Update vendored x/crypto/poly1305

- Adds "poly1305: drop broken arm assembly"

vendor/golang.org/x/crypto/poly1305/bits_compat.go

@@ -0,0 +1,39 @@
+// Copyright 2019 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+// +build !go1.13
+
+package poly1305
+
+// Generic fallbacks for the math/bits intrinsics, copied from
+// src/math/bits/bits.go. They were added in Go 1.12, but Add64 and Sum64 had
+// variable time fallbacks until Go 1.13.
+
+func bitsAdd64(x, y, carry uint64) (sum, carryOut uint64) {
+	sum = x + y + carry
+	carryOut = ((x & y) | ((x | y) &^ sum)) >> 63
+	return
+}
+
+func bitsSub64(x, y, borrow uint64) (diff, borrowOut uint64) {
+	diff = x - y - borrow
+	borrowOut = ((^x & y) | (^(x ^ y) & diff)) >> 63
+	return
+}
+
+func bitsMul64(x, y uint64) (hi, lo uint64) {
+	const mask32 = 1<<32 - 1
+	x0 := x & mask32
+	x1 := x >> 32
+	y0 := y & mask32
+	y1 := y >> 32
+	w0 := x0 * y0
+	t := x1*y0 + w0>>32
+	w1 := t & mask32
+	w2 := t >> 32
+	w1 += x0 * y1
+	hi = x1*y1 + w2 + w1>>32
+	lo = x * y
+	return
+}

vendor/golang.org/x/crypto/poly1305/bits_go1.13.go

@@ -0,0 +1,21 @@
+// Copyright 2019 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+// +build go1.13
+
+package poly1305
+
+import "math/bits"
+
+func bitsAdd64(x, y, carry uint64) (sum, carryOut uint64) {
+	return bits.Add64(x, y, carry)
+}
+
+func bitsSub64(x, y, borrow uint64) (diff, borrowOut uint64) {
+	return bits.Sub64(x, y, borrow)
+}
+
+func bitsMul64(x, y uint64) (hi, lo uint64) {
+	return bits.Mul64(x, y)
+}

vendor/golang.org/x/crypto/poly1305/mac_noasm.go

@@ -2,7 +2,7 @@
 // Use of this source code is governed by a BSD-style
 // license that can be found in the LICENSE file.
 
-// +build !amd64 gccgo appengine
+// +build !amd64,!ppc64le gccgo appengine
 
 package poly1305
 

vendor/golang.org/x/crypto/poly1305/poly1305.go

@@ -22,8 +22,14 @@ import "crypto/subtle"
 // TagSize is the size, in bytes, of a poly1305 authenticator.
 const TagSize = 16
 
-// Verify returns true if mac is a valid authenticator for m with the given
-// key.
+// Sum generates an authenticator for msg using a one-time key and puts the
+// 16-byte result into out. Authenticating two different messages with the same
+// key allows an attacker to forge messages at will.
+func Sum(out *[16]byte, m []byte, key *[32]byte) {
+	sum(out, m, key)
+}
+
+// Verify returns true if mac is a valid authenticator for m with the given key.
 func Verify(mac *[16]byte, m []byte, key *[32]byte) bool {
 	var tmp [16]byte
 	Sum(&tmp, m, key)

vendor/golang.org/x/crypto/poly1305/sum_amd64.go

@@ -7,62 +7,52 @@
 package poly1305
 
 //go:noescape
-func initialize(state *[7]uint64, key *[32]byte)
+func update(state *macState, msg []byte)
 
-//go:noescape
-func update(state *[7]uint64, msg []byte)
-
-//go:noescape
-func finalize(tag *[TagSize]byte, state *[7]uint64)
-
-// Sum generates an authenticator for m using a one-time key and puts the
-// 16-byte result into out. Authenticating two different messages with the same
-// key allows an attacker to forge messages at will.
-func Sum(out *[16]byte, m []byte, key *[32]byte) {
+func sum(out *[16]byte, m []byte, key *[32]byte) {
 	h := newMAC(key)
 	h.Write(m)
 	h.Sum(out)
 }
 
 func newMAC(key *[32]byte) (h mac) {
-	initialize(&h.state, key)
+	initialize(key, &h.r, &h.s)
 	return
 }
 
-type mac struct {
-	state [7]uint64 // := uint64{ h0, h1, h2, r0, r1, pad0, pad1 }
-
-	buffer [TagSize]byte
-	offset int
-}
+// mac is a wrapper for macGeneric that redirects calls that would have gone to
+// updateGeneric to update.
+//
+// Its Write and Sum methods are otherwise identical to the macGeneric ones, but
+// using function pointers would carry a major performance cost.
+type mac struct{ macGeneric }
 
-func (h *mac) Write(p []byte) (n int, err error) {
-	n = len(p)
+func (h *mac) Write(p []byte) (int, error) {
+	nn := len(p)
 	if h.offset > 0 {
-		remaining := TagSize - h.offset
-		if n < remaining {
-			h.offset += copy(h.buffer[h.offset:], p)
-			return n, nil
+		n := copy(h.buffer[h.offset:], p)
+		if h.offset+n < TagSize {
+			h.offset += n
+			return nn, nil
 		}
-		copy(h.buffer[h.offset:], p[:remaining])
-		p = p[remaining:]
+		p = p[n:]
 		h.offset = 0
-		update(&h.state, h.buffer[:])
+		update(&h.macState, h.buffer[:])
 	}
-	if nn := len(p) - (len(p) % TagSize); nn > 0 {
-		update(&h.state, p[:nn])
-		p = p[nn:]
+	if n := len(p) - (len(p) % TagSize); n > 0 {
+		update(&h.macState, p[:n])
+		p = p[n:]
 	}
 	if len(p) > 0 {
 		h.offset += copy(h.buffer[h.offset:], p)
 	}
-	return n, nil
+	return nn, nil
 }
 
 func (h *mac) Sum(out *[16]byte) {
-	state := h.state
+	state := h.macState
 	if h.offset > 0 {
 		update(&state, h.buffer[:h.offset])
 	}
-	finalize(out, &state)
+	finalize(out, &state.h, &state.s)
 }

vendor/golang.org/x/crypto/poly1305/sum_amd64.s

@@ -54,10 +54,6 @@
 	ADCQ  t3, h1;                  \
 	ADCQ  $0, h2
 
-DATA ·poly1305Mask<>+0x00(SB)/8, $0x0FFFFFFC0FFFFFFF
-DATA ·poly1305Mask<>+0x08(SB)/8, $0x0FFFFFFC0FFFFFFC
-GLOBL ·poly1305Mask<>(SB), RODATA, $16
-
 // func update(state *[7]uint64, msg []byte)
 TEXT ·update(SB), $0-32
 	MOVQ state+0(FP), DI
@@ -110,39 +106,3 @@ done:
 	MOVQ R9, 8(DI)
 	MOVQ R10, 16(DI)
 	RET
-
-// func initialize(state *[7]uint64, key *[32]byte)
-TEXT ·initialize(SB), $0-16
-	MOVQ state+0(FP), DI
-	MOVQ key+8(FP), SI
-
-	// state[0...7] is initialized with zero
-	MOVOU 0(SI), X0
-	MOVOU 16(SI), X1
-	MOVOU ·poly1305Mask<>(SB), X2
-	PAND  X2, X0
-	MOVOU X0, 24(DI)
-	MOVOU X1, 40(DI)
-	RET
-
-// func finalize(tag *[TagSize]byte, state *[7]uint64)
-TEXT ·finalize(SB), $0-16
-	MOVQ tag+0(FP), DI
-	MOVQ state+8(FP), SI
-
-	MOVQ    0(SI), AX
-	MOVQ    8(SI), BX
-	MOVQ    16(SI), CX
-	MOVQ    AX, R8
-	MOVQ    BX, R9
-	SUBQ    $0xFFFFFFFFFFFFFFFB, AX
-	SBBQ    $0xFFFFFFFFFFFFFFFF, BX
-	SBBQ    $3, CX
-	CMOVQCS R8, AX
-	CMOVQCS R9, BX
-	ADDQ    40(SI), AX
-	ADCQ    48(SI), BX
-
-	MOVQ AX, 0(DI)
-	MOVQ BX, 8(DI)
-	RET

vendor/golang.org/x/crypto/poly1305/sum_arm.go

@@ -1,22 +0,0 @@
-// Copyright 2015 The Go Authors. All rights reserved.
-// Use of this source code is governed by a BSD-style
-// license that can be found in the LICENSE file.
-
-// +build arm,!gccgo,!appengine,!nacl
-
-package poly1305
-
-// This function is implemented in sum_arm.s
-//go:noescape
-func poly1305_auth_armv6(out *[16]byte, m *byte, mlen uint32, key *[32]byte)
-
-// Sum generates an authenticator for m using a one-time key and puts the
-// 16-byte result into out. Authenticating two different messages with the same
-// key allows an attacker to forge messages at will.
-func Sum(out *[16]byte, m []byte, key *[32]byte) {
-	var mPtr *byte
-	if len(m) > 0 {
-		mPtr = &m[0]
-	}
-	poly1305_auth_armv6(out, mPtr, uint32(len(m)), key)
-}

vendor/golang.org/x/crypto/poly1305/sum_arm.s

@@ -1,427 +0,0 @@
-// Copyright 2015 The Go Authors. All rights reserved.
-// Use of this source code is governed by a BSD-style
-// license that can be found in the LICENSE file.
-
-// +build arm,!gccgo,!appengine,!nacl
-
-#include "textflag.h"
-
-// This code was translated into a form compatible with 5a from the public
-// domain source by Andrew Moon: github.com/floodyberry/poly1305-opt/blob/master/app/extensions/poly1305.
-
-DATA ·poly1305_init_constants_armv6<>+0x00(SB)/4, $0x3ffffff
-DATA ·poly1305_init_constants_armv6<>+0x04(SB)/4, $0x3ffff03
-DATA ·poly1305_init_constants_armv6<>+0x08(SB)/4, $0x3ffc0ff
-DATA ·poly1305_init_constants_armv6<>+0x0c(SB)/4, $0x3f03fff
-DATA ·poly1305_init_constants_armv6<>+0x10(SB)/4, $0x00fffff
-GLOBL ·poly1305_init_constants_armv6<>(SB), 8, $20
-
-// Warning: the linker may use R11 to synthesize certain instructions. Please
-// take care and verify that no synthetic instructions use it.
-
-TEXT poly1305_init_ext_armv6<>(SB), NOSPLIT, $0
-	// Needs 16 bytes of stack and 64 bytes of space pointed to by R0.  (It
-	// might look like it's only 60 bytes of space but the final four bytes
-	// will be written by another function.) We need to skip over four
-	// bytes of stack because that's saving the value of 'g'.
-	ADD       $4, R13, R8
-	MOVM.IB   [R4-R7], (R8)
-	MOVM.IA.W (R1), [R2-R5]
-	MOVW      $·poly1305_init_constants_armv6<>(SB), R7
-	MOVW      R2, R8
-	MOVW      R2>>26, R9
-	MOVW      R3>>20, g
-	MOVW      R4>>14, R11
-	MOVW      R5>>8, R12
-	ORR       R3<<6, R9, R9
-	ORR       R4<<12, g, g
-	ORR       R5<<18, R11, R11
-	MOVM.IA   (R7), [R2-R6]
-	AND       R8, R2, R2
-	AND       R9, R3, R3
-	AND       g, R4, R4
-	AND       R11, R5, R5
-	AND       R12, R6, R6
-	MOVM.IA.W [R2-R6], (R0)
-	EOR       R2, R2, R2
-	EOR       R3, R3, R3
-	EOR       R4, R4, R4
-	EOR       R5, R5, R5
-	EOR       R6, R6, R6
-	MOVM.IA.W [R2-R6], (R0)
-	MOVM.IA.W (R1), [R2-R5]
-	MOVM.IA   [R2-R6], (R0)
-	ADD       $20, R13, R0
-	MOVM.DA   (R0), [R4-R7]
-	RET
-
-#define MOVW_UNALIGNED(Rsrc, Rdst, Rtmp, offset) \
-	MOVBU (offset+0)(Rsrc), Rtmp; \
-	MOVBU Rtmp, (offset+0)(Rdst); \
-	MOVBU (offset+1)(Rsrc), Rtmp; \
-	MOVBU Rtmp, (offset+1)(Rdst); \
-	MOVBU (offset+2)(Rsrc), Rtmp; \
-	MOVBU Rtmp, (offset+2)(Rdst); \
-	MOVBU (offset+3)(Rsrc), Rtmp; \
-	MOVBU Rtmp, (offset+3)(Rdst)
-
-TEXT poly1305_blocks_armv6<>(SB), NOSPLIT, $0
-	// Needs 24 bytes of stack for saved registers and then 88 bytes of
-	// scratch space after that. We assume that 24 bytes at (R13) have
-	// already been used: four bytes for the link register saved in the
-	// prelude of poly1305_auth_armv6, four bytes for saving the value of g
-	// in that function and 16 bytes of scratch space used around
-	// poly1305_finish_ext_armv6_skip1.
-	ADD     $24, R13, R12
-	MOVM.IB [R4-R8, R14], (R12)
-	MOVW    R0, 88(R13)
-	MOVW    R1, 92(R13)
-	MOVW    R2, 96(R13)
-	MOVW    R1, R14
-	MOVW    R2, R12
-	MOVW    56(R0), R8
-	WORD    $0xe1180008                // TST R8, R8 not working see issue 5921
-	EOR     R6, R6, R6
-	MOVW.EQ $(1<<24), R6
-	MOVW    R6, 84(R13)
-	ADD     $116, R13, g
-	MOVM.IA (R0), [R0-R9]
-	MOVM.IA [R0-R4], (g)
-	CMP     $16, R12
-	BLO     poly1305_blocks_armv6_done
-
-poly1305_blocks_armv6_mainloop:
-	WORD    $0xe31e0003                            // TST R14, #3 not working see issue 5921
-	BEQ     poly1305_blocks_armv6_mainloop_aligned
-	ADD     $100, R13, g
-	MOVW_UNALIGNED(R14, g, R0, 0)
-	MOVW_UNALIGNED(R14, g, R0, 4)
-	MOVW_UNALIGNED(R14, g, R0, 8)
-	MOVW_UNALIGNED(R14, g, R0, 12)
-	MOVM.IA (g), [R0-R3]
-	ADD     $16, R14
-	B       poly1305_blocks_armv6_mainloop_loaded
-
-poly1305_blocks_armv6_mainloop_aligned:
-	MOVM.IA.W (R14), [R0-R3]
-
-poly1305_blocks_armv6_mainloop_loaded:
-	MOVW    R0>>26, g
-	MOVW    R1>>20, R11
-	MOVW    R2>>14, R12
-	MOVW    R14, 92(R13)
-	MOVW    R3>>8, R4
-	ORR     R1<<6, g, g
-	ORR     R2<<12, R11, R11
-	ORR     R3<<18, R12, R12
-	BIC     $0xfc000000, R0, R0
-	BIC     $0xfc000000, g, g
-	MOVW    84(R13), R3
-	BIC     $0xfc000000, R11, R11
-	BIC     $0xfc000000, R12, R12
-	ADD     R0, R5, R5
-	ADD     g, R6, R6
-	ORR     R3, R4, R4
-	ADD     R11, R7, R7
-	ADD     $116, R13, R14
-	ADD     R12, R8, R8
-	ADD     R4, R9, R9
-	MOVM.IA (R14), [R0-R4]
-	MULLU   R4, R5, (R11, g)
-	MULLU   R3, R5, (R14, R12)
-	MULALU  R3, R6, (R11, g)
-	MULALU  R2, R6, (R14, R12)
-	MULALU  R2, R7, (R11, g)
-	MULALU  R1, R7, (R14, R12)
-	ADD     R4<<2, R4, R4
-	ADD     R3<<2, R3, R3
-	MULALU  R1, R8, (R11, g)
-	MULALU  R0, R8, (R14, R12)
-	MULALU  R0, R9, (R11, g)
-	MULALU  R4, R9, (R14, R12)
-	MOVW    g, 76(R13)
-	MOVW    R11, 80(R13)
-	MOVW    R12, 68(R13)
-	MOVW    R14, 72(R13)
-	MULLU   R2, R5, (R11, g)
-	MULLU   R1, R5, (R14, R12)
-	MULALU  R1, R6, (R11, g)
-	MULALU  R0, R6, (R14, R12)
-	MULALU  R0, R7, (R11, g)
-	MULALU  R4, R7, (R14, R12)
-	ADD     R2<<2, R2, R2
-	ADD     R1<<2, R1, R1
-	MULALU  R4, R8, (R11, g)
-	MULALU  R3, R8, (R14, R12)
-	MULALU  R3, R9, (R11, g)
-	MULALU  R2, R9, (R14, R12)
-	MOVW    g, 60(R13)
-	MOVW    R11, 64(R13)
-	MOVW    R12, 52(R13)
-	MOVW    R14, 56(R13)
-	MULLU   R0, R5, (R11, g)
-	MULALU  R4, R6, (R11, g)
-	MULALU  R3, R7, (R11, g)
-	MULALU  R2, R8, (R11, g)
-	MULALU  R1, R9, (R11, g)
-	ADD     $52, R13, R0
-	MOVM.IA (R0), [R0-R7]
-	MOVW    g>>26, R12
-	MOVW    R4>>26, R14
-	ORR     R11<<6, R12, R12
-	ORR     R5<<6, R14, R14
-	BIC     $0xfc000000, g, g
-	BIC     $0xfc000000, R4, R4
-	ADD.S   R12, R0, R0
-	ADC     $0, R1, R1
-	ADD.S   R14, R6, R6
-	ADC     $0, R7, R7
-	MOVW    R0>>26, R12
-	MOVW    R6>>26, R14
-	ORR     R1<<6, R12, R12
-	ORR     R7<<6, R14, R14
-	BIC     $0xfc000000, R0, R0
-	BIC     $0xfc000000, R6, R6
-	ADD     R14<<2, R14, R14
-	ADD.S   R12, R2, R2
-	ADC     $0, R3, R3
-	ADD     R14, g, g
-	MOVW    R2>>26, R12
-	MOVW    g>>26, R14
-	ORR     R3<<6, R12, R12
-	BIC     $0xfc000000, g, R5
-	BIC     $0xfc000000, R2, R7
-	ADD     R12, R4, R4
-	ADD     R14, R0, R0
-	MOVW    R4>>26, R12
-	BIC     $0xfc000000, R4, R8
-	ADD     R12, R6, R9
-	MOVW    96(R13), R12
-	MOVW    92(R13), R14
-	MOVW    R0, R6
-	CMP     $32, R12
-	SUB     $16, R12, R12
-	MOVW    R12, 96(R13)
-	BHS     poly1305_blocks_armv6_mainloop
-
-poly1305_blocks_armv6_done:
-	MOVW    88(R13), R12
-	MOVW    R5, 20(R12)
-	MOVW    R6, 24(R12)
-	MOVW    R7, 28(R12)
-	MOVW    R8, 32(R12)
-	MOVW    R9, 36(R12)
-	ADD     $48, R13, R0
-	MOVM.DA (R0), [R4-R8, R14]
-	RET
-
-#define MOVHUP_UNALIGNED(Rsrc, Rdst, Rtmp) \
-	MOVBU.P 1(Rsrc), Rtmp; \
-	MOVBU.P Rtmp, 1(Rdst); \
-	MOVBU.P 1(Rsrc), Rtmp; \
-	MOVBU.P Rtmp, 1(Rdst)
-
-#define MOVWP_UNALIGNED(Rsrc, Rdst, Rtmp) \
-	MOVHUP_UNALIGNED(Rsrc, Rdst, Rtmp); \
-	MOVHUP_UNALIGNED(Rsrc, Rdst, Rtmp)
-
-// func poly1305_auth_armv6(out *[16]byte, m *byte, mlen uint32, key *[32]key)
-TEXT ·poly1305_auth_armv6(SB), $196-16
-	// The value 196, just above, is the sum of 64 (the size of the context
-	// structure) and 132 (the amount of stack needed).
-	//
-	// At this point, the stack pointer (R13) has been moved down. It
-	// points to the saved link register and there's 196 bytes of free
-	// space above it.
-	//
-	// The stack for this function looks like:
-	//
-	// +---------------------
-	// |
-	// | 64 bytes of context structure
-	// |
-	// +---------------------
-	// |
-	// | 112 bytes for poly1305_blocks_armv6
-	// |
-	// +---------------------
-	// | 16 bytes of final block, constructed at
-	// | poly1305_finish_ext_armv6_skip8
-	// +---------------------
-	// | four bytes of saved 'g'
-	// +---------------------
-	// | lr, saved by prelude    <- R13 points here
-	// +---------------------
-	MOVW g, 4(R13)
-
-	MOVW out+0(FP), R4
-	MOVW m+4(FP), R5
-	MOVW mlen+8(FP), R6
-	MOVW key+12(FP), R7
-
-	ADD  $136, R13, R0 // 136 = 4 + 4 + 16 + 112
-	MOVW R7, R1
-
-	// poly1305_init_ext_armv6 will write to the stack from R13+4, but
-	// that's ok because none of the other values have been written yet.
-	BL    poly1305_init_ext_armv6<>(SB)
-	BIC.S $15, R6, R2
-	BEQ   poly1305_auth_armv6_noblocks
-	ADD   $136, R13, R0
-	MOVW  R5, R1
-	ADD   R2, R5, R5
-	SUB   R2, R6, R6
-	BL    poly1305_blocks_armv6<>(SB)
-
-poly1305_auth_armv6_noblocks:
-	ADD  $136, R13, R0
-	MOVW R5, R1
-	MOVW R6, R2
-	MOVW R4, R3
-
-	MOVW  R0, R5
-	MOVW  R1, R6
-	MOVW  R2, R7
-	MOVW  R3, R8
-	AND.S R2, R2, R2
-	BEQ   poly1305_finish_ext_armv6_noremaining
-	EOR   R0, R0
-	ADD   $8, R13, R9                           // 8 = offset to 16 byte scratch space
-	MOVW  R0, (R9)
-	MOVW  R0, 4(R9)
-	MOVW  R0, 8(R9)
-	MOVW  R0, 12(R9)
-	WORD  $0xe3110003                           // TST R1, #3 not working see issue 5921
-	BEQ   poly1305_finish_ext_armv6_aligned
-	WORD  $0xe3120008                           // TST R2, #8 not working see issue 5921
-	BEQ   poly1305_finish_ext_armv6_skip8
-	MOVWP_UNALIGNED(R1, R9, g)
-	MOVWP_UNALIGNED(R1, R9, g)
-
-poly1305_finish_ext_armv6_skip8:
-	WORD $0xe3120004                     // TST $4, R2 not working see issue 5921
-	BEQ  poly1305_finish_ext_armv6_skip4
-	MOVWP_UNALIGNED(R1, R9, g)
-
-poly1305_finish_ext_armv6_skip4:
-	WORD $0xe3120002                     // TST $2, R2 not working see issue 5921
-	BEQ  poly1305_finish_ext_armv6_skip2
-	MOVHUP_UNALIGNED(R1, R9, g)
-	B    poly1305_finish_ext_armv6_skip2
-
-poly1305_finish_ext_armv6_aligned:
-	WORD      $0xe3120008                             // TST R2, #8 not working see issue 5921
-	BEQ       poly1305_finish_ext_armv6_skip8_aligned
-	MOVM.IA.W (R1), [g-R11]
-	MOVM.IA.W [g-R11], (R9)
-
-poly1305_finish_ext_armv6_skip8_aligned:
-	WORD   $0xe3120004                             // TST $4, R2 not working see issue 5921
-	BEQ    poly1305_finish_ext_armv6_skip4_aligned
-	MOVW.P 4(R1), g
-	MOVW.P g, 4(R9)
-
-poly1305_finish_ext_armv6_skip4_aligned:
-	WORD    $0xe3120002                     // TST $2, R2 not working see issue 5921
-	BEQ     poly1305_finish_ext_armv6_skip2
-	MOVHU.P 2(R1), g
-	MOVH.P  g, 2(R9)
-
-poly1305_finish_ext_armv6_skip2:
-	WORD    $0xe3120001                     // TST $1, R2 not working see issue 5921
-	BEQ     poly1305_finish_ext_armv6_skip1
-	MOVBU.P 1(R1), g
-	MOVBU.P g, 1(R9)
-
-poly1305_finish_ext_armv6_skip1:
-	MOVW  $1, R11
-	MOVBU R11, 0(R9)
-	MOVW  R11, 56(R5)
-	MOVW  R5, R0
-	ADD   $8, R13, R1
-	MOVW  $16, R2
-	BL    poly1305_blocks_armv6<>(SB)
-
-poly1305_finish_ext_armv6_noremaining:
-	MOVW      20(R5), R0
-	MOVW      24(R5), R1
-	MOVW      28(R5), R2
-	MOVW      32(R5), R3
-	MOVW      36(R5), R4
-	MOVW      R4>>26, R12
-	BIC       $0xfc000000, R4, R4
-	ADD       R12<<2, R12, R12
-	ADD       R12, R0, R0
-	MOVW      R0>>26, R12
-	BIC       $0xfc000000, R0, R0
-	ADD       R12, R1, R1
-	MOVW      R1>>26, R12
-	BIC       $0xfc000000, R1, R1
-	ADD       R12, R2, R2
-	MOVW      R2>>26, R12
-	BIC       $0xfc000000, R2, R2
-	ADD       R12, R3, R3
-	MOVW      R3>>26, R12
-	BIC       $0xfc000000, R3, R3
-	ADD       R12, R4, R4
-	ADD       $5, R0, R6
-	MOVW      R6>>26, R12
-	BIC       $0xfc000000, R6, R6
-	ADD       R12, R1, R7
-	MOVW      R7>>26, R12
-	BIC       $0xfc000000, R7, R7
-	ADD       R12, R2, g
-	MOVW      g>>26, R12
-	BIC       $0xfc000000, g, g
-	ADD       R12, R3, R11
-	MOVW      $-(1<<26), R12
-	ADD       R11>>26, R12, R12
-	BIC       $0xfc000000, R11, R11
-	ADD       R12, R4, R9
-	MOVW      R9>>31, R12
-	SUB       $1, R12
-	AND       R12, R6, R6
-	AND       R12, R7, R7
-	AND       R12, g, g
-	AND       R12, R11, R11
-	AND       R12, R9, R9
-	MVN       R12, R12
-	AND       R12, R0, R0
-	AND       R12, R1, R1
-	AND       R12, R2, R2
-	AND       R12, R3, R3
-	AND       R12, R4, R4
-	ORR       R6, R0, R0
-	ORR       R7, R1, R1
-	ORR       g, R2, R2
-	ORR       R11, R3, R3
-	ORR       R9, R4, R4
-	ORR       R1<<26, R0, R0
-	MOVW      R1>>6, R1
-	ORR       R2<<20, R1, R1
-	MOVW      R2>>12, R2
-	ORR       R3<<14, R2, R2
-	MOVW      R3>>18, R3
-	ORR       R4<<8, R3, R3
-	MOVW      40(R5), R6
-	MOVW      44(R5), R7
-	MOVW      48(R5), g
-	MOVW      52(R5), R11
-	ADD.S     R6, R0, R0
-	ADC.S     R7, R1, R1
-	ADC.S     g, R2, R2
-	ADC.S     R11, R3, R3
-	MOVM.IA   [R0-R3], (R8)
-	MOVW      R5, R12
-	EOR       R0, R0, R0
-	EOR       R1, R1, R1
-	EOR       R2, R2, R2
-	EOR       R3, R3, R3
-	EOR       R4, R4, R4
-	EOR       R5, R5, R5
-	EOR       R6, R6, R6
-	EOR       R7, R7, R7
-	MOVM.IA.W [R0-R7], (R12)
-	MOVM.IA   [R0-R7], (R12)
-	MOVW      4(R13), g
-	RET

vendor/golang.org/x/crypto/poly1305/sum_generic.go

@@ -2,18 +2,29 @@
 // Use of this source code is governed by a BSD-style
 // license that can be found in the LICENSE file.
 
+// This file provides the generic implementation of Sum and MAC. Other files
+// might provide optimized assembly implementations of some of this code.
+
 package poly1305
 
 import "encoding/binary"
 
-const (
-	msgBlock   = uint32(1 << 24)
-	finalBlock = uint32(0)
-)
+// Poly1305 [RFC 7539] is a relatively simple algorithm: the authentication tag
+// for a 64 bytes message is approximately
+//
+//     s + m[0:16] * r⁴ + m[16:32] * r³ + m[32:48] * r² + m[48:64] * r  mod  2¹³⁰ - 5
+//
+// for some secret r and s. It can be computed sequentially like
+//
+//     for len(msg) > 0:
+//         h += read(msg, 16)
+//         h *= r
+//         h %= 2¹³⁰ - 5
+//     return h + s
+//
+// All the complexity is about doing performant constant-time math on numbers
+// larger than any available numeric type.
 
-// sumGeneric generates an authenticator for msg using a one-time key and
-// puts the 16-byte result into out. This is the generic implementation of
-// Sum and should be called if no assembly implementation is available.
 func sumGeneric(out *[TagSize]byte, msg []byte, key *[32]byte) {
 	h := newMACGeneric(key)
 	h.Write(msg)
@@ -21,152 +32,276 @@ func sumGeneric(out *[TagSize]byte, msg []byte, key *[32]byte) {
 }
 
 func newMACGeneric(key *[32]byte) (h macGeneric) {
-	h.r[0] = binary.LittleEndian.Uint32(key[0:]) & 0x3ffffff
-	h.r[1] = (binary.LittleEndian.Uint32(key[3:]) >> 2) & 0x3ffff03
-	h.r[2] = (binary.LittleEndian.Uint32(key[6:]) >> 4) & 0x3ffc0ff
-	h.r[3] = (binary.LittleEndian.Uint32(key[9:]) >> 6) & 0x3f03fff
-	h.r[4] = (binary.LittleEndian.Uint32(key[12:]) >> 8) & 0x00fffff
-
-	h.s[0] = binary.LittleEndian.Uint32(key[16:])
-	h.s[1] = binary.LittleEndian.Uint32(key[20:])
-	h.s[2] = binary.LittleEndian.Uint32(key[24:])
-	h.s[3] = binary.LittleEndian.Uint32(key[28:])
+	initialize(key, &h.r, &h.s)
 	return
 }
 
+// macState holds numbers in saturated 64-bit little-endian limbs. That is,
+// the value of [x0, x1, x2] is x[0] + x[1] * 2⁶⁴ + x[2] * 2¹²⁸.
+type macState struct {
+	// h is the main accumulator. It is to be interpreted modulo 2¹³⁰ - 5, but
+	// can grow larger during and after rounds.
+	h [3]uint64
+	// r and s are the private key components.
+	r [2]uint64
+	s [2]uint64
+}
+
 type macGeneric struct {
-	h, r [5]uint32
-	s    [4]uint32
+	macState
 
 	buffer [TagSize]byte
 	offset int
 }
 
-func (h *macGeneric) Write(p []byte) (n int, err error) {
-	n = len(p)
+// Write splits the incoming message into TagSize chunks, and passes them to
+// update. It buffers incomplete chunks.
+func (h *macGeneric) Write(p []byte) (int, error) {
+	nn := len(p)
 	if h.offset > 0 {
-		remaining := TagSize - h.offset
-		if n < remaining {
-			h.offset += copy(h.buffer[h.offset:], p)
-			return n, nil
+		n := copy(h.buffer[h.offset:], p)
+		if h.offset+n < TagSize {
+			h.offset += n
+			return nn, nil
 		}
-		copy(h.buffer[h.offset:], p[:remaining])
-		p = p[remaining:]
+		p = p[n:]
 		h.offset = 0
-		updateGeneric(h.buffer[:], msgBlock, &(h.h), &(h.r))
+		updateGeneric(&h.macState, h.buffer[:])
 	}
-	if nn := len(p) - (len(p) % TagSize); nn > 0 {
-		updateGeneric(p, msgBlock, &(h.h), &(h.r))
-		p = p[nn:]
+	if n := len(p) - (len(p) % TagSize); n > 0 {
+		updateGeneric(&h.macState, p[:n])
+		p = p[n:]
 	}
 	if len(p) > 0 {
 		h.offset += copy(h.buffer[h.offset:], p)
 	}
-	return n, nil
+	return nn, nil
 }
 
-func (h *macGeneric) Sum(out *[16]byte) {
-	H, R := h.h, h.r
+// Sum flushes the last incomplete chunk from the buffer, if any, and generates
+// the MAC output. It does not modify its state, in order to allow for multiple
+// calls to Sum, even if no Write is allowed after Sum.
+func (h *macGeneric) Sum(out *[TagSize]byte) {
+	state := h.macState
 	if h.offset > 0 {
-		var buffer [TagSize]byte
-		copy(buffer[:], h.buffer[:h.offset])
-		buffer[h.offset] = 1 // invariant: h.offset < TagSize
-		updateGeneric(buffer[:], finalBlock, &H, &R)
+		updateGeneric(&state, h.buffer[:h.offset])
 	}
-	finalizeGeneric(out, &H, &(h.s))
+	finalize(out, &state.h, &state.s)
+}
+
+// [rMask0, rMask1] is the specified Poly1305 clamping mask in little-endian. It
+// clears some bits of the secret coefficient to make it possible to implement
+// multiplication more efficiently.
+const (
+	rMask0 = 0x0FFFFFFC0FFFFFFF
+	rMask1 = 0x0FFFFFFC0FFFFFFC
+)
+
+func initialize(key *[32]byte, r, s *[2]uint64) {
+	r[0] = binary.LittleEndian.Uint64(key[0:8]) & rMask0
+	r[1] = binary.LittleEndian.Uint64(key[8:16]) & rMask1
+	s[0] = binary.LittleEndian.Uint64(key[16:24])
+	s[1] = binary.LittleEndian.Uint64(key[24:32])
+}
+
+// uint128 holds a 128-bit number as two 64-bit limbs, for use with the
+// bits.Mul64 and bits.Add64 intrinsics.
+type uint128 struct {
+	lo, hi uint64
+}
+
+func mul64(a, b uint64) uint128 {
+	hi, lo := bitsMul64(a, b)
+	return uint128{lo, hi}
 }
 
-func updateGeneric(msg []byte, flag uint32, h, r *[5]uint32) {
-	h0, h1, h2, h3, h4 := h[0], h[1], h[2], h[3], h[4]
-	r0, r1, r2, r3, r4 := uint64(r[0]), uint64(r[1]), uint64(r[2]), uint64(r[3]), uint64(r[4])
-	R1, R2, R3, R4 := r1*5, r2*5, r3*5, r4*5
-
-	for len(msg) >= TagSize {
-		// h += msg
-		h0 += binary.LittleEndian.Uint32(msg[0:]) & 0x3ffffff
-		h1 += (binary.LittleEndian.Uint32(msg[3:]) >> 2) & 0x3ffffff
-		h2 += (binary.LittleEndian.Uint32(msg[6:]) >> 4) & 0x3ffffff
-		h3 += (binary.LittleEndian.Uint32(msg[9:]) >> 6) & 0x3ffffff
-		h4 += (binary.LittleEndian.Uint32(msg[12:]) >> 8) | flag
-
-		// h *= r
-		d0 := (uint64(h0) * r0) + (uint64(h1) * R4) + (uint64(h2) * R3) + (uint64(h3) * R2) + (uint64(h4) * R1)
-		d1 := (d0 >> 26) + (uint64(h0) * r1) + (uint64(h1) * r0) + (uint64(h2) * R4) + (uint64(h3) * R3) + (uint64(h4) * R2)
-		d2 := (d1 >> 26) + (uint64(h0) * r2) + (uint64(h1) * r1) + (uint64(h2) * r0) + (uint64(h3) * R4) + (uint64(h4) * R3)
-		d3 := (d2 >> 26) + (uint64(h0) * r3) + (uint64(h1) * r2) + (uint64(h2) * r1) + (uint64(h3) * r0) + (uint64(h4) * R4)
-		d4 := (d3 >> 26) + (uint64(h0) * r4) + (uint64(h1) * r3) + (uint64(h2) * r2) + (uint64(h3) * r1) + (uint64(h4) * r0)
-
-		// h %= p
-		h0 = uint32(d0) & 0x3ffffff
-		h1 = uint32(d1) & 0x3ffffff
-		h2 = uint32(d2) & 0x3ffffff
-		h3 = uint32(d3) & 0x3ffffff
-		h4 = uint32(d4) & 0x3ffffff
-
-		h0 += uint32(d4>>26) * 5
-		h1 += h0 >> 26
-		h0 = h0 & 0x3ffffff
-
-		msg = msg[TagSize:]
+func add128(a, b uint128) uint128 {
+	lo, c := bitsAdd64(a.lo, b.lo, 0)
+	hi, c := bitsAdd64(a.hi, b.hi, c)
+	if c != 0 {
+		panic("poly1305: unexpected overflow")
 	}
+	return uint128{lo, hi}
+}
 
-	h[0], h[1], h[2], h[3], h[4] = h0, h1, h2, h3, h4
+func shiftRightBy2(a uint128) uint128 {
+	a.lo = a.lo>>2 | (a.hi&3)<<62
+	a.hi = a.hi >> 2
+	return a
 }
 
-func finalizeGeneric(out *[TagSize]byte, h *[5]uint32, s *[4]uint32) {
-	h0, h1, h2, h3, h4 := h[0], h[1], h[2], h[3], h[4]
-
-	// h %= p reduction
-	h2 += h1 >> 26
-	h1 &= 0x3ffffff
-	h3 += h2 >> 26
-	h2 &= 0x3ffffff
-	h4 += h3 >> 26
-	h3 &= 0x3ffffff
-	h0 += 5 * (h4 >> 26)
-	h4 &= 0x3ffffff
-	h1 += h0 >> 26
-	h0 &= 0x3ffffff
-
-	// h - p
-	t0 := h0 + 5
-	t1 := h1 + (t0 >> 26)
-	t2 := h2 + (t1 >> 26)
-	t3 := h3 + (t2 >> 26)
-	t4 := h4 + (t3 >> 26) - (1 << 26)
-	t0 &= 0x3ffffff
-	t1 &= 0x3ffffff
-	t2 &= 0x3ffffff
-	t3 &= 0x3ffffff
-
-	// select h if h < p else h - p
-	t_mask := (t4 >> 31) - 1
-	h_mask := ^t_mask
-	h0 = (h0 & h_mask) | (t0 & t_mask)
-	h1 = (h1 & h_mask) | (t1 & t_mask)
-	h2 = (h2 & h_mask) | (t2 & t_mask)
-	h3 = (h3 & h_mask) | (t3 & t_mask)
-	h4 = (h4 & h_mask) | (t4 & t_mask)
-
-	// h %= 2^128
-	h0 |= h1 << 26
-	h1 = ((h1 >> 6) | (h2 << 20))
-	h2 = ((h2 >> 12) | (h3 << 14))
-	h3 = ((h3 >> 18) | (h4 << 8))
-
-	// s: the s part of the key
-	// tag = (h + s) % (2^128)
-	t := uint64(h0) + uint64(s[0])
-	h0 = uint32(t)
-	t = uint64(h1) + uint64(s[1]) + (t >> 32)
-	h1 = uint32(t)
-	t = uint64(h2) + uint64(s[2]) + (t >> 32)
-	h2 = uint32(t)
-	t = uint64(h3) + uint64(s[3]) + (t >> 32)
-	h3 = uint32(t)
-
-	binary.LittleEndian.PutUint32(out[0:], h0)
-	binary.LittleEndian.PutUint32(out[4:], h1)
-	binary.LittleEndian.PutUint32(out[8:], h2)
-	binary.LittleEndian.PutUint32(out[12:], h3)
+// updateGeneric absorbs msg into the state.h accumulator. For each chunk m of
+// 128 bits of message, it computes
+//
+//     h₊ = (h + m) * r  mod  2¹³⁰ - 5
+//
+// If the msg length is not a multiple of TagSize, it assumes the last
+// incomplete chunk is the final one.
+func updateGeneric(state *macState, msg []byte) {
+	h0, h1, h2 := state.h[0], state.h[1], state.h[2]
+	r0, r1 := state.r[0], state.r[1]
+
+	for len(msg) > 0 {
+		var c uint64
+
+		// For the first step, h + m, we use a chain of bits.Add64 intrinsics.
+		// The resulting value of h might exceed 2¹³⁰ - 5, but will be partially
+		// reduced at the end of the multiplication below.
+		//
+		// The spec requires us to set a bit just above the message size, not to
+		// hide leading zeroes. For full chunks, that's 1 << 128, so we can just
+		// add 1 to the most significant (2¹²⁸) limb, h2.
+		if len(msg) >= TagSize {
+			h0, c = bitsAdd64(h0, binary.LittleEndian.Uint64(msg[0:8]), 0)
+			h1, c = bitsAdd64(h1, binary.LittleEndian.Uint64(msg[8:16]), c)
+			h2 += c + 1
+
+			msg = msg[TagSize:]
+		} else {
+			var buf [TagSize]byte
+			copy(buf[:], msg)
+			buf[len(msg)] = 1
+
+			h0, c = bitsAdd64(h0, binary.LittleEndian.Uint64(buf[0:8]), 0)
+			h1, c = bitsAdd64(h1, binary.LittleEndian.Uint64(buf[8:16]), c)
+			h2 += c
+
+			msg = nil
+		}
+
+		// Multiplication of big number limbs is similar to elementary school
+		// columnar multiplication. Instead of digits, there are 64-bit limbs.
+		//
+		// We are multiplying a 3 limbs number, h, by a 2 limbs number, r.
+		//
+		//                        h2    h1    h0  x
+		//                              r1    r0  =
+		//                       ----------------
+		//                      h2r0  h1r0  h0r0     <-- individual 128-bit products
+		//            +   h2r1  h1r1  h0r1
+		//               ------------------------
+		//                 m3    m2    m1    m0      <-- result in 128-bit overlapping limbs
+		//               ------------------------
+		//         m3.hi m2.hi m1.hi m0.hi           <-- carry propagation
+		//     +         m3.lo m2.lo m1.lo m0.lo
+		//        -------------------------------
+		//           t4    t3    t2    t1    t0      <-- final result in 64-bit limbs
+		//
+		// The main difference from pen-and-paper multiplication is that we do
+		// carry propagation in a separate step, as if we wrote two digit sums
+		// at first (the 128-bit limbs), and then carried the tens all at once.
+
+		h0r0 := mul64(h0, r0)
+		h1r0 := mul64(h1, r0)
+		h2r0 := mul64(h2, r0)
+		h0r1 := mul64(h0, r1)
+		h1r1 := mul64(h1, r1)
+		h2r1 := mul64(h2, r1)
+
+		// Since h2 is known to be at most 7 (5 + 1 + 1), and r0 and r1 have their
+		// top 4 bits cleared by rMask{0,1}, we know that their product is not going
+		// to overflow 64 bits, so we can ignore the high part of the products.
+		//
+		// This also means that the product doesn't have a fifth limb (t4).
+		if h2r0.hi != 0 {
+			panic("poly1305: unexpected overflow")
+		}
+		if h2r1.hi != 0 {
+			panic("poly1305: unexpected overflow")
+		}
+
+		m0 := h0r0
+		m1 := add128(h1r0, h0r1) // These two additions don't overflow thanks again
+		m2 := add128(h2r0, h1r1) // to the 4 masked bits at the top of r0 and r1.
+		m3 := h2r1
+
+		t0 := m0.lo
+		t1, c := bitsAdd64(m1.lo, m0.hi, 0)
+		t2, c := bitsAdd64(m2.lo, m1.hi, c)
+		t3, _ := bitsAdd64(m3.lo, m2.hi, c)
+
+		// Now we have the result as 4 64-bit limbs, and we need to reduce it
+		// modulo 2¹³⁰ - 5. The special shape of this Crandall prime lets us do
+		// a cheap partial reduction according to the reduction identity
+		//
+		//     c * 2¹³⁰ + n  =  c * 5 + n  mod  2¹³⁰ - 5
+		//
+		// because 2¹³⁰ = 5 mod 2¹³⁰ - 5. Partial reduction since the result is
+		// likely to be larger than 2¹³⁰ - 5, but still small enough to fit the
+		// assumptions we make about h in the rest of the code.
+		//
+		// See also https://speakerdeck.com/gtank/engineering-prime-numbers?slide=23
+
+		// We split the final result at the 2¹³⁰ mark into h and cc, the carry.
+		// Note that the carry bits are effectively shifted left by 2, in other
+		// words, cc = c * 4 for the c in the reduction identity.
+		h0, h1, h2 = t0, t1, t2&maskLow2Bits
+		cc := uint128{t2 & maskNotLow2Bits, t3}
+
+		// To add c * 5 to h, we first add cc = c * 4, and then add (cc >> 2) = c.
+
+		h0, c = bitsAdd64(h0, cc.lo, 0)
+		h1, c = bitsAdd64(h1, cc.hi, c)
+		h2 += c
+
+		cc = shiftRightBy2(cc)
+
+		h0, c = bitsAdd64(h0, cc.lo, 0)
+		h1, c = bitsAdd64(h1, cc.hi, c)
+		h2 += c
+
+		// h2 is at most 3 + 1 + 1 = 5, making the whole of h at most
+		//
+		//     5 * 2¹²⁸ + (2¹²⁸ - 1) = 6 * 2¹²⁸ - 1
+	}
+
+	state.h[0], state.h[1], state.h[2] = h0, h1, h2
+}
+
+const (
+	maskLow2Bits    uint64 = 0x0000000000000003
+	maskNotLow2Bits uint64 = ^maskLow2Bits
+)
+
+// select64 returns x if v == 1 and y if v == 0, in constant time.
+func select64(v, x, y uint64) uint64 { return ^(v-1)&x | (v-1)&y }
+
+// [p0, p1, p2] is 2¹³⁰ - 5 in little endian order.
+const (
+	p0 = 0xFFFFFFFFFFFFFFFB
+	p1 = 0xFFFFFFFFFFFFFFFF
+	p2 = 0x0000000000000003
+)
+
+// finalize completes the modular reduction of h and computes
+//
+//     out = h + s  mod  2¹²⁸
+//
+func finalize(out *[TagSize]byte, h *[3]uint64, s *[2]uint64) {
+	h0, h1, h2 := h[0], h[1], h[2]
+
+	// After the partial reduction in updateGeneric, h might be more than
+	// 2¹³⁰ - 5, but will be less than 2 * (2¹³⁰ - 5). To complete the reduction
+	// in constant time, we compute t = h - (2¹³⁰ - 5), and select h as the
+	// result if the subtraction underflows, and t otherwise.
+
+	hMinusP0, b := bitsSub64(h0, p0, 0)
+	hMinusP1, b := bitsSub64(h1, p1, b)
+	_, b = bitsSub64(h2, p2, b)
+
+	// h = h if h < p else h - p
+	h0 = select64(b, h0, hMinusP0)
+	h1 = select64(b, h1, hMinusP1)
+
+	// Finally, we compute the last Poly1305 step
+	//
+	//     tag = h + s  mod  2¹²⁸
+	//
+	// by just doing a wide addition with the 128 low bits of h and discarding
+	// the overflow.
+	h0, c := bitsAdd64(h0, s[0], 0)
+	h1, _ = bitsAdd64(h1, s[1], c)
+
+	binary.LittleEndian.PutUint64(out[0:8], h0)
+	binary.LittleEndian.PutUint64(out[8:16], h1)
 }

vendor/golang.org/x/crypto/poly1305/sum_noasm.go

@@ -2,14 +2,11 @@
 // Use of this source code is governed by a BSD-style
 // license that can be found in the LICENSE file.
 
-// +build s390x,!go1.11 !arm,!amd64,!s390x gccgo appengine nacl
+// +build s390x,!go1.11 !amd64,!s390x,!ppc64le gccgo appengine nacl
 
 package poly1305
 
-// Sum generates an authenticator for msg using a one-time key and puts the
-// 16-byte result into out. Authenticating two different messages with the same
-// key allows an attacker to forge messages at will.
-func Sum(out *[TagSize]byte, msg []byte, key *[32]byte) {
+func sum(out *[TagSize]byte, msg []byte, key *[32]byte) {
 	h := newMAC(key)
 	h.Write(msg)
 	h.Sum(out)

vendor/golang.org/x/crypto/poly1305/sum_ppc64le.go

@@ -0,0 +1,58 @@
+// Copyright 2019 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+// +build ppc64le,!gccgo,!appengine
+
+package poly1305
+
+//go:noescape
+func update(state *macState, msg []byte)
+
+func sum(out *[16]byte, m []byte, key *[32]byte) {
+	h := newMAC(key)
+	h.Write(m)
+	h.Sum(out)
+}
+
+func newMAC(key *[32]byte) (h mac) {
+	initialize(key, &h.r, &h.s)
+	return
+}
+
+// mac is a wrapper for macGeneric that redirects calls that would have gone to
+// updateGeneric to update.
+//
+// Its Write and Sum methods are otherwise identical to the macGeneric ones, but
+// using function pointers would carry a major performance cost.
+type mac struct{ macGeneric }
+
+func (h *mac) Write(p []byte) (int, error) {
+	nn := len(p)
+	if h.offset > 0 {
+		n := copy(h.buffer[h.offset:], p)
+		if h.offset+n < TagSize {
+			h.offset += n
+			return nn, nil
+		}
+		p = p[n:]
+		h.offset = 0
+		update(&h.macState, h.buffer[:])
+	}
+	if n := len(p) - (len(p) % TagSize); n > 0 {
+		update(&h.macState, p[:n])
+		p = p[n:]
+	}
+	if len(p) > 0 {
+		h.offset += copy(h.buffer[h.offset:], p)
+	}
+	return nn, nil
+}
+
+func (h *mac) Sum(out *[16]byte) {
+	state := h.macState
+	if h.offset > 0 {
+		update(&state, h.buffer[:h.offset])
+	}
+	finalize(out, &state.h, &state.s)
+}

vendor/golang.org/x/crypto/poly1305/sum_ppc64le.s

@@ -0,0 +1,181 @@
+// Copyright 2019 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+// +build ppc64le,!gccgo,!appengine
+
+#include "textflag.h"
+
+// This was ported from the amd64 implementation.
+
+#define POLY1305_ADD(msg, h0, h1, h2, t0, t1, t2) \
+	MOVD (msg), t0;  \
+	MOVD 8(msg), t1; \
+	MOVD $1, t2;     \
+	ADDC t0, h0, h0; \
+	ADDE t1, h1, h1; \
+	ADDE t2, h2;     \
+	ADD  $16, msg
+
+#define POLY1305_MUL(h0, h1, h2, r0, r1, t0, t1, t2, t3, t4, t5) \
+	MULLD  r0, h0, t0;  \
+	MULLD  r0, h1, t4;  \
+	MULHDU r0, h0, t1;  \
+	MULHDU r0, h1, t5;  \
+	ADDC   t4, t1, t1;  \
+	MULLD  r0, h2, t2;  \
+	ADDZE  t5;          \
+	MULHDU r1, h0, t4;  \
+	MULLD  r1, h0, h0;  \
+	ADD    t5, t2, t2;  \
+	ADDC   h0, t1, t1;  \
+	MULLD  h2, r1, t3;  \
+	ADDZE  t4, h0;      \
+	MULHDU r1, h1, t5;  \
+	MULLD  r1, h1, t4;  \
+	ADDC   t4, t2, t2;  \
+	ADDE   t5, t3, t3;  \
+	ADDC   h0, t2, t2;  \
+	MOVD   $-4, t4;     \
+	MOVD   t0, h0;      \
+	MOVD   t1, h1;      \
+	ADDZE  t3;          \
+	ANDCC  $3, t2, h2;  \
+	AND    t2, t4, t0;  \
+	ADDC   t0, h0, h0;  \
+	ADDE   t3, h1, h1;  \
+	SLD    $62, t3, t4; \
+	SRD    $2, t2;      \
+	ADDZE  h2;          \
+	OR     t4, t2, t2;  \
+	SRD    $2, t3;      \
+	ADDC   t2, h0, h0;  \
+	ADDE   t3, h1, h1;  \
+	ADDZE  h2
+
+DATA ·poly1305Mask<>+0x00(SB)/8, $0x0FFFFFFC0FFFFFFF
+DATA ·poly1305Mask<>+0x08(SB)/8, $0x0FFFFFFC0FFFFFFC
+GLOBL ·poly1305Mask<>(SB), RODATA, $16
+
+// func update(state *[7]uint64, msg []byte)
+TEXT ·update(SB), $0-32
+	MOVD state+0(FP), R3
+	MOVD msg_base+8(FP), R4
+	MOVD msg_len+16(FP), R5
+
+	MOVD 0(R3), R8   // h0
+	MOVD 8(R3), R9   // h1
+	MOVD 16(R3), R10 // h2
+	MOVD 24(R3), R11 // r0
+	MOVD 32(R3), R12 // r1
+
+	CMP R5, $16
+	BLT bytes_between_0_and_15
+
+loop:
+	POLY1305_ADD(R4, R8, R9, R10, R20, R21, R22)
+
+multiply:
+	POLY1305_MUL(R8, R9, R10, R11, R12, R16, R17, R18, R14, R20, R21)
+	ADD $-16, R5
+	CMP R5, $16
+	BGE loop
+
+bytes_between_0_and_15:
+	CMP  $0, R5
+	BEQ  done
+	MOVD $0, R16 // h0
+	MOVD $0, R17 // h1
+
+flush_buffer:
+	CMP R5, $8
+	BLE just1
+
+	MOVD $8, R21
+	SUB  R21, R5, R21
+
+	// Greater than 8 -- load the rightmost remaining bytes in msg
+	// and put into R17 (h1)
+	MOVD (R4)(R21), R17
+	MOVD $16, R22
+
+	// Find the offset to those bytes
+	SUB R5, R22, R22
+	SLD $3, R22
+
+	// Shift to get only the bytes in msg
+	SRD R22, R17, R17
+
+	// Put 1 at high end
+	MOVD $1, R23
+	SLD  $3, R21
+	SLD  R21, R23, R23
+	OR   R23, R17, R17
+
+	// Remainder is 8
+	MOVD $8, R5
+
+just1:
+	CMP R5, $8
+	BLT less8
+
+	// Exactly 8
+	MOVD (R4), R16
+
+	CMP $0, R17
+
+	// Check if we've already set R17; if not
+	// set 1 to indicate end of msg.
+	BNE  carry
+	MOVD $1, R17
+	BR   carry
+
+less8:
+	MOVD  $0, R16   // h0
+	MOVD  $0, R22   // shift count
+	CMP   R5, $4
+	BLT   less4
+	MOVWZ (R4), R16
+	ADD   $4, R4
+	ADD   $-4, R5
+	MOVD  $32, R22
+
+less4:
+	CMP   R5, $2
+	BLT   less2
+	MOVHZ (R4), R21
+	SLD   R22, R21, R21
+	OR    R16, R21, R16
+	ADD   $16, R22
+	ADD   $-2, R5
+	ADD   $2, R4
+
+less2:
+	CMP   $0, R5
+	BEQ   insert1
+	MOVBZ (R4), R21
+	SLD   R22, R21, R21
+	OR    R16, R21, R16
+	ADD   $8, R22
+
+insert1:
+	// Insert 1 at end of msg
+	MOVD $1, R21
+	SLD  R22, R21, R21
+	OR   R16, R21, R16
+
+carry:
+	// Add new values to h0, h1, h2
+	ADDC R16, R8
+	ADDE R17, R9
+	ADDE $0, R10
+	MOVD $16, R5
+	ADD  R5, R4
+	BR   multiply
+
+done:
+	// Save h0, h1, h2 in state
+	MOVD R8, 0(R3)
+	MOVD R9, 8(R3)
+	MOVD R10, 16(R3)
+	RET

vendor/golang.org/x/crypto/poly1305/sum_s390x.go

@@ -22,10 +22,7 @@ func poly1305vx(out *[16]byte, m *byte, mlen uint64, key *[32]byte)
 //go:noescape
 func poly1305vmsl(out *[16]byte, m *byte, mlen uint64, key *[32]byte)
 
-// Sum generates an authenticator for m using a one-time key and puts the
-// 16-byte result into out. Authenticating two different messages with the same
-// key allows an attacker to forge messages at will.
-func Sum(out *[16]byte, m []byte, key *[32]byte) {
+func sum(out *[16]byte, m []byte, key *[32]byte) {
 	if cpu.S390X.HasVX {
 		var mPtr *byte
 		if len(m) > 0 {

vendor/vendor.json

@@ -640,10 +640,10 @@
 			"revisionTime": "2019-02-05T21:23:42Z"
 		},
 		{
-			"checksumSHA1": "vEQUUlb4vRR6zD4mDwZMKV/QP+M=",
+			"checksumSHA1": "gM7I3f4twvtCL2rSLu/QNIQnZf8=",
 			"path": "golang.org/x/crypto/poly1305",
-			"revision": "a1f597ede03a7bef967a422b5b3a5bd08805a01e",
-			"revisionTime": "2019-02-05T21:23:42Z"
+			"revision": "1d94cc7ab1c630336ab82ccb9c9cda72a875c382",
+			"revisionTime": "2020-02-12T18:29:22Z"
 		},
 		{
 			"checksumSHA1": "yQNckVsxPX36aYbCk1ftg2w9ufw=",