crypto: arm64/aes-ce - implement 5 way interleave for ECB, CBC and CTR

This implements 5-way interleaving for ECB, CBC decryption and CTR,
resulting in a speedup of ~11% on Marvell ThunderX2, which has a
very deep pipeline and therefore a high issue latency for NEON
instructions operating on the same registers.

Note that XTS is left alone: implementing 5-way interleave there
would either involve spilling of the calculated tweaks to the
stack, or recalculating them after the encryption operation, and
doing either of those would most likely penalize low end cores.

For ECB, this is not a concern at all, given that we have plenty
of spare registers. For CTR and CBC decryption, we take advantage
of the fact that v16 is not used by the CE version of the code
(which is the only one targeted by the optimization), and so we
can reshuffle the code a bit and avoid having to spill to memory
(with the exception of one extra reload in the CBC routine)

Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

3 files changed, 75 insertions, 31 deletions

diff --git a/arch/arm64/crypto/aes-ce.S b/arch/arm64/crypto/aes-ce.S
index 0fca5f4634067d..dbfc04ea20c75c 100644
--- a/arch/arm64/crypto/aes-ce.S
+++ b/arch/arm64/crypto/aes-ce.S
@@ -18,6 +18,8 @@
 	.arch		armv8-a+crypto
 
 	xtsmask		.req	v16
+	cbciv		.req	v16
+	vctr		.req	v16
 
 	.macro		xts_reload_mask, tmp
 	.endm
diff --git a/arch/arm64/crypto/aes-modes.S b/arch/arm64/crypto/aes-modes.S
index add6267f9e3a22..3f60f7b152bf33 100644
--- a/arch/arm64/crypto/aes-modes.S
+++ b/arch/arm64/crypto/aes-modes.S
@@ -17,6 +17,14 @@
 #define MAX_STRIDE	4
 #endif
 
+#if MAX_STRIDE == 4
+#define ST4(x...) x
+#define ST5(x...)
+#else
+#define ST4(x...)
+#define ST5(x...) x
+#endif
+
 aes_encrypt_block4x:
 	encrypt_block4x	v0, v1, v2, v3, w3, x2, x8, w7
 	ret
@@ -53,14 +61,17 @@ AES_ENTRY(aes_ecb_encrypt)
 	enc_prepare	w3, x2, x5
 
 .LecbencloopNx:
-	subs		w4, w4, #4
+	subs		w4, w4, #MAX_STRIDE
 	bmi		.Lecbenc1x
 	ld1		{v0.16b-v3.16b}, [x1], #64	/* get 4 pt blocks */
-	bl		aes_encrypt_block4x
+ST4(	bl		aes_encrypt_block4x		)
+ST5(	ld1		{v4.16b}, [x1], #16		)
+ST5(	bl		aes_encrypt_block5x		)
 	st1		{v0.16b-v3.16b}, [x0], #64
+ST5(	st1		{v4.16b}, [x0], #16		)
 	b		.LecbencloopNx
 .Lecbenc1x:
-	adds		w4, w4, #4
+	adds		w4, w4, #MAX_STRIDE
 	beq		.Lecbencout
 .Lecbencloop:
 	ld1		{v0.16b}, [x1], #16		/* get next pt block */
@@ -81,14 +92,17 @@ AES_ENTRY(aes_ecb_decrypt)
 	dec_prepare	w3, x2, x5
 
 .LecbdecloopNx:
-	subs		w4, w4, #4
+	subs		w4, w4, #MAX_STRIDE
 	bmi		.Lecbdec1x
 	ld1		{v0.16b-v3.16b}, [x1], #64	/* get 4 ct blocks */
-	bl		aes_decrypt_block4x
+ST4(	bl		aes_decrypt_block4x		)
+ST5(	ld1		{v4.16b}, [x1], #16		)
+ST5(	bl		aes_decrypt_block5x		)
 	st1		{v0.16b-v3.16b}, [x0], #64
+ST5(	st1		{v4.16b}, [x0], #16		)
 	b		.LecbdecloopNx
 .Lecbdec1x:
-	adds		w4, w4, #4
+	adds		w4, w4, #MAX_STRIDE
 	beq		.Lecbdecout
 .Lecbdecloop:
 	ld1		{v0.16b}, [x1], #16		/* get next ct block */
@@ -148,39 +162,56 @@ AES_ENTRY(aes_cbc_decrypt)
 	stp		x29, x30, [sp, #-16]!
 	mov		x29, sp
 
-	ld1		{v7.16b}, [x5]			/* get iv */
+	ld1		{cbciv.16b}, [x5]		/* get iv */
 	dec_prepare	w3, x2, x6
 
 .LcbcdecloopNx:
-	subs		w4, w4, #4
+	subs		w4, w4, #MAX_STRIDE
 	bmi		.Lcbcdec1x
 	ld1		{v0.16b-v3.16b}, [x1], #64	/* get 4 ct blocks */
+#if MAX_STRIDE == 5
+	ld1		{v4.16b}, [x1], #16		/* get 1 ct block */
+	mov		v5.16b, v0.16b
+	mov		v6.16b, v1.16b
+	mov		v7.16b, v2.16b
+	bl		aes_decrypt_block5x
+	sub		x1, x1, #32
+	eor		v0.16b, v0.16b, cbciv.16b
+	eor		v1.16b, v1.16b, v5.16b
+	ld1		{v5.16b}, [x1], #16		/* reload 1 ct block */
+	ld1		{cbciv.16b}, [x1], #16		/* reload 1 ct block */
+	eor		v2.16b, v2.16b, v6.16b
+	eor		v3.16b, v3.16b, v7.16b
+	eor		v4.16b, v4.16b, v5.16b
+#else
 	mov		v4.16b, v0.16b
 	mov		v5.16b, v1.16b
 	mov		v6.16b, v2.16b
 	bl		aes_decrypt_block4x
 	sub		x1, x1, #16
-	eor		v0.16b, v0.16b, v7.16b
+	eor		v0.16b, v0.16b, cbciv.16b
 	eor		v1.16b, v1.16b, v4.16b
-	ld1		{v7.16b}, [x1], #16		/* reload 1 ct block */
+	ld1		{cbciv.16b}, [x1], #16		/* reload 1 ct block */
 	eor		v2.16b, v2.16b, v5.16b
 	eor		v3.16b, v3.16b, v6.16b
+#endif
 	st1		{v0.16b-v3.16b}, [x0], #64
+ST5(	st1		{v4.16b}, [x0], #16		)
 	b		.LcbcdecloopNx
 .Lcbcdec1x:
-	adds		w4, w4, #4
+	adds		w4, w4, #MAX_STRIDE
 	beq		.Lcbcdecout
 .Lcbcdecloop:
 	ld1		{v1.16b}, [x1], #16		/* get next ct block */
 	mov		v0.16b, v1.16b			/* ...and copy to v0 */
 	decrypt_block	v0, w3, x2, x6, w7
-	eor		v0.16b, v0.16b, v7.16b		/* xor with iv => pt */
-	mov		v7.16b, v1.16b			/* ct is next iv */
+	eor		v0.16b, v0.16b, cbciv.16b	/* xor with iv => pt */
+	mov		cbciv.16b, v1.16b		/* ct is next iv */
 	st1		{v0.16b}, [x0], #16
 	subs		w4, w4, #1
 	bne		.Lcbcdecloop
 .Lcbcdecout:
-	st1		{v7.16b}, [x5]			/* return iv */
+	st1		{cbciv.16b}, [x5]		/* return iv */
 	ldp		x29, x30, [sp], #16
 	ret
 AES_ENDPROC(aes_cbc_decrypt)
@@ -274,51 +305,60 @@ AES_ENTRY(aes_ctr_encrypt)
 	mov		x29, sp
 
 	enc_prepare	w3, x2, x6
-	ld1		{v4.16b}, [x5]
+	ld1		{vctr.16b}, [x5]
 
-	umov		x6, v4.d[1]		/* keep swabbed ctr in reg */
+	umov		x6, vctr.d[1]		/* keep swabbed ctr in reg */
 	rev		x6, x6
 	cmn		w6, w4			/* 32 bit overflow? */
 	bcs		.Lctrloop
 .LctrloopNx:
-	subs		w4, w4, #4
+	subs		w4, w4, #MAX_STRIDE
 	bmi		.Lctr1x
 	add		w7, w6, #1
-	mov		v0.16b, v4.16b
+	mov		v0.16b, vctr.16b
 	add		w8, w6, #2
-	mov		v1.16b, v4.16b
+	mov		v1.16b, vctr.16b
+	add		w9, w6, #3
+	mov		v2.16b, vctr.16b
 	add		w9, w6, #3
-	mov		v2.16b, v4.16b
 	rev		w7, w7
-	mov		v3.16b, v4.16b
+	mov		v3.16b, vctr.16b
 	rev		w8, w8
+ST5(	mov		v4.16b, vctr.16b		)
 	mov		v1.s[3], w7
 	rev		w9, w9
+ST5(	add		w10, w6, #4			)
 	mov		v2.s[3], w8
+ST5(	rev		w10, w10			)
 	mov		v3.s[3], w9
+ST5(	mov		v4.s[3], w10			)
 	ld1		{v5.16b-v7.16b}, [x1], #48	/* get 3 input blocks */
-	bl		aes_encrypt_block4x
+ST4(	bl		aes_encrypt_block4x		)
+ST5(	bl		aes_encrypt_block5x		)
 	eor		v0.16b, v5.16b, v0.16b
-	ld1		{v5.16b}, [x1], #16		/* get 1 input block  */
+ST4(	ld1		{v5.16b}, [x1], #16		)
 	eor		v1.16b, v6.16b, v1.16b
+ST5(	ld1		{v5.16b-v6.16b}, [x1], #32	)
 	eor		v2.16b, v7.16b, v2.16b
 	eor		v3.16b, v5.16b, v3.16b
+ST5(	eor		v4.16b, v6.16b, v4.16b		)
 	st1		{v0.16b-v3.16b}, [x0], #64
-	add		x6, x6, #4
+ST5(	st1		{v4.16b}, [x0], #16		)
+	add		x6, x6, #MAX_STRIDE
 	rev		x7, x6
-	ins		v4.d[1], x7
+	ins		vctr.d[1], x7
 	cbz		w4, .Lctrout
 	b		.LctrloopNx
 .Lctr1x:
-	adds		w4, w4, #4
+	adds		w4, w4, #MAX_STRIDE
 	beq		.Lctrout
 .Lctrloop:
-	mov		v0.16b, v4.16b
+	mov		v0.16b, vctr.16b
 	encrypt_block	v0, w3, x2, x8, w7
 
 	adds		x6, x6, #1		/* increment BE ctr */
 	rev		x7, x6
-	ins		v4.d[1], x7
+	ins		vctr.d[1], x7
 	bcs		.Lctrcarry		/* overflow? */
 
 .Lctrcarrydone:
@@ -330,7 +370,7 @@ AES_ENTRY(aes_ctr_encrypt)
 	bne		.Lctrloop
 
 .Lctrout:
-	st1		{v4.16b}, [x5]		/* return next CTR value */
+	st1		{vctr.16b}, [x5]	/* return next CTR value */
 	ldp		x29, x30, [sp], #16
 	ret
 
@@ -339,11 +379,11 @@ AES_ENTRY(aes_ctr_encrypt)
 	b		.Lctrout
 
 .Lctrcarry:
-	umov		x7, v4.d[0]		/* load upper word of ctr  */
+	umov		x7, vctr.d[0]		/* load upper word of ctr  */
 	rev		x7, x7			/* ... to handle the carry */
 	add		x7, x7, #1
 	rev		x7, x7
-	ins		v4.d[0], x7
+	ins		vctr.d[0], x7
 	b		.Lctrcarrydone
 AES_ENDPROC(aes_ctr_encrypt)
 
diff --git a/arch/arm64/crypto/aes-neon.S b/arch/arm64/crypto/aes-neon.S
index 33bb6af309a351..8bd66a6c4749ed 100644
--- a/arch/arm64/crypto/aes-neon.S
+++ b/arch/arm64/crypto/aes-neon.S
@@ -15,6 +15,8 @@
 #define AES_ENDPROC(func)	ENDPROC(neon_ ## func)
 
 	xtsmask		.req	v7
+	cbciv		.req	v7
+	vctr		.req	v4
 
 	.macro		xts_reload_mask, tmp
 	xts_load_mask	\tmp