#64]
+	stp	x27,x28,[sp,#80]
+	sub	sp,sp,#64
+
+	ldp	@d[0],@d[1],[@x[0]]		// load sigma
+	ldp	@d[2],@d[3],[$key]		// load key
+	ldp	@d[4],@d[5],[$key,#16]
+	ldp	@d[6],@d[7],[$ctr]		// load counter
+#ifdef	__AARCH64EB__
+	ror	@d[2],@d[2],#32
+	ror	@d[3],@d[3],#32
+	ror	@d[4],@d[4],#32
+	ror	@d[5],@d[5],#32
+	ror	@d[6],@d[6],#32
+	ror	@d[7],@d[7],#32
+#endif
+
+.Loop_outer:
+	mov.32	@x[0],@d[0]			// unpack key block
+	lsr	@x[1],@d[0],#32
+	mov.32	@x[2],@d[1]
+	lsr	@x[3],@d[1],#32
+	mov.32	@x[4],@d[2]
+	lsr	@x[5],@d[2],#32
+	mov.32	@x[6],@d[3]
+	lsr	@x[7],@d[3],#32
+	mov.32	@x[8],@d[4]
+	lsr	@x[9],@d[4],#32
+	mov.32	@x[10],@d[5]
+	lsr	@x[11],@d[5],#32
+	mov.32	@x[12],@d[6]
+	lsr	@x[13],@d[6],#32
+	mov.32	@x[14],@d[7]
+	lsr	@x[15],@d[7],#32
+
+	mov	$ctr,#10
+	subs	$len,$len,#64
+.Loop:
+	sub	$ctr,$ctr,#1
+___
+	foreach (&ROUND(0, 4, 8,12)) { eval; }
+	foreach (&ROUND(0, 5,10,15)) { eval; }
+$code.=<<___;
+	cbnz	$ctr,.Loop
+
+	add.32	@x[0],@x[0],@d[0]		// accumulate key block
+	add	@x[1],@x[1],@d[0],lsr#32
+	add.32	@x[2],@x[2],@d[1]
+	add	@x[3],@x[3],@d[1],lsr#32
+	add.32	@x[4],@x[4],@d[2]
+	add	@x[5],@x[5],@d[2],lsr#32
+	add.32	@x[6],@x[6],@d[3]
+	add	@x[7],@x[7],@d[3],lsr#32
+	add.32	@x[8],@x[8],@d[4]
+	add	@x[9],@x[9],@d[4],lsr#32
+	add.32	@x[10],@x[10],@d[5]
+	add	@x[11],@x[11],@d[5],lsr#32
+	add.32	@x[12],@x[12],@d[6]
+	add	@x[13],@x[13],@d[6],lsr#32
+	add.32	@x[14],@x[14],@d[7]
+	add	@x[15],@x[15],@d[7],lsr#32
+
+	b.lo	.Ltail
+
+	add	@x[0],@x[0],@x[1],lsl#32	// pack
+	add	@x[2],@x[2],@x[3],lsl#32
+	ldp	@x[1],@x[3],[$inp,#0]		// load input
+	add	@x[4],@x[4],@x[5],lsl#32
+	add	@x[6],@x[6],@x[7],lsl#32
+	ldp	@x[5],@x[7],[$inp,#16]
+	add	@x[8],@x[8],@x[9],lsl#32
+	add	@x[10],@x[10],@x[11],lsl#32
+	ldp	@x[9],@x[11],[$inp,#32]
+	add	@x[12],@x[12],@x[13],lsl#32
+	add	@x[14],@x[14],@x[15],lsl#32
+	ldp	@x[13],@x[15],[$inp,#48]
+	add	$inp,$inp,#64
+#ifdef	__AARCH64EB__
+	rev	@x[0],@x[0]
+	rev	@x[2],@x[2]
+	rev	@x[4],@x[4]
+	rev	@x[6],@x[6]
+	rev	@x[8],@x[8]
+	rev	@x[10],@x[10]
+	rev	@x[12],@x[12]
+	rev	@x[14],@x[14]
+#endif
+	eor	@x[0],@x[0],@x[1]
+	eor	@x[2],@x[2],@x[3]
+	eor	@x[4],@x[4],@x[5]
+	eor	@x[6],@x[6],@x[7]
+	eor	@x[8],@x[8],@x[9]
+	eor	@x[10],@x[10],@x[11]
+	eor	@x[12],@x[12],@x[13]
+	eor	@x[14],@x[14],@x[15]
+
+	stp	@x[0],@x[2],[$out,#0]		// store output
+	 add	@d[6],@d[6],#1			// increment counter
+	stp	@x[4],@x[6],[$out,#16]
+	stp	@x[8],@x[10],[$out,#32]
+	stp	@x[12],@x[14],[$out,#48]
+	add	$out,$out,#64
+
+	b.hi	.Loop_outer
+
+	ldp	x19,x20,[x29,#16]
+	add	sp,sp,#64
+	ldp	x21,x22,[x29,#32]
+	ldp	x23,x24,[x29,#48]
+	ldp	x25,x26,[x29,#64]
+	ldp	x27,x28,[x29,#80]
+	ldp	x29,x30,[sp],#96
+.Labort:
+	ret
+
+.align	4
+.Ltail:
+	add	$len,$len,#64
+.Less_than_64:
+	sub	$out,$out,#1
+	add	$inp,$inp,$len
+	add	$out,$out,$len
+	add	$ctr,sp,$len
+	neg	$len,$len
+
+	add	@x[0],@x[0],@x[1],lsl#32	// pack
+	add	@x[2],@x[2],@x[3],lsl#32
+	add	@x[4],@x[4],@x[5],lsl#32
+	add	@x[6],@x[6],@x[7],lsl#32
+	add	@x[8],@x[8],@x[9],lsl#32
+	add	@x[10],@x[10],@x[11],lsl#32
+	add	@x[12],@x[12],@x[13],lsl#32
+	add	@x[14],@x[14],@x[15],lsl#32
+#ifdef	__AARCH64EB__
+	rev	@x[0],@x[0]
+	rev	@x[2],@x[2]
+	rev	@x[4],@x[4]
+	rev	@x[6],@x[6]
+	rev	@x[8],@x[8]
+	rev	@x[10],@x[10]
+	rev	@x[12],@x[12]
+	rev	@x[14],@x[14]
+#endif
+	stp	@x[0],@x[2],[sp,#0]
+	stp	@x[4],@x[6],[sp,#16]
+	stp	@x[8],@x[10],[sp,#32]
+	stp	@x[12],@x[14],[sp,#48]
+
+.Loop_tail:
+	ldrb	w10,[$inp,$len]
+	ldrb	w11,[$ctr,$len]
+	add	$len,$len,#1
+	eor	w10,w10,w11
+	strb	w10,[$out,$len]
+	cbnz	$len,.Loop_tail
+
+	stp	xzr,xzr,[sp,#0]
+	stp	xzr,xzr,[sp,#16]
+	stp	xzr,xzr,[sp,#32]
+	stp	xzr,xzr,[sp,#48]
+
+	ldp	x19,x20,[x29,#16]
+	add	sp,sp,#64
+	ldp	x21,x22,[x29,#32]
+	ldp	x23,x24,[x29,#48]
+	ldp	x25,x26,[x29,#64]
+	ldp	x27,x28,[x29,#80]
+	ldp	x29,x30,[sp],#96
+	ret
+.size	ChaCha20_ctr32,.-ChaCha20_ctr32
+___
+
+{{{
+my ($A0,$B0,$C0,$D0,$A1,$B1,$C1,$D1,$A2,$B2,$C2,$D2,$T0,$T1,$T2,$T3) =
+    map("v$_.4s",(0..7,16..23));
+my (@K)=map("v$_.4s",(24..30));
+my $ONE="v31.4s";
+
+sub NEONROUND {
+my $odd = pop;
+my ($a,$b,$c,$d,$t)=@_;
+
+	(
+	"&add		('$a','$a','$b')",
+	"&eor		('$d','$d','$a')",
+	"&rev32_16	('$d','$d')",		# vrot ($d,16)
+
+	"&add		('$c','$c','$d')",
+	"&eor		('$t','$b','$c')",
+	"&ushr		('$b','$t',20)",
+	"&sli		('$b','$t',12)",
+
+	"&add		('$a','$a','$b')",
+	"&eor		('$t','$d','$a')",
+	"&ushr		('$d','$t',24)",
+	"&sli		('$d','$t',8)",
+
+	"&add		('$c','$c','$d')",
+	"&eor		('$t','$b','$c')",
+	"&ushr		('$b','$t',25)",
+	"&sli		('$b','$t',7)",
+
+	"&ext		('$c','$c','$c',8)",
+	"&ext		('$d','$d','$d',$odd?4:12)",
+	"&ext		('$b','$b','$b',$odd?12:4)"
+	);
+}
+
+$code.=<<___;
+#if !defined(__KERNEL__) || defined(CONFIG_KERNEL_MODE_NEON)
+#ifdef __KERNEL__
+.globl  ChaCha20_neon
+.type   ChaCha20_neon,%function
+#endif
+.type	ChaCha20_neon,%function
+.align	5
+ChaCha20_neon:
+	stp	x29,x30,[sp,#-96]!
+	add	x29,sp,#0
+
+	adr	@x[0],.Lsigma
+	stp	x19,x20,[sp,#16]
+	stp	x21,x22,[sp,#32]
+	stp	x23,x24,[sp,#48]
+	stp	x25,x26,[sp,#64]
+	stp	x27,x28,[sp,#80]
+#ifdef	__APPLE__
+	cmp	$len,#512
+	b.hs	.L512_or_more_neon
+#endif
+
+	sub	sp,sp,#64
+
+	ldp	@d[0],@d[1],[@x[0]]		// load sigma
+	ld1	{@K[0]},[@x[0]],#16
+	ldp	@d[2],@d[3],[$key]		// load key
+	ldp	@d[4],@d[5],[$key,#16]
+	ld1	{@K[1],@K[2]},[$key]
+	ldp	@d[6],@d[7],[$ctr]		// load counter
+	ld1	{@K[3]},[$ctr]
+	ld1	{$ONE},[@x[0]]
+#ifdef	__AARCH64EB__
+	rev64	@K[0],@K[0]
+	ror	@d[2],@d[2],#32
+	ror	@d[3],@d[3],#32
+	ror	@d[4],@d[4],#32
+	ror	@d[5],@d[5],#32
+	ror	@d[6],@d[6],#32
+	ror	@d[7],@d[7],#32
+#endif
+	add	@K[3],@K[3],$ONE		// += 1
+	add	@K[4],@K[3],$ONE
+	add	@K[5],@K[4],$ONE
+	shl	$ONE,$ONE,#2			// 1 -> 4
+
+.Loop_outer_neon:
+	mov.32	@x[0],@d[0]			// unpack key block
+	lsr	@x[1],@d[0],#32
+	 mov	$A0,@K[0]
+	mov.32	@x[2],@d[1]
+	lsr	@x[3],@d[1],#32
+	 mov	$A1,@K[0]
+	mov.32	@x[4],@d[2]
+	lsr	@x[5],@d[2],#32
+	 mov	$A2,@K[0]
+	mov.32	@x[6],@d[3]
+	 mov	$B0,@K[1]
+	lsr	@x[7],@d[3],#32
+	 mov	$B1,@K[1]
+	mov.32	@x[8],@d[4]
+	 mov	$B2,@K[1]
+	lsr	@x[9],@d[4],#32
+	 mov	$D0,@K[3]
+	mov.32	@x[10],@d[5]
+	 mov	$D1,@K[4]
+	lsr	@x[11],@d[5],#32
+	 mov	$D2,@K[5]
+	mov.32	@x[12],@d[6]
+	 mov	$C0,@K[2]
+	lsr	@x[13],@d[6],#32
+	 mov	$C1,@K[2]
+	mov.32	@x[14],@d[7]
+	 mov	$C2,@K[2]
+	lsr	@x[15],@d[7],#32
+
+	mov	$ctr,#10
+	subs	$len,$len,#256
+.Loop_neon:
+	sub	$ctr,$ctr,#1
+___
+	my @thread0=&NEONROUND($A0,$B0,$C0,$D0,$T0,0);
+	my @thread1=&NEONROUND($A1,$B1,$C1,$D1,$T1,0);
+	my @thread2=&NEONROUND($A2,$B2,$C2,$D2,$T2,0);
+	my @thread3=&ROUND(0,4,8,12);
+
+	foreach (@thread0) {
+		eval;			eval(shift(@thread3));
+		eval(shift(@thread1));	eval(shift(@thread3));
+		eval(shift(@thread2));	eval(shift(@thread3));
+	}
+
+	@thread0=&NEONROUND($A0,$B0,$C0,$D0,$T0,1);
+	@thread1=&NEONROUND($A1,$B1,$C1,$D1,$T1,1);
+	@thread2=&NEONROUND($A2,$B2,$C2,$D2,$T2,1);
+	@thread3=&ROUND(0,5,10,15);
+
+	foreach (@thread0) {
+		eval;			eval(shift(@thread3));
+		eval(shift(@thread1));	eval(shift(@thread3));
+		eval(shift(@thread2));	eval(shift(@thread3));
+	}
+$code.=<<___;
+	cbnz	$ctr,.Loop_neon
+
+	add.32	@x[0],@x[0],@d[0]		// accumulate key block
+	 add	$A0,$A0,@K[0]
+	add	@x[1],@x[1],@d[0],lsr#32
+	 add	$A1,$A1,@K[0]
+	add.32	@x[2],@x[2],@d[1]
+	 add	$A2,$A2,@K[0]
+	add	@x[3],@x[3],@d[1],lsr#32
+	 add	$C0,$C0,@K[2]
+	add.32	@x[4],@x[4],@d[2]
+	 add	$C1,$C1,@K[2]
+	add	@x[5],@x[5],@d[2],lsr#32
+	 add	$C2,$C2,@K[2]
+	add.32	@x[6],@x[6],@d[3]
+	 add	$D0,$D0,@K[3]
+	add	@x[7],@x[7],@d[3],lsr#32
+	add.32	@x[8],@x[8],@d[4]
+	 add	$D1,$D1,@K[4]
+	add	@x[9],@x[9],@d[4],lsr#32
+	add.32	@x[10],@x[10],@d[5]
+	 add	$D2,$D2,@K[5]
+	add	@x[11],@x[11],@d[5],lsr#32
+	add.32	@x[12],@x[12],@d[6]
+	 add	$B0,$B0,@K[1]
+	add	@x[13],@x[13],@d[6],lsr#32
+	add.32	@x[14],@x[14],@d[7]
+	 add	$B1,$B1,@K[1]
+	add	@x[15],@x[15],@d[7],lsr#32
+	 add	$B2,$B2,@K[1]
+
+	b.lo	.Ltail_neon
+
+	add	@x[0],@x[0],@x[1],lsl#32	// pack
+	add	@x[2],@x[2],@x[3],lsl#32
+	ldp	@x[1],@x[3],[$inp,#0]		// load input
+	add	@x[4],@x[4],@x[5],lsl#32
+	add	@x[6],@x[6],@x[7],lsl#32
+	ldp	@x[5],@x[7],[$inp,#16]
+	add	@x[8],@x[8],@x[9],lsl#32
+	add	@x[10],@x[10],@x[11],lsl#32
+	ldp	@x[9],@x[11],[$inp,#32]
+	add	@x[12],@x[12],@x[13],lsl#32
+	add	@x[14],@x[14],@x[15],lsl#32
+	ldp	@x[13],@x[15],[$inp,#48]
+	add	$inp,$inp,#64
+#ifdef	__AARCH64EB__
+	rev	@x[0],@x[0]
+	rev	@x[2],@x[2]
+	rev	@x[4],@x[4]
+	rev	@x[6],@x[6]
+	rev	@x[8],@x[8]
+	rev	@x[10],@x[10]
+	rev	@x[12],@x[12]
+	rev	@x[14],@x[14]
+#endif
+	ld1.8	{$T0-$T3},[$inp],#64
+	eor	@x[0],@x[0],@x[1]
+	eor	@x[2],@x[2],@x[3]
+	eor	@x[4],@x[4],@x[5]
+	eor	@x[6],@x[6],@x[7]
+	eor	@x[8],@x[8],@x[9]
+	 eor	$A0,$A0,$T0
+	eor	@x[10],@x[10],@x[11]
+	 eor	$B0,$B0,$T1
+	eor	@x[12],@x[12],@x[13]
+	 eor	$C0,$C0,$T2
+	eor	@x[14],@x[14],@x[15]
+	 eor	$D0,$D0,$T3
+	 ld1.8	{$T0-$T3},[$inp],#64
+
+	stp	@x[0],@x[2],[$out,#0]		// store output
+	 add	@d[6],@d[6],#4			// increment counter
+	stp	@x[4],@x[6],[$out,#16]
+	 add	@K[3],@K[3],$ONE		// += 4
+	stp	@x[8],@x[10],[$out,#32]
+	 add	@K[4],@K[4],$ONE
+	stp	@x[12],@x[14],[$out,#48]
+	 add	@K[5],@K[5],$ONE
+	add	$out,$out,#64
+
+	st1.8	{$A0-$D0},[$out],#64
+	ld1.8	{$A0-$D0},[$inp],#64
+
+	eor	$A1,$A1,$T0
+	eor	$B1,$B1,$T1
+	eor	$C1,$C1,$T2
+	eor	$D1,$D1,$T3
+	st1.8	{$A1-$D1},[$out],#64
+
+	eor	$A2,$A2,$A0
+	eor	$B2,$B2,$B0
+	eor	$C2,$C2,$C0
+	eor	$D2,$D2,$D0
+	st1.8	{$A2-$D2},[$out],#64
+
+	b.hi	.Loop_outer_neon
+
+	ldp	x19,x20,[x29,#16]
+	add	sp,sp,#64
+	ldp	x21,x22,[x29,#32]
+	ldp	x23,x24,[x29,#48]
+	ldp	x25,x26,[x29,#64]
+	ldp	x27,x28,[x29,#80]
+	ldp	x29,x30,[sp],#96
+	ret
+
+.Ltail_neon:
+	add	$len,$len,#256
+	cmp	$len,#64
+	b.lo	.Less_than_64
+
+	add	@x[0],@x[0],@x[1],lsl#32	// pack
+	add	@x[2],@x[2],@x[3],lsl#32
+	ldp	@x[1],@x[3],[$inp,#0]		// load input
+	add	@x[4],@x[4],@x[5],lsl#32
+	add	@x[6],@x[6],@x[7],lsl#32
+	ldp	@x[5],@x[7],[$inp,#16]
+	add	@x[8],@x[8],@x[9],lsl#32
+	add	@x[10],@x[10],@x[11],lsl#32
+	ldp	@x[9],@x[11],[$inp,#32]
+	add	@x[12],@x[12],@x[13],lsl#32
+	add	@x[14],@x[14],@x[15],lsl#32
+	ldp	@x[13],@x[15],[$inp,#48]
+	add	$inp,$inp,#64
+#ifdef	__AARCH64EB__
+	rev	@x[0],@x[0]
+	rev	@x[2],@x[2]
+	rev	@x[4],@x[4]
+	rev	@x[6],@x[6]
+	rev	@x[8],@x[8]
+	rev	@x[10],@x[10]
+	rev	@x[12],@x[12]
+	rev	@x[14],@x[14]
+#endif
+	eor	@x[0],@x[0],@x[1]
+	eor	@x[2],@x[2],@x[3]
+	eor	@x[4],@x[4],@x[5]
+	eor	@x[6],@x[6],@x[7]
+	eor	@x[8],@x[8],@x[9]
+	eor	@x[10],@x[10],@x[11]
+	eor	@x[12],@x[12],@x[13]
+	eor	@x[14],@x[14],@x[15]
+
+	stp	@x[0],@x[2],[$out,#0]		// store output
+	 add	@d[6],@d[6],#4			// increment counter
+	stp	@x[4],@x[6],[$out,#16]
+	stp	@x[8],@x[10],[$out,#32]
+	stp	@x[12],@x[14],[$out,#48]
+	add	$out,$out,#64
+	b.eq	.Ldone_neon
+	sub	$len,$len,#64
+	cmp	$len,#64
+	b.lo	.Less_than_128
+
+	ld1.8	{$T0-$T3},[$inp],#64
+	eor	$A0,$A0,$T0
+	eor	$B0,$B0,$T1
+	eor	$C0,$C0,$T2
+	eor	$D0,$D0,$T3
+	st1.8	{$A0-$D0},[$out],#64
+	b.eq	.Ldone_neon
+	sub	$len,$len,#64
+	cmp	$len,#64
+	b.lo	.Less_than_192
+
+	ld1.8	{$T0-$T3},[$inp],#64
+	eor	$A1,$A1,$T0
+	eor	$B1,$B1,$T1
+	eor	$C1,$C1,$T2
+	eor	$D1,$D1,$T3
+	st1.8	{$A1-$D1},[$out],#64
+	b.eq	.Ldone_neon
+	sub	$len,$len,#64
+
+	st1.8	{$A2-$D2},[sp]
+	b	.Last_neon
+
+.Less_than_128:
+	st1.8	{$A0-$D0},[sp]
+	b	.Last_neon
+.Less_than_192:
+	st1.8	{$A1-$D1},[sp]
+	b	.Last_neon
+
+.align	4
+.Last_neon:
+	sub	$out,$out,#1
+	add	$inp,$inp,$len
+	add	$out,$out,$len
+	add	$ctr,sp,$len
+	neg	$len,$len
+
+.Loop_tail_neon:
+	ldrb	w10,[$inp,$len]
+	ldrb	w11,[$ctr,$len]
+	add	$len,$len,#1
+	eor	w10,w10,w11
+	strb	w10,[$out,$len]
+	cbnz	$len,.Loop_tail_neon
+
+	stp	xzr,xzr,[sp,#0]
+	stp	xzr,xzr,[sp,#16]
+	stp	xzr,xzr,[sp,#32]
+	stp	xzr,xzr,[sp,#48]
+
+.Ldone_neon:
+	ldp	x19,x20,[x29,#16]
+	add	sp,sp,#64
+	ldp	x21,x22,[x29,#32]
+	ldp	x23,x24,[x29,#48]
+	ldp	x25,x26,[x29,#64]
+	ldp	x27,x28,[x29,#80]
+	ldp	x29,x30,[sp],#96
+	ret
+.size	ChaCha20_neon,.-ChaCha20_neon
+___
+{
+my ($T0,$T1,$T2,$T3,$T4,$T5)=@K;
+my ($A0,$B0,$C0,$D0,$A1,$B1,$C1,$D1,$A2,$B2,$C2,$D2,
+    $A3,$B3,$C3,$D3,$A4,$B4,$C4,$D4,$A5,$B5,$C5,$D5) = map("v$_.4s",(0..23));
+
+$code.=<<___;
+#ifdef	__APPLE__
+.type	ChaCha20_512_neon,%function
+.align	5
+ChaCha20_512_neon:
+	stp	x29,x30,[sp,#-96]!
+	add	x29,sp,#0
+
+	adr	@x[0],.Lsigma
+	stp	x19,x20,[sp,#16]
+	stp	x21,x22,[sp,#32]
+	stp	x23,x24,[sp,#48]
+	stp	x25,x26,[sp,#64]
+	stp	x27,x28,[sp,#80]
+
+.L512_or_more_neon:
+	sub	sp,sp,#128+64
+
+	ldp	@d[0],@d[1],[@x[0]]		// load sigma
+	ld1	{@K[0]},[@x[0]],#16
+	ldp	@d[2],@d[3],[$key]		// load key
+	ldp	@d[4],@d[5],[$key,#16]
+	ld1	{@K[1],@K[2]},[$key]
+	ldp	@d[6],@d[7],[$ctr]		// load counter
+	ld1	{@K[3]},[$ctr]
+	ld1	{$ONE},[@x[0]]
+# ifdef	__AARCH64EB__
+	rev64	@K[0],@K[0]
+	ror	@d[2],@d[2],#32
+	ror	@d[3],@d[3],#32
+	ror	@d[4],@d[4],#32
+	ror	@d[5],@d[5],#32
+	ror	@d[6],@d[6],#32
+	ror	@d[7],@d[7],#32
+# endif
+	add	@K[3],@K[3],$ONE		// += 1
+	stp	@K[0],@K[1],[sp,#0]		// off-load key block, invariant part
+	add	@K[3],@K[3],$ONE		// not typo
+	str	@K[2],[sp,#32]
+	add	@K[4],@K[3],$ONE
+	add	@K[5],@K[4],$ONE
+	add	@K[6],@K[5],$ONE
+	shl	$ONE,$ONE,#2			// 1 -> 4
+
+	stp	d8,d9,[sp,#128+0]		// meet ABI requirements
+	stp	d10,d11,[sp,#128+16]
+	stp	d12,d13,[sp,#128+32]
+	stp	d14,d15,[sp,#128+48]
+
+	sub	$len,$len,#512			// not typo
+
+.Loop_outer_512_neon:
+	 mov	$A0,@K[0]
+	 mov	$A1,@K[0]
+	 mov	$A2,@K[0]
+	 mov	$A3,@K[0]
+	 mov	$A4,@K[0]
+	 mov	$A5,@K[0]
+	 mov	$B0,@K[1]
+	mov.32	@x[0],@d[0]			// unpack key block
+	 mov	$B1,@K[1]
+	lsr	@x[1],@d[0],#32
+	 mov	$B2,@K[1]
+	mov.32	@x[2],@d[1]
+	 mov	$B3,@K[1]
+	lsr	@x[3],@d[1],#32
+	 mov	$B4,@K[1]
+	mov.32	@x[4],@d[2]
+	 mov	$B5,@K[1]
+	lsr	@x[5],@d[2],#32
+	 mov	$D0,@K[3]
+	mov.32	@x[6],@d[3]
+	 mov	$D1,@K[4]
+	lsr	@x[7],@d[3],#32
+	 mov	$D2,@K[5]
+	mov.32	@x[8],@d[4]
+	 mov	$D3,@K[6]
+	lsr	@x[9],@d[4],#32
+	 mov	$C0,@K[2]
+	mov.32	@x[10],@d[5]
+	 mov	$C1,@K[2]
+	lsr	@x[11],@d[5],#32
+	 add	$D4,$D0,$ONE			// +4
+	mov.32	@x[12],@d[6]
+	 add	$D5,$D1,$ONE			// +4
+	lsr	@x[13],@d[6],#32
+	 mov	$C2,@K[2]
+	mov.32	@x[14],@d[7]
+	 mov	$C3,@K[2]
+	lsr	@x[15],@d[7],#32
+	 mov	$C4,@K[2]
+	 stp	@K[3],@K[4],[sp,#48]		// off-load key block, variable part
+	 mov	$C5,@K[2]
+	 str	@K[5],[sp,#80]
+
+	mov	$ctr,#5
+	subs	$len,$len,#512
+.Loop_upper_neon:
+	sub	$ctr,$ctr,#1
+___
+	my @thread0=&NEONROUND($A0,$B0,$C0,$D0,$T0,0);
+	my @thread1=&NEONROUND($A1,$B1,$C1,$D1,$T1,0);
+	my @thread2=&NEONROUND($A2,$B2,$C2,$D2,$T2,0);
+	my @thread3=&NEONROUND($A3,$B3,$C3,$D3,$T3,0);
+	my @thread4=&NEONROUND($A4,$B4,$C4,$D4,$T4,0);
+	my @thread5=&NEONROUND($A5,$B5,$C5,$D5,$T5,0);
+	my @thread67=(&ROUND(0,4,8,12),&ROUND(0,5,10,15));
+	my $diff = ($#thread0+1)*6 - $#thread67 - 1;
+	my $i = 0;
+
+	foreach (@thread0) {
+		eval;			eval(shift(@thread67));
+		eval(shift(@thread1));	eval(shift(@thread67));
+		eval(shift(@thread2));	eval(shift(@thread67));
+		eval(shift(@thread3));	eval(shift(@thread67));
+		eval(shift(@thread4));	eval(shift(@thread67));
+		eval(shift(@thread5));	eval(shift(@thread67));
+	}
+
+	@thread0=&NEONROUND($A0,$B0,$C0,$D0,$T0,1);
+	@thread1=&NEONROUND($A1,$B1,$C1,$D1,$T1,1);
+	@thread2=&NEONROUND($A2,$B2,$C2,$D2,$T2,1);
+	@thread3=&NEONROUND($A3,$B3,$C3,$D3,$T3,1);
+	@thread4=&NEONROUND($A4,$B4,$C4,$D4,$T4,1);
+	@thread5=&NEONROUND($A5,$B5,$C5,$D5,$T5,1);
+	@thread67=(&ROUND(0,4,8,12),&ROUND(0,5,10,15));
+
+	foreach (@thread0) {
+		eval;			eval(shift(@thread67));
+		eval(shift(@thread1));	eval(shift(@thread67));
+		eval(shift(@thread2));	eval(shift(@thread67));
+		eval(shift(@thread3));	eval(shift(@thread67));
+		eval(shift(@thread4));	eval(shift(@thread67));
+		eval(shift(@thread5));	eval(shift(@thread67));
+	}
+$code.=<<___;
+	cbnz	$ctr,.Loop_upper_neon
+
+	add.32	@x[0],@x[0],@d[0]		// accumulate key block
+	add	@x[1],@x[1],@d[0],lsr#32
+	add.32	@x[2],@x[2],@d[1]
+	add	@x[3],@x[3],@d[1],lsr#32
+	add.32	@x[4],@x[4],@d[2]
+	add	@x[5],@x[5],@d[2],lsr#32
+	add.32	@x[6],@x[6],@d[3]
+	add	@x[7],@x[7],@d[3],lsr#32
+	add.32	@x[8],@x[8],@d[4]
+	add	@x[9],@x[9],@d[4],lsr#32
+	add.32	@x[10],@x[10],@d[5]
+	add	@x[11],@x[11],@d[5],lsr#32
+	add.32	@x[12],@x[12],@d[6]
+	add	@x[13],@x[13],@d[6],lsr#32
+	add.32	@x[14],@x[14],@d[7]
+	add	@x[15],@x[15],@d[7],lsr#32
+
+	add	@x[0],@x[0],@x[1],lsl#32	// pack
+	add	@x[2],@x[2],@x[3],lsl#32
+	ldp	@x[1],@x[3],[$inp,#0]		// load input
+	add	@x[4],@x[4],@x[5],lsl#32
+	add	@x[6],@x[6],@x[7],lsl#32
+	ldp	@x[5],@x[7],[$inp,#16]
+	add	@x[8],@x[8],@x[9],lsl#32
+	add	@x[10],@x[10],@x[11],lsl#32
+	ldp	@x[9],@x[11],[$inp,#32]
+	add	@x[12],@x[12],@x[13],lsl#32
+	add	@x[14],@x[14],@x[15],lsl#32
+	ldp	@x[13],@x[15],[$inp,#48]
+	add	$inp,$inp,#64
+# ifdef	__AARCH64EB__
+	rev	@x[0],@x[0]
+	rev	@x[2],@x[2]
+	rev	@x[4],@x[4]
+	rev	@x[6],@x[6]
+	rev	@x[8],@x[8]
+	rev	@x[10],@x[10]
+	rev	@x[12],@x[12]
+	rev	@x[14],@x[14]
+# endif
+	eor	@x[0],@x[0],@x[1]
+	eor	@x[2],@x[2],@x[3]
+	eor	@x[4],@x[4],@x[5]
+	eor	@x[6],@x[6],@x[7]
+	eor	@x[8],@x[8],@x[9]
+	eor	@x[10],@x[10],@x[11]
+	eor	@x[12],@x[12],@x[13]
+	eor	@x[14],@x[14],@x[15]
+
+	 stp	@x[0],@x[2],[$out,#0]		// store output
+	 add	@d[6],@d[6],#1			// increment counter
+	mov.32	@x[0],@d[0]			// unpack key block
+	lsr	@x[1],@d[0],#32
+	 stp	@x[4],@x[6],[$out,#16]
+	mov.32	@x[2],@d[1]
+	lsr	@x[3],@d[1],#32
+	 stp	@x[8],@x[10],[$out,#32]
+	mov.32	@x[4],@d[2]
+	lsr	@x[5],@d[2],#32
+	 stp	@x[12],@x[14],[$out,#48]
+	 add	$out,$out,#64
+	mov.32	@x[6],@d[3]
+	lsr	@x[7],@d[3],#32
+	mov.32	@x[8],@d[4]
+	lsr	@x[9],@d[4],#32
+	mov.32	@x[10],@d[5]
+	lsr	@x[11],@d[5],#32
+	mov.32	@x[12],@d[6]
+	lsr	@x[13],@d[6],#32
+	mov.32	@x[14],@d[7]
+	lsr	@x[15],@d[7],#32
+
+	mov	$ctr,#5
+.Loop_lower_neon:
+	sub	$ctr,$ctr,#1
+___
+	@thread0=&NEONROUND($A0,$B0,$C0,$D0,$T0,0);
+	@thread1=&NEONROUND($A1,$B1,$C1,$D1,$T1,0);
+	@thread2=&NEONROUND($A2,$B2,$C2,$D2,$T2,0);
+	@thread3=&NEONROUND($A3,$B3,$C3,$D3,$T3,0);
+	@thread4=&NEONROUND($A4,$B4,$C4,$D4,$T4,0);
+	@thread5=&NEONROUND($A5,$B5,$C5,$D5,$T5,0);
+	@thread67=(&ROUND(0,4,8,12),&ROUND(0,5,10,15));
+
+	foreach (@thread0) {
+		eval;			eval(shift(@thread67));
+		eval(shift(@thread1));	eval(shift(@thread67));
+		eval(shift(@thread2));	eval(shift(@thread67));
+		eval(shift(@thread3));	eval(shift(@thread67));
+		eval(shift(@thread4));	eval(shift(@thread67));
+		eval(shift(@thread5));	eval(shift(@thread67));
+	}
+
+	@thread0=&NEONROUND($A0,$B0,$C0,$D0,$T0,1);
+	@thread1=&NEONROUND($A1,$B1,$C1,$D1,$T1,1);
+	@thread2=&NEONROUND($A2,$B2,$C2,$D2,$T2,1);
+	@thread3=&NEONROUND($A3,$B3,$C3,$D3,$T3,1);
+	@thread4=&NEONROUND($A4,$B4,$C4,$D4,$T4,1);
+	@thread5=&NEONROUND($A5,$B5,$C5,$D5,$T5,1);
+	@thread67=(&ROUND(0,4,8,12),&ROUND(0,5,10,15));
+
+	foreach (@thread0) {
+		eval;			eval(shift(@thread67));
+		eval(shift(@thread1));	eval(shift(@thread67));
+		eval(shift(@thread2));	eval(shift(@thread67));
+		eval(shift(@thread3));	eval(shift(@thread67));
+		eval(shift(@thread4));	eval(shift(@thread67));
+		eval(shift(@thread5));	eval(shift(@thread67));
+	}
+$code.=<<___;
+	cbnz	$ctr,.Loop_lower_neon
+
+	add.32	@x[0],@x[0],@d[0]		// accumulate key block
+	 ldp	@K[0],@K[1],[sp,#0]
+	add	@x[1],@x[1],@d[0],lsr#32
+	 ldp	@K[2],@K[3],[sp,#32]
+	add.32	@x[2],@x[2],@d[1]
+	 ldp	@K[4],@K[5],[sp,#64]
+	add	@x[3],@x[3],@d[1],lsr#32
+	 add	$A0,$A0,@K[0]
+	add.32	@x[4],@x[4],@d[2]
+	 add	$A1,$A1,@K[0]
+	add	@x[5],@x[5],@d[2],lsr#32
+	 add	$A2,$A2,@K[0]
+	add.32	@x[6],@x[6],@d[3]
+	 add	$A3,$A3,@K[0]
+	add	@x[7],@x[7],@d[3],lsr#32
+	 add	$A4,$A4,@K[0]
+	add.32	@x[8],@x[8],@d[4]
+	 add	$A5,$A5,@K[0]
+	add	@x[9],@x[9],@d[4],lsr#32
+	 add	$C0,$C0,@K[2]
+	add.32	@x[10],@x[10],@d[5]
+	 add	$C1,$C1,@K[2]
+	add	@x[11],@x[11],@d[5],lsr#32
+	 add	$C2,$C2,@K[2]
+	add.32	@x[12],@x[12],@d[6]
+	 add	$C3,$C3,@K[2]
+	add	@x[13],@x[13],@d[6],lsr#32
+	 add	$C4,$C4,@K[2]
+	add.32	@x[14],@x[14],@d[7]
+	 add	$C5,$C5,@K[2]
+	add	@x[15],@x[15],@d[7],lsr#32
+	 add	$D4,$D4,$ONE			// +4
+	add	@x[0],@x[0],@x[1],lsl#32	// pack
+	 add	$D5,$D5,$ONE			// +4
+	add	@x[2],@x[2],@x[3],lsl#32
+	 add	$D0,$D0,@K[3]
+	ldp	@x[1],@x[3],[$inp,#0]		// load input
+	 add	$D1,$D1,@K[4]
+	add	@x[4],@x[4],@x[5],lsl#32
+	 add	$D2,$D2,@K[5]
+	add	@x[6],@x[6],@x[7],lsl#32
+	 add	$D3,$D3,@K[6]
+	ldp	@x[5],@x[7],[$inp,#16]
+	 add	$D4,$D4,@K[3]
+	add	@x[8],@x[8],@x[9],lsl#32
+	 add	$D5,$D5,@K[4]
+	add	@x[10],@x[10],@x[11],lsl#32
+	 add	$B0,$B0,@K[1]
+	ldp	@x[9],@x[11],[$inp,#32]
+	 add	$B1,$B1,@K[1]
+	add	@x[12],@x[12],@x[13],lsl#32
+	 add	$B2,$B2,@K[1]
+	add	@x[14],@x[14],@x[15],lsl#32
+	 add	$B3,$B3,@K[1]
+	ldp	@x[13],@x[15],[$inp,#48]
+	 add	$B4,$B4,@K[1]
+	add	$inp,$inp,#64
+	 add	$B5,$B5,@K[1]
+
+# ifdef	__AARCH64EB__
+	rev	@x[0],@x[0]
+	rev	@x[2],@x[2]
+	rev	@x[4],@x[4]
+	rev	@x[6],@x[6]
+	rev	@x[8],@x[8]
+	rev	@x[10],@x[10]
+	rev	@x[12],@x[12]
+	rev	@x[14],@x[14]
+# endif
+	ld1.8	{$T0-$T3},[$inp],#64
+	eor	@x[0],@x[0],@x[1]
+	eor	@x[2],@x[2],@x[3]
+	eor	@x[4],@x[4],@x[5]
+	eor	@x[6],@x[6],@x[7]
+	eor	@x[8],@x[8],@x[9]
+	 eor	$A0,$A0,$T0
+	eor	@x[10],@x[10],@x[11]
+	 eor	$B0,$B0,$T1
+	eor	@x[12],@x[12],@x[13]
+	 eor	$C0,$C0,$T2
+	eor	@x[14],@x[14],@x[15]
+	 eor	$D0,$D0,$T3
+	 ld1.8	{$T0-$T3},[$inp],#64
+
+	stp	@x[0],@x[2],[$out,#0]		// store output
+	 add	@d[6],@d[6],#7			// increment counter
+	stp	@x[4],@x[6],[$out,#16]
+	stp	@x[8],@x[10],[$out,#32]
+	stp	@x[12],@x[14],[$out,#48]
+	add	$out,$out,#64
+	st1.8	{$A0-$D0},[$out],#64
+
+	ld1.8	{$A0-$D0},[$inp],#64
+	eor	$A1,$A1,$T0
+	eor	$B1,$B1,$T1
+	eor	$C1,$C1,$T2
+	eor	$D1,$D1,$T3
+	st1.8	{$A1-$D1},[$out],#64
+
+	ld1.8	{$A1-$D1},[$inp],#64
+	eor	$A2,$A2,$A0
+	 ldp	@K[0],@K[1],[sp,#0]
+	eor	$B2,$B2,$B0
+	 ldp	@K[2],@K[3],[sp,#32]
+	eor	$C2,$C2,$C0
+	eor	$D2,$D2,$D0
+	st1.8	{$A2-$D2},[$out],#64
+
+	ld1.8	{$A2-$D2},[$inp],#64
+	eor	$A3,$A3,$A1
+	eor	$B3,$B3,$B1
+	eor	$C3,$C3,$C1
+	eor	$D3,$D3,$D1
+	st1.8	{$A3-$D3},[$out],#64
+
+	ld1.8	{$A3-$D3},[$inp],#64
+	eor	$A4,$A4,$A2
+	eor	$B4,$B4,$B2
+	eor	$C4,$C4,$C2
+	eor	$D4,$D4,$D2
+	st1.8	{$A4-$D4},[$out],#64
+
+	shl	$A0,$ONE,#1			// 4 -> 8
+	eor	$A5,$A5,$A3
+	eor	$B5,$B5,$B3
+	eor	$C5,$C5,$C3
+	eor	$D5,$D5,$D3
+	st1.8	{$A5-$D5},[$out],#64
+
+	add	@K[3],@K[3],$A0			// += 8
+	add	@K[4],@K[4],$A0
+	add	@K[5],@K[5],$A0
+	add	@K[6],@K[6],$A0
+
+	b.hs	.Loop_outer_512_neon
+
+	adds	$len,$len,#512
+	ushr	$A0,$ONE,#2			// 4 -> 1
+
+	ldp	d8,d9,[sp,#128+0]		// meet ABI requirements
+	ldp	d10,d11,[sp,#128+16]
+	ldp	d12,d13,[sp,#128+32]
+	ldp	d14,d15,[sp,#128+48]
+
+	stp	@K[0],$ONE,[sp,#0]		// wipe off-load area
+	stp	@K[0],$ONE,[sp,#32]
+	stp	@K[0],$ONE,[sp,#64]
+
+	b.eq	.Ldone_512_neon
+
+	cmp	$len,#192
+	sub	@K[3],@K[3],$A0			// -= 1
+	sub	@K[4],@K[4],$A0
+	sub	@K[5],@K[5],$A0
+	add	sp,sp,#128
+	b.hs	.Loop_outer_neon
+
+	eor	@K[1],@K[1],@K[1]
+	eor	@K[2],@K[2],@K[2]
+	eor	@K[3],@K[3],@K[3]
+	eor	@K[4],@K[4],@K[4]
+	eor	@K[5],@K[5],@K[5]
+	eor	@K[6],@K[6],@K[6]
+	b	.Loop_outer
+
+.Ldone_512_neon:
+	ldp	x19,x20,[x29,#16]
+	add	sp,sp,#128+64
+	ldp	x21,x22,[x29,#32]
+	ldp	x23,x24,[x29,#48]
+	ldp	x25,x26,[x29,#64]
+	ldp	x27,x28,[x29,#80]
+	ldp	x29,x30,[sp],#96
+	ret
+.size	ChaCha20_512_neon,.-ChaCha20_512_neon
+#endif
+#endif
+___
+}
+}}}
+
+open SELF,$0;
+while(<SELF>) {
+	next if (/^#!/);
+	last if (!s/^#/\/\// and !/^$/);
+	print;
+}
+close SELF;
+
+foreach (split("\n",$code)) {
+	s/\`([^\`]*)\`/eval $1/geo;
+
+	(s/\b([a-z]+)\.32\b/$1/ and (s/x([0-9]+)/w$1/g or 1))	or
+	(m/\b(eor|ext|mov)\b/ and (s/\.4s/\.16b/g or 1))	or
+	(s/\b((?:ld|st)1)\.8\b/$1/ and (s/\.4s/\.16b/g or 1))	or
+	(m/\b(ld|st)[rp]\b/ and (s/v([0-9]+)\.4s/q$1/g or 1))	or
+	(s/\brev32\.16\b/rev32/ and (s/\.4s/\.8h/g or 1));
+
+	print $_,"\n";
+}
+close STDOUT;	# flush

@@ -0,0 +1,461 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+/*
+ * Copyright (C) 2018 Google, Inc.
+ */
+
+#include <linux/linkage.h>
+#include <asm/assembler.h>
+
+/*
+ * Design notes:
+ *
+ * 16 registers would be needed to hold the state matrix, but only 14 are
+ * available because 'sp' and 'pc' cannot be used.  So we spill the elements
+ * (x8, x9) to the stack and swap them out with (x10, x11).  This adds one
+ * 'ldrd' and one 'strd' instruction per round.
+ *
+ * All rotates are performed using the implicit rotate operand accepted by the
+ * 'add' and 'eor' instructions.  This is faster than using explicit rotate
+ * instructions.  To make this work, we allow the values in the second and last
+ * rows of the ChaCha state matrix (rows 'b' and 'd') to temporarily have the
+ * wrong rotation amount.  The rotation amount is then fixed up just in time
+ * when the values are used.  'brot' is the number of bits the values in row 'b'
+ * need to be rotated right to arrive at the correct values, and 'drot'
+ * similarly for row 'd'.  (brot, drot) start out as (0, 0) but we make it such
+ * that they end up as (25, 24) after every round.
+ */
+
+	// ChaCha state registers
+	X0	.req	r0
+	X1	.req	r1
+	X2	.req	r2
+	X3	.req	r3
+	X4	.req	r4
+	X5	.req	r5
+	X6	.req	r6
+	X7	.req	r7
+	X8_X10	.req	r8	// shared by x8 and x10
+	X9_X11	.req	r9	// shared by x9 and x11
+	X12	.req	r10
+	X13	.req	r11
+	X14	.req	r12
+	X15	.req	r14
+
+.Lexpand_32byte_k:
+	// "expand 32-byte k"
+	.word	0x61707865, 0x3320646e, 0x79622d32, 0x6b206574
+
+#ifdef __thumb2__
+#  define adrl adr
+#endif
+
+.macro __rev		out, in,  t0, t1, t2
+.if __LINUX_ARM_ARCH__ >= 6
+	rev		\out, \in
+.else
+	lsl		\t0, \in, #24
+	and		\t1, \in, #0xff00
+	and		\t2, \in, #0xff0000
+	orr		\out, \t0, \in, lsr #24
+	orr		\out, \out, \t1, lsl #8
+	orr		\out, \out, \t2, lsr #8
+.endif
+.endm
+
+.macro _le32_bswap	x,  t0, t1, t2
+#ifdef __ARMEB__
+	__rev		\x, \x,  \t0, \t1, \t2
+#endif
+.endm
+
+.macro _le32_bswap_4x	a, b, c, d,  t0, t1, t2
+	_le32_bswap	\a,  \t0, \t1, \t2
+	_le32_bswap	\b,  \t0, \t1, \t2
+	_le32_bswap	\c,  \t0, \t1, \t2
+	_le32_bswap	\d,  \t0, \t1, \t2
+.endm
+
+.macro __ldrd		a, b, src, offset
+#if __LINUX_ARM_ARCH__ >= 6
+	ldrd		\a, \b, [\src, #\offset]
+#else
+	ldr		\a, [\src, #\offset]
+	ldr		\b, [\src, #\offset + 4]
+#endif
+.endm
+
+.macro __strd		a, b, dst, offset
+#if __LINUX_ARM_ARCH__ >= 6
+	strd		\a, \b, [\dst, #\offset]
+#else
+	str		\a, [\dst, #\offset]
+	str		\b, [\dst, #\offset + 4]
+#endif
+.endm
+
+.macro _halfround	a1, b1, c1, d1,  a2, b2, c2, d2
+
+	// a += b; d ^= a; d = rol(d, 16);
+	add		\a1, \a1, \b1, ror #brot
+	add		\a2, \a2, \b2, ror #brot
+	eor		\d1, \a1, \d1, ror #drot
+	eor		\d2, \a2, \d2, ror #drot
+	// drot == 32 - 16 == 16
+
+	// c += d; b ^= c; b = rol(b, 12);
+	add		\c1, \c1, \d1, ror #16
+	add		\c2, \c2, \d2, ror #16
+	eor		\b1, \c1, \b1, ror #brot
+	eor		\b2, \c2, \b2, ror #brot
+	// brot == 32 - 12 == 20
+
+	// a += b; d ^= a; d = rol(d, 8);
+	add		\a1, \a1, \b1, ror #20
+	add		\a2, \a2, \b2, ror #20
+	eor		\d1, \a1, \d1, ror #16
+	eor		\d2, \a2, \d2, ror #16
+	// drot == 32 - 8 == 24
+
+	// c += d; b ^= c; b = rol(b, 7);
+	add		\c1, \c1, \d1, ror #24
+	add		\c2, \c2, \d2, ror #24
+	eor		\b1, \c1, \b1, ror #20
+	eor		\b2, \c2, \b2, ror #20
+	// brot == 32 - 7 == 25
+.endm
+
+.macro _doubleround
+
+	// column round
+
+	// quarterrounds: (x0, x4, x8, x12) and (x1, x5, x9, x13)
+	_halfround	X0, X4, X8_X10, X12,  X1, X5, X9_X11, X13
+
+	// save (x8, x9); restore (x10, x11)
+	__strd		X8_X10, X9_X11, sp, 0
+	__ldrd		X8_X10, X9_X11, sp, 8
+
+	// quarterrounds: (x2, x6, x10, x14) and (x3, x7, x11, x15)
+	_halfround	X2, X6, X8_X10, X14,  X3, X7, X9_X11, X15
+
+	.set brot, 25
+	.set drot, 24
+
+	// diagonal round
+
+	// quarterrounds: (x0, x5, x10, x15) and (x1, x6, x11, x12)
+	_halfround	X0, X5, X8_X10, X15,  X1, X6, X9_X11, X12
+
+	// save (x10, x11); restore (x8, x9)
+	__strd		X8_X10, X9_X11, sp, 8
+	__ldrd		X8_X10, X9_X11, sp, 0
+
+	// quarterrounds: (x2, x7, x8, x13) and (x3, x4, x9, x14)
+	_halfround	X2, X7, X8_X10, X13,  X3, X4, X9_X11, X14
+.endm
+
+.macro _chacha_permute	nrounds
+	.set brot, 0
+	.set drot, 0
+	.rept \nrounds / 2
+	 _doubleround
+	.endr
+.endm
+
+.macro _chacha		nrounds
+
+.Lnext_block\@:
+	// Stack: unused0-unused1 x10-x11 x0-x15 OUT IN LEN
+	// Registers contain x0-x9,x12-x15.
+
+	// Do the core ChaCha permutation to update x0-x15.
+	_chacha_permute	\nrounds
+
+	add		sp, #8
+	// Stack: x10-x11 orig_x0-orig_x15 OUT IN LEN
+	// Registers contain x0-x9,x12-x15.
+	// x4-x7 are rotated by 'brot'; x12-x15 are rotated by 'drot'.
+
+	// Free up some registers (r8-r12,r14) by pushing (x8-x9,x12-x15).
+	push		{X8_X10, X9_X11, X12, X13, X14, X15}
+
+	// Load (OUT, IN, LEN).
+	ldr		r14, [sp, #96]
+	ldr		r12, [sp, #100]
+	ldr		r11, [sp, #104]
+
+	orr		r10, r14, r12
+
+	// Use slow path if fewer than 64 bytes remain.
+	cmp		r11, #64
+	blt		.Lxor_slowpath\@
+
+	// Use slow path if IN and/or OUT isn't 4-byte aligned.  Needed even on
+	// ARMv6+, since ldmia and stmia (used below) still require alignment.
+	tst		r10, #3
+	bne		.Lxor_slowpath\@
+
+	// Fast path: XOR 64 bytes of aligned data.
+
+	// Stack: x8-x9 x12-x15 x10-x11 orig_x0-orig_x15 OUT IN LEN
+	// Registers: r0-r7 are x0-x7; r8-r11 are free; r12 is IN; r14 is OUT.
+	// x4-x7 are rotated by 'brot'; x12-x15 are rotated by 'drot'.
+
+	// x0-x3
+	__ldrd		r8, r9, sp, 32
+	__ldrd		r10, r11, sp, 40
+	add		X0, X0, r8
+	add		X1, X1, r9
+	add		X2, X2, r10
+	add		X3, X3, r11
+	_le32_bswap_4x	X0, X1, X2, X3,  r8, r9, r10
+	ldmia		r12!, {r8-r11}
+	eor		X0, X0, r8
+	eor		X1, X1, r9
+	eor		X2, X2, r10
+	eor		X3, X3, r11
+	stmia		r14!, {X0-X3}
+
+	// x4-x7
+	__ldrd		r8, r9, sp, 48
+	__ldrd		r10, r11, sp, 56
+	add		X4, r8, X4, ror #brot
+	add		X5, r9, X5, ror #brot
+	ldmia		r12!, {X0-X3}
+	add		X6, r10, X6, ror #brot
+	add		X7, r11, X7, ror #brot
+	_le32_bswap_4x	X4, X5, X6, X7,  r8, r9, r10
+	eor		X4, X4, X0
+	eor		X5, X5, X1
+	eor		X6, X6, X2
+	eor		X7, X7, X3
+	stmia		r14!, {X4-X7}
+
+	// x8-x15
+	pop		{r0-r7}			// (x8-x9,x12-x15,x10-x11)
+	__ldrd		r8, r9, sp, 32
+	__ldrd		r10, r11, sp, 40
+	add		r0, r0, r8		// x8
+	add		r1, r1, r9		// x9
+	add		r6, r6, r10		// x10
+	add		r7, r7, r11		// x11
+	_le32_bswap_4x	r0, r1, r6, r7,  r8, r9, r10
+	ldmia		r12!, {r8-r11}
+	eor		r0, r0, r8		// x8
+	eor		r1, r1, r9		// x9
+	eor		r6, r6, r10		// x10
+	eor		r7, r7, r11		// x11
+	stmia		r14!, {r0,r1,r6,r7}
+	ldmia		r12!, {r0,r1,r6,r7}
+	__ldrd		r8, r9, sp, 48
+	__ldrd		r10, r11, sp, 56
+	add		r2, r8, r2, ror #drot	// x12
+	add		r3, r9, r3, ror #drot	// x13
+	add		r4, r10, r4, ror #drot	// x14
+	add		r5, r11, r5, ror #drot	// x15
+	_le32_bswap_4x	r2, r3, r4, r5,  r9, r10, r11
+	  ldr		r9, [sp, #72]		// load LEN
+	eor		r2, r2, r0		// x12
+	eor		r3, r3, r1		// x13
+	eor		r4, r4, r6		// x14
+	eor		r5, r5, r7		// x15
+	  subs		r9, #64			// decrement and check LEN
+	stmia		r14!, {r2-r5}
+
+	beq		.Ldone\@
+
+.Lprepare_for_next_block\@:
+
+	// Stack: x0-x15 OUT IN LEN
+
+	// Increment block counter (x12)
+	add		r8, #1
+
+	// Store updated (OUT, IN, LEN)
+	str		r14, [sp, #64]
+	str		r12, [sp, #68]
+	str		r9, [sp, #72]
+
+	  mov		r14, sp
+
+	// Store updated block counter (x12)
+	str		r8, [sp, #48]
+
+	  sub		sp, #16
+
+	// Reload state and do next block
+	ldmia		r14!, {r0-r11}		// load x0-x11
+	__strd		r10, r11, sp, 8		// store x10-x11 before state
+	ldmia		r14, {r10-r12,r14}	// load x12-x15
+	b		.Lnext_block\@
+
+.Lxor_slowpath\@:
+	// Slow path: < 64 bytes remaining, or unaligned input or output buffer.
+	// We handle it by storing the 64 bytes of keystream to the stack, then
+	// XOR-ing the needed portion with the data.
+
+	// Allocate keystream buffer
+	sub		sp, #64
+	mov		r14, sp
+
+	// Stack: ks0-ks15 x8-x9 x12-x15 x10-x11 orig_x0-orig_x15 OUT IN LEN
+	// Registers: r0-r7 are x0-x7; r8-r11 are free; r12 is IN; r14 is &ks0.
+	// x4-x7 are rotated by 'brot'; x12-x15 are rotated by 'drot'.
+
+	// Save keystream for x0-x3
+	__ldrd		r8, r9, sp, 96
+	__ldrd		r10, r11, sp, 104
+	add		X0, X0, r8
+	add		X1, X1, r9
+	add		X2, X2, r10
+	add		X3, X3, r11
+	_le32_bswap_4x	X0, X1, X2, X3,  r8, r9, r10
+	stmia		r14!, {X0-X3}
+
+	// Save keystream for x4-x7
+	__ldrd		r8, r9, sp, 112
+	__ldrd		r10, r11, sp, 120
+	add		X4, r8, X4, ror #brot
+	add		X5, r9, X5, ror #brot
+	add		X6, r10, X6, ror #brot
+	add		X7, r11, X7, ror #brot
+	_le32_bswap_4x	X4, X5, X6, X7,  r8, r9, r10
+	  add		r8, sp, #64
+	stmia		r14!, {X4-X7}
+
+	// Save keystream for x8-x15
+	ldm		r8, {r0-r7}		// (x8-x9,x12-x15,x10-x11)
+	__ldrd		r8, r9, sp, 128
+	__ldrd		r10, r11, sp, 136
+	add		r0, r0, r8		// x8
+	add		r1, r1, r9		// x9
+	add		r6, r6, r10		// x10
+	add		r7, r7, r11		// x11
+	_le32_bswap_4x	r0, r1, r6, r7,  r8, r9, r10
+	stmia		r14!, {r0,r1,r6,r7}
+	__ldrd		r8, r9, sp, 144
+	__ldrd		r10, r11, sp, 152
+	add		r2, r8, r2, ror #drot	// x12
+	add		r3, r9, r3, ror #drot	// x13
+	add		r4, r10, r4, ror #drot	// x14
+	add		r5, r11, r5, ror #drot	// x15
+	_le32_bswap_4x	r2, r3, r4, r5,  r9, r10, r11
+	stmia		r14, {r2-r5}
+
+	// Stack: ks0-ks15 unused0-unused7 x0-x15 OUT IN LEN
+	// Registers: r8 is block counter, r12 is IN.
+
+	ldr		r9, [sp, #168]		// LEN
+	ldr		r14, [sp, #160]		// OUT
+	cmp		r9, #64
+	  mov		r0, sp
+	movle		r1, r9
+	movgt		r1, #64
+	// r1 is number of bytes to XOR, in range [1, 64]
+
+.if __LINUX_ARM_ARCH__ < 6
+	orr		r2, r12, r14
+	tst		r2, #3			// IN or OUT misaligned?
+	bne		.Lxor_next_byte\@
+.endif
+
+	// XOR a word at a time
+.rept 16
+	subs		r1, #4
+	blt		.Lxor_words_done\@
+	ldr		r2, [r12], #4
+	ldr		r3, [r0], #4
+	eor		r2, r2, r3
+	str		r2, [r14], #4
+.endr
+	b		.Lxor_slowpath_done\@
+.Lxor_words_done\@:
+	ands		r1, r1, #3
+	beq		.Lxor_slowpath_done\@
+
+	// XOR a byte at a time
+.Lxor_next_byte\@:
+	ldrb		r2, [r12], #1
+	ldrb		r3, [r0], #1
+	eor		r2, r2, r3
+	strb		r2, [r14], #1
+	subs		r1, #1
+	bne		.Lxor_next_byte\@
+
+.Lxor_slowpath_done\@:
+	subs		r9, #64
+	add		sp, #96
+	bgt		.Lprepare_for_next_block\@
+
+.Ldone\@:
+.endm	// _chacha
+
+/*
+ * void chacha20_arm(u8 *out, const u8 *in, size_t len, const u32 key[8],
+ *		     const u32 iv[4]);
+ */
+ENTRY(chacha20_arm)
+	cmp		r2, #0			// len == 0?
+	reteq		lr
+
+	push		{r0-r2,r4-r11,lr}
+
+	// Push state x0-x15 onto stack.
+	// Also store an extra copy of x10-x11 just before the state.
+
+	ldr		r4, [sp, #48]		// iv
+	mov		r0, sp
+	sub		sp, #80
+
+	// iv: x12-x15
+	ldm		r4, {X12,X13,X14,X15}
+	stmdb		r0!, {X12,X13,X14,X15}
+
+	// key: x4-x11
+	__ldrd		X8_X10, X9_X11, r3, 24
+	__strd		X8_X10, X9_X11, sp, 8
+	stmdb		r0!, {X8_X10, X9_X11}
+	ldm		r3, {X4-X9_X11}
+	stmdb		r0!, {X4-X9_X11}
+
+	// constants: x0-x3
+	adrl		X3, .Lexpand_32byte_k
+	ldm		X3, {X0-X3}
+	__strd		X0, X1, sp, 16
+	__strd		X2, X3, sp, 24
+
+	_chacha		20
+
+	add		sp, #76
+	pop		{r4-r11, pc}
+ENDPROC(chacha20_arm)
+
+/*
+ * void hchacha20_arm(const u32 state[16], u32 out[8]);
+ */
+ENTRY(hchacha20_arm)
+	push		{r1,r4-r11,lr}
+
+	mov		r14, r0
+	ldmia		r14!, {r0-r11}		// load x0-x11
+	push		{r10-r11}		// store x10-x11 to stack
+	ldm		r14, {r10-r12,r14}	// load x12-x15
+	sub		sp, #8
+
+	_chacha_permute	20
+
+	// Skip over (unused0-unused1, x10-x11)
+	add		sp, #16
+
+	// Fix up rotations of x12-x15
+	ror		X12, X12, #drot
+	ror		X13, X13, #drot
+	  pop		{r4}			// load 'out'
+	ror		X14, X14, #drot
+	ror		X15, X15, #drot
+
+	// Store (x0-x3,x12-x15) to 'out'
+	stm		r4, {X0,X1,X2,X3,X12,X13,X14,X15}
+
+	pop		{r4-r11,pc}
+ENDPROC(hchacha20_arm)

@@ -18,6 +18,8 @@
 
 #if defined(CONFIG_ZINC_ARCH_X86_64)
 #include "chacha20-x86_64-glue.c"
+#elif defined(CONFIG_ZINC_ARCH_ARM) || defined(CONFIG_ZINC_ARCH_ARM64)
+#include "chacha20-arm-glue.c"
 #else
 static bool *const chacha20_nobs[] __initconst = { };
 static void __init chacha20_fpu_init(void)