Don't assume alignment of ChaCha key on ARM.

When addressing [1], I checked the AEAD code but brain-farted: a key is
aligned in that code, but it's the Poly1305 key, which doesn't matter
here.

It would be nice to align the ChaCha key too, but Android doesn't have
|posix_memalign| in the versions that we care about. It does have
|memalign|, but that's documented as "obsolete" and we don't have a
concept of an Android OS yet and I don't want to add one just for this.

So this change uses the buffer for loading the key again.

(Note that we never used to check for alignment of the |key| before
calling this. We must have gotten it for free somehow when checking the
alignment of |in| and |out|. But there are clearly some paths that don't
have an aligned key:
https://code.google.com/p/chromium/issues/detail?id=454308.)

At least the generation script started paying off immediately ☺.

[1] https://boringssl-review.googlesource.com/#/c/3132/1/crypto/chacha/chacha_vec.c@185

Change-Id: I4f893ba0733440fddd453f9636cc2aeaf05076ed
Reviewed-on: https://boringssl-review.googlesource.com/3270
Reviewed-by: Adam Langley <agl@google.com>

crypto/chacha/chacha_vec.c

@@ -40,6 +40,7 @@ typedef unsigned vec __attribute__((vector_size(16)));
  * This implementation supports parallel processing of multiple blocks,
  * including potentially using general-purpose registers. */
 #if __ARM_NEON__
+#include <string.h>
 #include <arm_neon.h>
 #define GPR_TOO 1
 #define VBPI 2
@@ -162,29 +163,15 @@ void CRYPTO_chacha_20(
 	uint8_t alignment_buffer[16] __attribute__((aligned(16)));
 #endif
 	vec s0, s1, s2, s3;
-#if !defined(__ARM_NEON__) && !defined(__SSE2__)
-	__attribute__ ((aligned (16))) unsigned key[8], nonce[4];
-#endif
 	__attribute__ ((aligned (16))) unsigned chacha_const[] =
 		{0x61707865,0x3320646E,0x79622D32,0x6B206574};
-#if defined(__ARM_NEON__) || defined(__SSE2__)
-	kp = (unsigned *)key;
-#else
-	((vec *)key)[0] = REVV_BE(((vec *)key)[0]);
-	((vec *)key)[1] = REVV_BE(((vec *)key)[1]);
-	nonce[0] = REVW_BE(((unsigned *)nonce)[0]);
-	nonce[1] = REVW_BE(((unsigned *)nonce)[1]);
-	nonce[2] = REVW_BE(((unsigned *)nonce)[2]);
-	nonce[3] = REVW_BE(((unsigned *)nonce)[3]);
 	kp = (unsigned *)key;
-	np = (unsigned *)nonce;
-#endif
 #if defined(__ARM_NEON__)
 	np = (unsigned*) nonce;
 #endif
 	s0 = LOAD_ALIGNED(chacha_const);
-	s1 = LOAD_ALIGNED(&((vec*)kp)[0]);
-	s2 = LOAD_ALIGNED(&((vec*)kp)[1]);
+	s1 = LOAD(&((vec*)kp)[0]);
+	s2 = LOAD(&((vec*)kp)[1]);
 	s3 = (vec){
 		counter & 0xffffffff,
 #if __ARM_NEON__ || defined(OPENSSL_X86)

crypto/chacha/chacha_vec_arm.S

@@ -62,74 +62,88 @@ CRYPTO_chacha_20_neon:
 	@ args = 8, pretend = 0, frame = 128
 	@ frame_needed = 1, uses_anonymous_args = 0
 	push	{r4, r5, r6, r7, r8, r9, r10, fp, lr}
-	mov	r4, r2
+	mov	ip, r3
 	vpush.64	{d8, d9, d10, d11, d12, d13, d14, d15}
+	mov	r9, r2
+	ldr	r4, .L92+16
+	mov	fp, r0
+	mov	r10, r1
+	mov	lr, ip
+.LPIC16:
+	add	r4, pc
 	movw	r8, #43691
-	movt	r8, 43690
-	mov	ip, r3
-	umull	r8, r9, r4, r8
 	sub	sp, sp, #132
 	add	r7, sp, #0
 	sub	sp, sp, #112
-	mov	fp, r0
-	mov	r10, r1
+	movt	r8, 43690
+	str	r0, [r7, #60]
+	str	r1, [r7, #12]
 	str	r2, [r7, #8]
+	ldmia	r4, {r0, r1, r2, r3}
 	add	r4, sp, #15
-	ldr	r2, .L92+16
 	bic	r4, r4, #15
-	ldr	r5, [r7, #232]
-	add	lr, r4, #64
-.LPIC16:
-	add	r2, pc
-	str	r0, [r7, #60]
-	str	r1, [r7, #12]
-	str	r3, [r7, #44]
-	ldmia	r2, {r0, r1, r2, r3}
-	ldr	r6, [r5]
-	str	r4, [r7, #72]
-	ldr	r5, [r5, #4]
-	ldr	r4, [r7, #236]
-	str	r6, [r7, #120]
-	str	r5, [r7, #124]
-	str	r4, [r7, #112]
-	stmia	lr, {r0, r1, r2, r3}
-	movs	r3, #0
-	ldr	r0, [r7, #72]
-	str	r3, [r7, #116]
-	lsrs	r3, r9, #7
+	str	ip, [r7, #44]
+	umull	r8, r9, r9, r8
+	mov	r6, r4
+	adds	r4, r4, #64
+	add	r5, r6, #80
+	str	r5, [r7, #68]
+	stmia	r4, {r0, r1, r2, r3}
+	movs	r4, #0
+	ldr	r0, [ip]	@ unaligned
+	ldr	r1, [ip, #4]	@ unaligned
+	ldr	r2, [ip, #8]	@ unaligned
+	ldr	r3, [ip, #12]	@ unaligned
+	vldr	d24, [r6, #64]
+	vldr	d25, [r6, #72]
+	str	r4, [r7, #116]
+	mov	r4, r5
+	stmia	r5!, {r0, r1, r2, r3}
+	ldr	r0, [lr, #16]!	@ unaligned
+	ldr	r3, [r7, #232]
+	str	r6, [r7, #72]
+	ldr	r2, [lr, #8]	@ unaligned
+	ldr	ip, [r3]
+	ldr	r6, [r7, #236]
+	ldr	r1, [lr, #4]	@ unaligned
+	ldr	r3, [lr, #12]	@ unaligned
+	ldr	r5, [r7, #72]
+	vldr	d26, [r5, #80]
+	vldr	d27, [r5, #88]
+	str	ip, [r7, #120]
+	stmia	r4!, {r0, r1, r2, r3}
+	lsrs	r2, r9, #7
+	ldr	r3, [r7, #232]
+	str	r6, [r7, #112]
+	vldr	d28, [r5, #80]
+	vldr	d29, [r5, #88]
+	ldr	r3, [r3, #4]
+	str	r3, [r7, #124]
 	vldr	d22, [r7, #112]
 	vldr	d23, [r7, #120]
-	vldr	d24, [r0, #64]
-	vldr	d25, [r0, #72]
-	vld1.64	{d26-d27}, [ip:64]
-	vldr	d28, [ip, #16]
-	vldr	d29, [ip, #24]
 	beq	.L26
-	ldr	r1, [r0, #64]
-	lsls	r2, r3, #8
-	sub	r3, r2, r3, lsl #6
-	str	r3, [r7, #4]
-	ldr	r2, [r0, #72]
-	str	r1, [r7, #40]
-	mov	r1, r3
-	ldr	r3, [r0, #68]
+	lsls	r1, r2, #8
+	ldr	r4, [r5, #64]
+	sub	r2, r1, r2, lsl #6
+	str	r2, [r7, #4]
 	vldr	d0, .L92
 	vldr	d1, .L92+8
-	str	r2, [r7, #32]
-	adds	r2, r4, #2
-	str	r3, [r7, #36]
-	ldr	r3, [r0, #76]
-	str	r2, [r7, #48]
-	mov	r2, r0
 	mov	r0, fp
+	mov	r1, r2
+	ldr	r2, [r5, #68]
+	str	r4, [r7, #40]
+	ldr	r4, [r5, #72]
+	str	r2, [r7, #36]
+	ldr	r2, [r5, #76]
+	str	r4, [r7, #32]
+	adds	r4, r6, #2
 	str	r10, [r7, #64]
-	str	r3, [r7, #28]
-	adds	r3, r0, r1
-	mov	r1, r6
-	str	r3, [r7, #16]
-	add	r3, r2, #80
-	mov	r2, r5
-	str	r3, [r7, #68]
+	str	r2, [r7, #28]
+	adds	r2, r0, r1
+	mov	r1, ip
+	str	r2, [r7, #16]
+	mov	r2, r3
+	str	r4, [r7, #48]
 .L4:
 	ldr	r0, [r7, #44]
 	add	r8, r7, #28
@@ -749,14 +763,12 @@ CRYPTO_chacha_20_neon:
 	rsb	fp, fp, r1
 	lsrs	fp, fp, #6
 	beq	.L6
-	ldr	r6, [r7, #72]
 	ldr	r5, [r7, #12]
 	ldr	r4, [r7, #16]
-	mov	r3, r6
-	adds	r3, r3, #80
+	ldr	r6, [r7, #72]
+	ldr	lr, [r7, #68]
 	vldr	d30, .L95
 	vldr	d31, .L95+8
-	mov	lr, r3
 	str	fp, [r7, #104]
 	str	fp, [r7, #108]
 .L8:
@@ -1299,18 +1311,15 @@ CRYPTO_chacha_20_neon:
 	vldm	sp!, {d8-d15}
 	pop	{r4, r5, r6, r7, r8, r9, r10, fp, pc}
 .L89:
-	ldr	r4, [r7, #12]
+	ldr	r5, [r7, #12]
 	vadd.i32	q12, q12, q10
-	ldr	r5, [r7, #72]
+	ldr	r4, [r7, #68]
 	cmp	r9, #31
-	ldr	r0, [r4]	@ unaligned
-	add	r6, r5, #80
-	ldr	r1, [r4, #4]	@ unaligned
-	ldr	r2, [r4, #8]	@ unaligned
-	mov	r5, r6
-	ldr	r3, [r4, #12]	@ unaligned
-	mov	r4, r6
-	str	r6, [r7, #68]
+	ldr	r0, [r5]	@ unaligned
+	ldr	r1, [r5, #4]	@ unaligned
+	mov	r6, r4
+	ldr	r2, [r5, #8]	@ unaligned
+	ldr	r3, [r5, #12]	@ unaligned
 	stmia	r6!, {r0, r1, r2, r3}
 	ldr	r2, [r7, #72]
 	ldr	r6, [r7, #16]
@@ -1355,13 +1364,13 @@ CRYPTO_chacha_20_neon:
 	str	fp, [r7, #16]
 	b	.L2
 .L90:
-	ldr	r3, [r7, #12]
+	mov	r3, r5
+	ldr	r4, [r7, #68]
+	ldr	r0, [r3, #16]!	@ unaligned
 	add	lr, r1, #16
-	mov	r4, r5
-	mov	r6, r5
 	mov	r5, r1
 	vadd.i32	q13, q13, q15
-	ldr	r0, [r3, #16]!	@ unaligned
+	mov	r6, r4
 	cmp	r9, #47
 	ldr	r1, [r3, #4]	@ unaligned
 	ldr	r2, [r3, #8]	@ unaligned