crypto: pass simd by reference

Signed-off-by: Jason A. Donenfeld <Jason@zx2c4.com>

14 files changed, 89 insertions, 83 deletions

diff --git a/src/crypto/include/zinc/chacha20.h b/src/crypto/include/zinc/chacha20.h
index 43e67cb..8b3cc8c 100644
--- a/src/crypto/include/zinc/chacha20.h
+++ b/src/crypto/include/zinc/chacha20.h
@@ -44,11 +44,11 @@ static inline void chacha20_init(struct chacha20_ctx *state,
 	state->counter[3] = nonce >> 32;
 }
 void chacha20(struct chacha20_ctx *state, u8 *dst, const u8 *src, u32 len,
-	      simd_context_t simd_context);
+	      simd_context_t *simd_context);
 
 /* Derived key should be 32-bit aligned */
 void hchacha20(u8 derived_key[CHACHA20_KEY_SIZE],
 	       const u8 nonce[HCHACHA20_NONCE_SIZE],
-	       const u8 key[HCHACHA20_KEY_SIZE], simd_context_t simd_context);
+	       const u8 key[HCHACHA20_KEY_SIZE], simd_context_t *simd_context);
 
 #endif /* _ZINC_CHACHA20_H */
diff --git a/src/crypto/include/zinc/chacha20poly1305.h b/src/crypto/include/zinc/chacha20poly1305.h
index 0212685..03979b6 100644
--- a/src/crypto/include/zinc/chacha20poly1305.h
+++ b/src/crypto/include/zinc/chacha20poly1305.h
@@ -25,7 +25,7 @@ void chacha20poly1305_encrypt(u8 *dst, const u8 *src, const size_t src_len,
 bool __must_check chacha20poly1305_encrypt_sg(
 	struct scatterlist *dst, struct scatterlist *src, const size_t src_len,
 	const u8 *ad, const size_t ad_len, const u64 nonce,
-	const u8 key[CHACHA20POLY1305_KEYLEN], simd_context_t simd_context);
+	const u8 key[CHACHA20POLY1305_KEYLEN], simd_context_t *simd_context);
 
 bool __must_check
 chacha20poly1305_decrypt(u8 *dst, const u8 *src, const size_t src_len,
@@ -35,7 +35,7 @@ chacha20poly1305_decrypt(u8 *dst, const u8 *src, const size_t src_len,
 bool __must_check chacha20poly1305_decrypt_sg(
 	struct scatterlist *dst, struct scatterlist *src, const size_t src_len,
 	const u8 *ad, const size_t ad_len, const u64 nonce,
-	const u8 key[CHACHA20POLY1305_KEYLEN], simd_context_t simd_context);
+	const u8 key[CHACHA20POLY1305_KEYLEN], simd_context_t *simd_context);
 
 void xchacha20poly1305_encrypt(u8 *dst, const u8 *src, const size_t src_len,
 			       const u8 *ad, const size_t ad_len,
diff --git a/src/crypto/include/zinc/poly1305.h b/src/crypto/include/zinc/poly1305.h
index a7f552c..1cdbadd 100644
--- a/src/crypto/include/zinc/poly1305.h
+++ b/src/crypto/include/zinc/poly1305.h
@@ -26,9 +26,9 @@ void poly1305_fpu_init(void);
 
 void poly1305_init(struct poly1305_ctx *ctx, const u8 key[POLY1305_KEY_SIZE]);
 void poly1305_update(struct poly1305_ctx *ctx, const u8 *input, size_t len,
-		     simd_context_t simd_context);
+		     simd_context_t *simd_context);
 void poly1305_final(struct poly1305_ctx *ctx, u8 mac[POLY1305_MAC_SIZE],
-		    simd_context_t simd_context);
+		    simd_context_t *simd_context);
 
 #ifdef DEBUG
 bool poly1305_selftest(void);
diff --git a/src/crypto/zinc/chacha20/chacha20-arm-glue.h b/src/crypto/zinc/chacha20/chacha20-arm-glue.h
index e661ed2..0c8c9d5 100644
--- a/src/crypto/zinc/chacha20/chacha20-arm-glue.h
+++ b/src/crypto/zinc/chacha20/chacha20-arm-glue.h
@@ -29,10 +29,10 @@ void __init chacha20_fpu_init(void)
 
 static inline bool chacha20_arch(u8 *dst, const u8 *src, const size_t len,
 				 const u32 key[8], const u32 counter[4],
-				 simd_context_t simd_context)
+				 simd_context_t *simd_context)
 {
 #if defined(ARM_USE_NEON)
-	if (simd_context == HAVE_FULL_SIMD && chacha20_use_neon) {
+	if (chacha20_use_neon && simd_use(simd_context)) {
 		chacha20_neon(dst, src, len, key, counter);
 		return true;
 	}
@@ -42,7 +42,7 @@ static inline bool chacha20_arch(u8 *dst, const u8 *src, const size_t len,
 }
 
 static inline bool hchacha20_arch(u8 *derived_key, const u8 *nonce,
-				  const u8 *key, simd_context_t simd_context)
+				  const u8 *key, simd_context_t *simd_context)
 {
 	return false;
 }
diff --git a/src/crypto/zinc/chacha20/chacha20-mips-glue.h b/src/crypto/zinc/chacha20/chacha20-mips-glue.h
index ef8e4ab..e4185e1 100644
--- a/src/crypto/zinc/chacha20/chacha20-mips-glue.h
+++ b/src/crypto/zinc/chacha20/chacha20-mips-glue.h
@@ -13,14 +13,14 @@ void __init chacha20_fpu_init(void)
 
 static inline bool chacha20_arch(u8 *dst, const u8 *src, const size_t len,
 				 const u32 key[8], const u32 counter[4],
-				 simd_context_t simd_context)
+				 simd_context_t *simd_context)
 {
 	chacha20_mips(dst, src, len, key, counter);
 	return true;
 }
 
 static inline bool hchacha20_arch(u8 *derived_key, const u8 *nonce,
-				  const u8 *key, simd_context_t simd_context)
+				  const u8 *key, simd_context_t *simd_context)
 {
 	return false;
 }
diff --git a/src/crypto/zinc/chacha20/chacha20-x86_64-glue.h b/src/crypto/zinc/chacha20/chacha20-x86_64-glue.h
index 78270d7..34919c7 100644
--- a/src/crypto/zinc/chacha20/chacha20-x86_64-glue.h
+++ b/src/crypto/zinc/chacha20/chacha20-x86_64-glue.h
@@ -59,9 +59,9 @@ void __init chacha20_fpu_init(void)
 
 static inline bool chacha20_arch(u8 *dst, const u8 *src, const size_t len,
 				 const u32 key[8], const u32 counter[4],
-				 simd_context_t simd_context)
+				 simd_context_t *simd_context)
 {
-	if (simd_context != HAVE_FULL_SIMD)
+	if (!simd_use(simd_context))
 		return false;
 
 #ifdef CONFIG_AS_AVX512
@@ -90,10 +90,10 @@ static inline bool chacha20_arch(u8 *dst, const u8 *src, const size_t len,
 }
 
 static inline bool hchacha20_arch(u8 *derived_key, const u8 *nonce,
-				  const u8 *key, simd_context_t simd_context)
+				  const u8 *key, simd_context_t *simd_context)
 {
 #if defined(CONFIG_AS_SSSE3)
-	if (simd_context == HAVE_FULL_SIMD && chacha20_use_ssse3) {
+	if (chacha20_use_ssse3 && simd_use(simd_context)) {
 		hchacha20_ssse3(derived_key, nonce, key);
 		return true;
 	}
diff --git a/src/crypto/zinc/chacha20/chacha20.c b/src/crypto/zinc/chacha20/chacha20.c
index fdfccdc..da04d5b 100644
--- a/src/crypto/zinc/chacha20/chacha20.c
+++ b/src/crypto/zinc/chacha20/chacha20.c
@@ -18,12 +18,12 @@ void __init chacha20_fpu_init(void)
 }
 static inline bool chacha20_arch(u8 *out, const u8 *in, const size_t len,
 				 const u32 key[8], const u32 counter[4],
-				 simd_context_t simd_context)
+				 simd_context_t *simd_context)
 {
 	return false;
 }
 static inline bool hchacha20_arch(u8 *derived_key, const u8 *nonce,
-				  const u8 *key, simd_context_t simd_context)
+				  const u8 *key, simd_context_t *simd_context)
 {
 	return false;
 }
@@ -113,7 +113,7 @@ static void chacha20_generic(u8 *out, const u8 *in, u32 len, const u32 key[8],
 }
 
 void chacha20(struct chacha20_ctx *state, u8 *dst, const u8 *src, u32 len,
-	      simd_context_t simd_context)
+	      simd_context_t *simd_context)
 {
 	if (!chacha20_arch(dst, src, len, state->key, state->counter,
 			   simd_context))
@@ -157,7 +157,7 @@ static void hchacha20_generic(u8 derived_key[CHACHA20_KEY_SIZE],
 /* Derived key should be 32-bit aligned */
 void hchacha20(u8 derived_key[CHACHA20_KEY_SIZE],
 	       const u8 nonce[HCHACHA20_NONCE_SIZE],
-	       const u8 key[HCHACHA20_KEY_SIZE], simd_context_t simd_context)
+	       const u8 key[HCHACHA20_KEY_SIZE], simd_context_t *simd_context)
 {
 	if (!hchacha20_arch(derived_key, nonce, key, simd_context))
 		hchacha20_generic(derived_key, nonce, key);
diff --git a/src/crypto/zinc/chacha20poly1305.c b/src/crypto/zinc/chacha20poly1305.c
index 92a5b9c..7a8e03f 100644
--- a/src/crypto/zinc/chacha20poly1305.c
+++ b/src/crypto/zinc/chacha20poly1305.c
@@ -33,7 +33,7 @@ static inline void
 __chacha20poly1305_encrypt(u8 *dst, const u8 *src, const size_t src_len,
 			   const u8 *ad, const size_t ad_len, const u64 nonce,
 			   const u8 key[CHACHA20POLY1305_KEYLEN],
-			   simd_context_t simd_context)
+			   simd_context_t *simd_context)
 {
 	struct poly1305_ctx poly1305_state;
 	struct chacha20_ctx chacha20_state;
@@ -75,10 +75,10 @@ void chacha20poly1305_encrypt(u8 *dst, const u8 *src, const size_t src_len,
 {
 	simd_context_t simd_context;
 
-	simd_context = simd_get();
+	simd_get(&simd_context);
 	__chacha20poly1305_encrypt(dst, src, src_len, ad, ad_len, nonce, key,
-				   simd_context);
-	simd_put(simd_context);
+				   &simd_context);
+	simd_put(&simd_context);
 }
 EXPORT_SYMBOL(chacha20poly1305_encrypt);
 
@@ -87,7 +87,7 @@ bool chacha20poly1305_encrypt_sg(struct scatterlist *dst,
 				 const u8 *ad, const size_t ad_len,
 				 const u64 nonce,
 				 const u8 key[CHACHA20POLY1305_KEYLEN],
-				 simd_context_t simd_context)
+				 simd_context_t *simd_context)
 {
 	struct poly1305_ctx poly1305_state;
 	struct chacha20_ctx chacha20_state;
@@ -155,7 +155,7 @@ static inline bool
 __chacha20poly1305_decrypt(u8 *dst, const u8 *src, const size_t src_len,
 			   const u8 *ad, const size_t ad_len, const u64 nonce,
 			   const u8 key[CHACHA20POLY1305_KEYLEN],
-			   simd_context_t simd_context)
+			   simd_context_t *simd_context)
 {
 	struct poly1305_ctx poly1305_state;
 	struct chacha20_ctx chacha20_state;
@@ -208,10 +208,10 @@ bool chacha20poly1305_decrypt(u8 *dst, const u8 *src, const size_t src_len,
 {
 	simd_context_t simd_context, ret;
 
-	simd_context = simd_get();
+	simd_get(&simd_context);
 	ret = __chacha20poly1305_decrypt(dst, src, src_len, ad, ad_len, nonce,
-					 key, simd_context);
-	simd_put(simd_context);
+					 key, &simd_context);
+	simd_put(&simd_context);
 	return ret;
 }
 EXPORT_SYMBOL(chacha20poly1305_decrypt);
@@ -221,7 +221,7 @@ bool chacha20poly1305_decrypt_sg(struct scatterlist *dst,
 				 const u8 *ad, const size_t ad_len,
 				 const u64 nonce,
 				 const u8 key[CHACHA20POLY1305_KEYLEN],
-				 simd_context_t simd_context)
+				 simd_context_t *simd_context)
 {
 	struct poly1305_ctx poly1305_state;
 	struct chacha20_ctx chacha20_state;
@@ -300,15 +300,16 @@ void xchacha20poly1305_encrypt(u8 *dst, const u8 *src, const size_t src_len,
 			       const u8 nonce[XCHACHA20POLY1305_NONCELEN],
 			       const u8 key[CHACHA20POLY1305_KEYLEN])
 {
-	simd_context_t simd_context = simd_get();
+	simd_context_t simd_context;
 	u8 derived_key[CHACHA20POLY1305_KEYLEN] __aligned(16);
 
-	hchacha20(derived_key, nonce, key, simd_context);
+	simd_get(&simd_context);
+	hchacha20(derived_key, nonce, key, &simd_context);
 	__chacha20poly1305_encrypt(dst, src, src_len, ad, ad_len,
 				   get_unaligned_le64(nonce + 16),
-				   derived_key, simd_context);
+				   derived_key, &simd_context);
 	memzero_explicit(derived_key, CHACHA20POLY1305_KEYLEN);
-	simd_put(simd_context);
+	simd_put(&simd_context);
 }
 EXPORT_SYMBOL(xchacha20poly1305_encrypt);
 
@@ -317,15 +318,17 @@ bool xchacha20poly1305_decrypt(u8 *dst, const u8 *src, const size_t src_len,
 			       const u8 nonce[XCHACHA20POLY1305_NONCELEN],
 			       const u8 key[CHACHA20POLY1305_KEYLEN])
 {
-	bool ret, simd_context = simd_get();
+	bool ret;
+	simd_context_t simd_context;
 	u8 derived_key[CHACHA20POLY1305_KEYLEN] __aligned(16);
 
-	hchacha20(derived_key, nonce, key, simd_context);
+	simd_get(&simd_context);
+	hchacha20(derived_key, nonce, key, &simd_context);
 	ret = __chacha20poly1305_decrypt(dst, src, src_len, ad, ad_len,
 					 get_unaligned_le64(nonce + 16),
-					 derived_key, simd_context);
+					 derived_key, &simd_context);
 	memzero_explicit(derived_key, CHACHA20POLY1305_KEYLEN);
-	simd_put(simd_context);
+	simd_put(&simd_context);
 	return ret;
 }
 EXPORT_SYMBOL(xchacha20poly1305_decrypt);
diff --git a/src/crypto/zinc/poly1305/poly1305-arm-glue.h b/src/crypto/zinc/poly1305/poly1305-arm-glue.h
index 8181703..6ec2fc8 100644
--- a/src/crypto/zinc/poly1305/poly1305-arm-glue.h
+++ b/src/crypto/zinc/poly1305/poly1305-arm-glue.h
@@ -39,10 +39,10 @@ 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,
-					simd_context_t simd_context)
+					simd_context_t *simd_context)
 {
 #if defined(ARM_USE_NEON)
-	if (simd_context == HAVE_FULL_SIMD && poly1305_use_neon) {
+	if (poly1305_use_neon && simd_use(simd_context)) {
 		poly1305_blocks_neon(ctx, inp, len, padbit);
 		return true;
 	}
@@ -53,10 +53,10 @@ static inline bool poly1305_blocks_arch(void *ctx, const u8 *inp,
 
 static inline bool poly1305_emit_arch(void *ctx, u8 mac[POLY1305_MAC_SIZE],
 				      const u32 nonce[4],
-				      simd_context_t simd_context)
+				      simd_context_t *simd_context)
 {
 #if defined(ARM_USE_NEON)
-	if (simd_context == HAVE_FULL_SIMD && poly1305_use_neon) {
+	if (poly1305_use_neon && simd_use(simd_context)) {
 		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 960abee..0e72c8b 100644
--- a/src/crypto/zinc/poly1305/poly1305-mips-glue.h
+++ b/src/crypto/zinc/poly1305/poly1305-mips-glue.h
@@ -22,7 +22,7 @@ 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,
-					simd_context_t simd_context)
+					simd_context_t *simd_context)
 {
 	poly1305_blocks_mips(ctx, inp, len, padbit);
 	return true;
@@ -30,7 +30,7 @@ static inline bool poly1305_blocks_arch(void *ctx, const u8 *inp,
 
 static inline bool poly1305_emit_arch(void *ctx, u8 mac[POLY1305_MAC_SIZE],
 				      const u32 nonce[4],
-				      simd_context_t simd_context)
+				      simd_context_t *simd_context)
 {
 	poly1305_emit_mips(ctx, mac, nonce);
 	return true;
diff --git a/src/crypto/zinc/poly1305/poly1305-x86_64-glue.h b/src/crypto/zinc/poly1305/poly1305-x86_64-glue.h
index 7f1af44..1afd1c5 100644
--- a/src/crypto/zinc/poly1305/poly1305-x86_64-glue.h
+++ b/src/crypto/zinc/poly1305/poly1305-x86_64-glue.h
@@ -63,20 +63,20 @@ 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,
-					simd_context_t simd_context)
+					simd_context_t *simd_context)
 {
 #ifdef CONFIG_AS_AVX512
-	if (poly1305_use_avx512 && simd_context == HAVE_FULL_SIMD)
+	if (poly1305_use_avx512 && simd_use(simd_context))
 		poly1305_blocks_avx512(ctx, inp, len, padbit);
 	else
 #endif
 #ifdef CONFIG_AS_AVX2
-	if (poly1305_use_avx2 && simd_context == HAVE_FULL_SIMD)
+	if (poly1305_use_avx2 && simd_use(simd_context))
 		poly1305_blocks_avx2(ctx, inp, len, padbit);
 	else
 #endif
 #ifdef CONFIG_AS_AVX
-	if (poly1305_use_avx && simd_context == HAVE_FULL_SIMD)
+	if (poly1305_use_avx && simd_use(simd_context))
 		poly1305_blocks_avx(ctx, inp, len, padbit);
 	else
 #endif
@@ -86,20 +86,20 @@ static inline bool poly1305_blocks_arch(void *ctx, const u8 *inp,
 
 static inline bool poly1305_emit_arch(void *ctx, u8 mac[POLY1305_MAC_SIZE],
 				      const u32 nonce[4],
-				      simd_context_t simd_context)
+				      simd_context_t *simd_context)
 {
 #ifdef CONFIG_AS_AVX512
-	if (poly1305_use_avx512 && simd_context == HAVE_FULL_SIMD)
+	if (poly1305_use_avx512 && simd_use(simd_context))
 		poly1305_emit_avx(ctx, mac, nonce);
 	else
 #endif
 #ifdef CONFIG_AS_AVX2
-	if (poly1305_use_avx2 && simd_context == HAVE_FULL_SIMD)
+	if (poly1305_use_avx2 && simd_use(simd_context))
 		poly1305_emit_avx(ctx, mac, nonce);
 	else
 #endif
 #ifdef CONFIG_AS_AVX
-	if (poly1305_use_avx && simd_context == HAVE_FULL_SIMD)
+	if (poly1305_use_avx && simd_use(simd_context))
 		poly1305_emit_avx(ctx, mac, nonce);
 	else
 #endif
diff --git a/src/crypto/zinc/poly1305/poly1305.c b/src/crypto/zinc/poly1305/poly1305.c
index d8c103f..a098b61 100644
--- a/src/crypto/zinc/poly1305/poly1305.c
+++ b/src/crypto/zinc/poly1305/poly1305.c
@@ -21,13 +21,13 @@ static inline bool poly1305_init_arch(void *ctx,
 }
 static inline bool poly1305_blocks_arch(void *ctx, const u8 *input,
 					const size_t len, const u32 padbit,
-					simd_context_t simd_context)
+					simd_context_t *simd_context)
 {
 	return false;
 }
 static inline bool poly1305_emit_arch(void *ctx, u8 mac[POLY1305_MAC_SIZE],
 				      const u32 nonce[4],
-				      simd_context_t simd_context)
+				      simd_context_t *simd_context)
 {
 	return false;
 }
@@ -58,7 +58,7 @@ EXPORT_SYMBOL(poly1305_init);
 
 static inline void poly1305_blocks(void *ctx, const u8 *input, const size_t len,
 				   const u32 padbit,
-				   simd_context_t simd_context)
+				   simd_context_t *simd_context)
 {
 	if (!poly1305_blocks_arch(ctx, input, len, padbit, simd_context))
 		poly1305_blocks_generic(ctx, input, len, padbit);
@@ -66,14 +66,14 @@ static inline void poly1305_blocks(void *ctx, const u8 *input, const size_t len,
 
 static inline void poly1305_emit(void *ctx, u8 mac[POLY1305_KEY_SIZE],
 				 const u32 nonce[4],
-				 simd_context_t simd_context)
+				 simd_context_t *simd_context)
 {
 	if (!poly1305_emit_arch(ctx, mac, nonce, simd_context))
 		poly1305_emit_generic(ctx, mac, nonce);
 }
 
 void poly1305_update(struct poly1305_ctx *ctx, const u8 *input, size_t len,
-		     simd_context_t simd_context)
+		     simd_context_t *simd_context)
 {
 	const size_t num = ctx->num % POLY1305_BLOCK_SIZE;
 	size_t rem;
@@ -108,7 +108,7 @@ void poly1305_update(struct poly1305_ctx *ctx, const u8 *input, size_t len,
 EXPORT_SYMBOL(poly1305_update);
 
 void poly1305_final(struct poly1305_ctx *ctx, u8 mac[POLY1305_MAC_SIZE],
-		    simd_context_t simd_context)
+		    simd_context_t *simd_context)
 {
 	size_t num = ctx->num % POLY1305_BLOCK_SIZE;
 
diff --git a/src/crypto/zinc/selftest/chacha20poly1305.h b/src/crypto/zinc/selftest/chacha20poly1305.h
index e5e8de5..9aae3f5 100644
--- a/src/crypto/zinc/selftest/chacha20poly1305.h
+++ b/src/crypto/zinc/selftest/chacha20poly1305.h
@@ -7635,7 +7635,7 @@ chacha20poly1305_selftest_encrypt_bignonce(u8 *dst, const u8 *src,
 					   const u8 nonce[12],
 					   const u8 key[CHACHA20POLY1305_KEYLEN])
 {
-	simd_context_t simd_context = simd_get();
+	simd_context_t simd_context;
 	struct poly1305_ctx poly1305_state;
 	struct chacha20_ctx chacha20_state;
 	union {
@@ -7643,26 +7643,27 @@ chacha20poly1305_selftest_encrypt_bignonce(u8 *dst, const u8 *src,
 		__le64 lens[2];
 	} b = {{ 0 }};
 
+	simd_get(&simd_context);
 	chacha20_init(&chacha20_state, key, 0);
 	chacha20_state.counter[1] = get_unaligned_le32(nonce + 0);
 	chacha20_state.counter[2] = get_unaligned_le32(nonce + 4);
 	chacha20_state.counter[3] = get_unaligned_le32(nonce + 8);
 	chacha20(&chacha20_state, b.block0, b.block0, sizeof(b.block0),
-		 simd_context);
+		 &simd_context);
 	poly1305_init(&poly1305_state, b.block0);
-	poly1305_update(&poly1305_state, ad, ad_len, simd_context);
+	poly1305_update(&poly1305_state, ad, ad_len, &simd_context);
 	poly1305_update(&poly1305_state, pad0, (0x10 - ad_len) & 0xf,
-			simd_context);
-	chacha20(&chacha20_state, dst, src, src_len, simd_context);
-	poly1305_update(&poly1305_state, dst, src_len, simd_context);
+			&simd_context);
+	chacha20(&chacha20_state, dst, src, src_len, &simd_context);
+	poly1305_update(&poly1305_state, dst, src_len, &simd_context);
 	poly1305_update(&poly1305_state, pad0, (0x10 - src_len) & 0xf,
-			simd_context);
+			&simd_context);
 	b.lens[0] = cpu_to_le64(ad_len);
 	b.lens[1] = cpu_to_le64(src_len);
 	poly1305_update(&poly1305_state, (u8 *)b.lens, sizeof(b.lens),
-			simd_context);
-	poly1305_final(&poly1305_state, dst + src_len, simd_context);
-	simd_put(simd_context);
+			&simd_context);
+	poly1305_final(&poly1305_state, dst + src_len, &simd_context);
+	simd_put(&simd_context);
 	memzero_explicit(&chacha20_state, sizeof(chacha20_state));
 	memzero_explicit(&b, sizeof(b));
 }
@@ -7698,7 +7699,8 @@ bool __init chacha20poly1305_selftest(void)
 {
 	size_t i;
 	u8 computed_result[MAXIMUM_TEST_BUFFER_LEN], *heap_src, *heap_dst;
-	bool success = true, ret, simd_context;
+	bool success = true, ret;
+	simd_context_t simd_context;
 	struct scatterlist sg_src, sg_dst;
 
 	heap_src = kmalloc(MAXIMUM_TEST_BUFFER_LEN, GFP_KERNEL);
@@ -7729,7 +7731,7 @@ bool __init chacha20poly1305_selftest(void)
 			success = false;
 		}
 	}
-	simd_context = simd_get();
+	simd_get(&simd_context);
 	for (i = 0; i < ARRAY_SIZE(chacha20poly1305_enc_vectors); ++i) {
 		if (chacha20poly1305_enc_vectors[i].nlen != 8)
 			continue;
@@ -7747,7 +7749,7 @@ bool __init chacha20poly1305_selftest(void)
 			chacha20poly1305_enc_vectors[i].alen,
 			get_unaligned_le64(chacha20poly1305_enc_vectors[i].nonce),
 			chacha20poly1305_enc_vectors[i].key,
-			simd_context);
+			&simd_context);
 		if (!ret || memcmp(heap_dst,
 				   chacha20poly1305_enc_vectors[i].result,
 				   chacha20poly1305_enc_vectors[i].ilen +
@@ -7757,7 +7759,7 @@ bool __init chacha20poly1305_selftest(void)
 			success = false;
 		}
 	}
-	simd_put(simd_context);
+	simd_put(&simd_context);
 	for (i = 0; i < ARRAY_SIZE(chacha20poly1305_dec_vectors); ++i) {
 		memset(computed_result, 0, sizeof(computed_result));
 		ret = chacha20poly1305_decrypt(computed_result,
@@ -7778,7 +7780,7 @@ bool __init chacha20poly1305_selftest(void)
 			success = false;
 		}
 	}
-	simd_context = simd_get();
+	simd_get(&simd_context);
 	for (i = 0; i < ARRAY_SIZE(chacha20poly1305_dec_vectors); ++i) {
 		memset(heap_dst, 0, MAXIMUM_TEST_BUFFER_LEN);
 		memcpy(heap_src, chacha20poly1305_dec_vectors[i].input,
@@ -7793,7 +7795,7 @@ bool __init chacha20poly1305_selftest(void)
 			chacha20poly1305_dec_vectors[i].assoc,
 			chacha20poly1305_dec_vectors[i].alen,
 			get_unaligned_le64(chacha20poly1305_dec_vectors[i].nonce),
-			chacha20poly1305_dec_vectors[i].key, simd_context);
+			chacha20poly1305_dec_vectors[i].key, &simd_context);
 		if (!decryption_success(ret,
 			chacha20poly1305_dec_vectors[i].failure,
 			memcmp(heap_dst, chacha20poly1305_dec_vectors[i].result,
@@ -7804,7 +7806,7 @@ bool __init chacha20poly1305_selftest(void)
 			success = false;
 		}
 	}
-	simd_put(simd_context);
+	simd_put(&simd_context);
 	for (i = 0; i < ARRAY_SIZE(xchacha20poly1305_enc_vectors); ++i) {
 		memset(computed_result, 0, sizeof(computed_result));
 		xchacha20poly1305_encrypt(computed_result,
diff --git a/src/crypto/zinc/selftest/poly1305.h b/src/crypto/zinc/selftest/poly1305.h
index d02941f..02cd4ba 100644
--- a/src/crypto/zinc/selftest/poly1305.h
+++ b/src/crypto/zinc/selftest/poly1305.h
@@ -820,10 +820,11 @@ static const struct poly1305_testvec poly1305_testvecs[] __initconst = {
 
 bool __init poly1305_selftest(void)
 {
-	simd_context_t simd_context = simd_get();
+	simd_context_t simd_context;
 	bool success = true;
 	size_t i, j;
 
+	simd_get(&simd_context);
 	for (i = 0; i < ARRAY_SIZE(poly1305_testvecs); ++i) {
 		struct poly1305_ctx poly1305;
 		u8 out[POLY1305_MAC_SIZE];
@@ -832,14 +833,14 @@ bool __init poly1305_selftest(void)
 		memset(&poly1305, 0, sizeof(poly1305));
 		poly1305_init(&poly1305, poly1305_testvecs[i].key);
 		poly1305_update(&poly1305, poly1305_testvecs[i].input,
-				poly1305_testvecs[i].ilen, simd_context);
-		poly1305_final(&poly1305, out, simd_context);
+				poly1305_testvecs[i].ilen, &simd_context);
+		poly1305_final(&poly1305, out, &simd_context);
 		if (memcmp(out, poly1305_testvecs[i].output,
 			   POLY1305_MAC_SIZE)) {
 			pr_info("poly1305 self-test %zu: FAIL\n", i + 1);
 			success = false;
 		}
-		simd_context = simd_relax(simd_context);
+		simd_relax(&simd_context);
 
 		if (poly1305_testvecs[i].ilen <= 1)
 			continue;
@@ -849,22 +850,22 @@ bool __init poly1305_selftest(void)
 			memset(&poly1305, 0, sizeof(poly1305));
 			poly1305_init(&poly1305, poly1305_testvecs[i].key);
 			poly1305_update(&poly1305, poly1305_testvecs[i].input,
-					j, simd_context);
+					j, &simd_context);
 			poly1305_update(&poly1305,
 					poly1305_testvecs[i].input + j,
 					poly1305_testvecs[i].ilen - j,
-					simd_context);
-			poly1305_final(&poly1305, out, simd_context);
+					&simd_context);
+			poly1305_final(&poly1305, out, &simd_context);
 			if (memcmp(out, poly1305_testvecs[i].output,
 				   POLY1305_MAC_SIZE)) {
 				pr_info("poly1305 self-test %zu (split %zu): FAIL\n",
 					i + 1, j);
 				success = false;
 			}
-			simd_context = simd_relax(simd_context);
+			simd_relax(&simd_context);
 		}
 	}
-	simd_put(simd_context);
+	simd_put(&simd_context);
 
 	if (success)
 		pr_info("poly1305 self-tests: pass\n");