poly1305: feed fpu functions PAGE_SIZE at a time

4 files changed, 49 insertions, 29 deletions

diff --git a/src/crypto/zinc/poly1305/poly1305-arm-glue.h b/src/crypto/zinc/poly1305/poly1305-arm-glue.h
index 15ad53f..ca2ed5a 100644
--- a/src/crypto/zinc/poly1305/poly1305-arm-glue.h
+++ b/src/crypto/zinc/poly1305/poly1305-arm-glue.h
@@ -55,7 +55,7 @@ static void convert_to_base2_64(void *ctx)
 	struct poly1305_arch_internal *state = ctx;
 	u32 cy;
 
-	if (!state->is_base2_26)
+	if (!IS_ENABLED(CONFIG_KERNEL_MODE_NEON) || !state->is_base2_26)
 		return;
 
 	cy = state->h[0] >> 26; state->h[0] &= 0x3ffffff; state->h[1] += cy;
@@ -87,18 +87,30 @@ static inline bool poly1305_init_arch(void *ctx,
 }
 
 static inline bool poly1305_blocks_arch(void *ctx, const u8 *inp,
-					const size_t len, const u32 padbit,
+					size_t len, const u32 padbit,
 					simd_context_t *simd_context)
 {
-	if (IS_ENABLED(CONFIG_KERNEL_MODE_NEON)) {
-		if (poly1305_use_neon && simd_use(simd_context)) {
-			poly1305_blocks_neon(ctx, inp, len, padbit);
-			return true;
-		}
+	/* SIMD disables preemption, so relax after processing each page. */
+	BUILD_BUG_ON(PAGE_SIZE < POLY1305_BLOCK_SIZE ||
+		     PAGE_SIZE % POLY1305_BLOCK_SIZE);
+
+	if (!IS_ENABLED(CONFIG_KERNEL_MODE_NEON) || !poly1305_use_neon ||
+	    !simd_use(simd_context)) {
 		convert_to_base2_64(ctx);
+		poly1305_blocks_arm(ctx, inp, len, padbit);
+		return true;
 	}
 
-	poly1305_blocks_arm(ctx, inp, len, padbit);
+	for (;;) {
+		const size_t bytes = min_t(size_t, len, PAGE_SIZE);
+
+		poly1305_blocks_neon(ctx, inp, bytes, padbit);
+		len -= bytes;
+		if (!len)
+			break;
+		inp += bytes;
+		simd_relax(simd_context);
+	}
 	return true;
 }
 
@@ -106,14 +118,11 @@ static inline bool poly1305_emit_arch(void *ctx, u8 mac[POLY1305_MAC_SIZE],
 				      const u32 nonce[4],
 				      simd_context_t *simd_context)
 {
-	if (IS_ENABLED(CONFIG_KERNEL_MODE_NEON)) {
-		if (poly1305_use_neon && simd_use(simd_context)) {
-			poly1305_emit_neon(ctx, mac, nonce);
-			return true;
-		}
+	if (!IS_ENABLED(CONFIG_KERNEL_MODE_NEON) || !poly1305_use_neon ||
+	    !simd_use(simd_context)) {
 		convert_to_base2_64(ctx);
-	}
-
-	poly1305_emit_arm(ctx, mac, nonce);
+		poly1305_emit_arm(ctx, mac, nonce);
+	} else
+		poly1305_emit_neon(ctx, mac, nonce);
 	return true;
 }
diff --git a/src/crypto/zinc/poly1305/poly1305-mips-glue.h b/src/crypto/zinc/poly1305/poly1305-mips-glue.h
index 4a0badb..f71c3f6 100644
--- a/src/crypto/zinc/poly1305/poly1305-mips-glue.h
+++ b/src/crypto/zinc/poly1305/poly1305-mips-glue.h
@@ -20,7 +20,7 @@ static inline bool poly1305_init_arch(void *ctx,
 }
 
 static inline bool poly1305_blocks_arch(void *ctx, const u8 *inp,
-					const size_t len, const u32 padbit,
+					size_t len, const u32 padbit,
 					simd_context_t *simd_context)
 {
 	poly1305_blocks_mips(ctx, inp, len, padbit);
diff --git a/src/crypto/zinc/poly1305/poly1305-x86_64-glue.h b/src/crypto/zinc/poly1305/poly1305-x86_64-glue.h
index 285cb31..b1248e8 100644
--- a/src/crypto/zinc/poly1305/poly1305-x86_64-glue.h
+++ b/src/crypto/zinc/poly1305/poly1305-x86_64-glue.h
@@ -93,11 +93,15 @@ static void convert_to_base2_64(void *ctx)
 }
 
 static inline bool poly1305_blocks_arch(void *ctx, const u8 *inp,
-					const size_t len, const u32 padbit,
+					size_t len, const u32 padbit,
 					simd_context_t *simd_context)
 {
 	struct poly1305_arch_internal *state = ctx;
 
+	/* SIMD disables preemption, so relax after processing each page. */
+	BUILD_BUG_ON(PAGE_SIZE < POLY1305_BLOCK_SIZE ||
+		     PAGE_SIZE % POLY1305_BLOCK_SIZE);
+
 	if (!IS_ENABLED(CONFIG_AS_AVX) || !poly1305_use_avx ||
 	    (len < (POLY1305_BLOCK_SIZE * 18) && !state->is_base2_26) ||
 	    !simd_use(simd_context)) {
@@ -106,12 +110,21 @@ static inline bool poly1305_blocks_arch(void *ctx, const u8 *inp,
 		return true;
 	}
 
-	if (IS_ENABLED(CONFIG_AS_AVX512) && poly1305_use_avx512)
-		poly1305_blocks_avx512(ctx, inp, len, padbit);
-	else if (IS_ENABLED(CONFIG_AS_AVX2) && poly1305_use_avx2)
-		poly1305_blocks_avx2(ctx, inp, len, padbit);
-	else
-		poly1305_blocks_avx(ctx, inp, len, padbit);
+	for (;;) {
+		const size_t bytes = min_t(size_t, len, PAGE_SIZE);
+
+		if (IS_ENABLED(CONFIG_AS_AVX512) && poly1305_use_avx512)
+			poly1305_blocks_avx512(ctx, inp, bytes, padbit);
+		else if (IS_ENABLED(CONFIG_AS_AVX2) && poly1305_use_avx2)
+			poly1305_blocks_avx2(ctx, inp, bytes, padbit);
+		else
+			poly1305_blocks_avx(ctx, inp, bytes, padbit);
+		len -= bytes;
+		if (!len)
+			break;
+		inp += bytes;
+		simd_relax(simd_context);
+	}
 
 	return true;
 }
@@ -126,9 +139,7 @@ static inline bool poly1305_emit_arch(void *ctx, u8 mac[POLY1305_MAC_SIZE],
 	    !state->is_base2_26 || !simd_use(simd_context)) {
 		convert_to_base2_64(ctx);
 		poly1305_emit_x86_64(ctx, mac, nonce);
-		return true;
-	}
-
-	poly1305_emit_avx(ctx, mac, nonce);
+	} else
+		poly1305_emit_avx(ctx, mac, nonce);
 	return true;
 }
diff --git a/src/crypto/zinc/poly1305/poly1305.c b/src/crypto/zinc/poly1305/poly1305.c
index d094ca7..ad758b5 100644
--- a/src/crypto/zinc/poly1305/poly1305.c
+++ b/src/crypto/zinc/poly1305/poly1305.c
@@ -28,7 +28,7 @@ static inline bool poly1305_init_arch(void *ctx,
 	return false;
 }
 static inline bool poly1305_blocks_arch(void *ctx, const u8 *input,
-					const size_t len, const u32 padbit,
+					size_t len, const u32 padbit,
 					simd_context_t *simd_context)
 {
 	return false;