Add a control to disable the Poly1305 NEON code.

Some phones have a buggy NEON unit and the Poly1305 NEON code fails on
them, even though other NEON code appears to work fine.

This change:

1) Fixes a bug where NEON was assumed even when the code wasn't compiled
   in NEON mode.

2) Adds a second NEON control bit that can be disabled in order to run
   NEON code, but not the Poly1305 NEON code.

https://code.google.com/p/chromium/issues/detail?id=341598

Change-Id: Icb121bf8dba47c7a46c7667f676ff7a4bc973625
Reviewed-on: https://boringssl-review.googlesource.com/1351
Reviewed-by: Adam Langley <agl@google.com>

crypto/arm_arch.h

@@ -88,13 +88,22 @@
 #endif
 
 #if !__ASSEMBLER__
+
 /* OPENSSL_armcap_P contains flags describing the capabilities of the CPU and
  * is easy for assembly code to acesss. For C code, see the functions in
  * |cpu.h|. */
-extern unsigned int OPENSSL_armcap_P;
+extern uint32_t OPENSSL_armcap_P;
 
-#define ARMV7_NEON      (1<<0)
-#endif
+/* ARMV7_NEON is true when a NEON unit is present in the current CPU. */
+#define ARMV7_NEON (1 << 0)
+
+/* ARMV7_NEON_FUNCTIONAL is true when the NEON unit doesn't contain subtle bugs.
+ * The Poly1305 NEON code is known to trigger bugs in the NEON units of some
+ * phones. If this bit isn't set then the Poly1305 NEON code won't be used.
+ * See https://code.google.com/p/chromium/issues/detail?id=341598. */
+#define ARMV7_NEON_FUNCTIONAL (1 << 1)
+
+#endif  /* !__ASSEMBLER__ */
 
 
 #endif  /* OPENSSL_HEADER_THREAD_H */

crypto/cpu-arm.c

@@ -64,9 +64,9 @@
 #include "arm_arch.h"
 
 #if defined(__ARM_NEON__)
-uint32_t OPENSSL_armcap_P = ARMV7_NEON;
+uint32_t OPENSSL_armcap_P = ARMV7_NEON | ARMV7_NEON_FUNCTIONAL;
 #else
-uint32_t OPENSSL_armcap_P = ARMV7_NEON;
+uint32_t OPENSSL_armcap_P = ARMV7_NEON_FUNCTIONAL;
 #endif
 
 char CRYPTO_is_NEON_capable() {
@@ -81,4 +81,17 @@ void CRYPTO_set_NEON_capable(char neon_capable) {
   }
 }
 
+char CRYPTO_is_NEON_functional() {
+  static const uint32_t kWantFlags = ARMV7_NEON | ARMV7_NEON_FUNCTIONAL;
+  return (OPENSSL_armcap_P & kWantFlags) == kWantFlags;
+}
+
+void CRYPTO_set_NEON_functional(char neon_functional) {
+  if (neon_functional) {
+    OPENSSL_armcap_P |= ARMV7_NEON_FUNCTIONAL;
+  } else {
+    OPENSSL_armcap_P &= ~ARMV7_NEON_FUNCTIONAL;
+  }
+}
+
 #endif  /* defined(OPENSSL_ARM) */

crypto/poly1305/poly1305.c

@@ -166,7 +166,7 @@ void CRYPTO_poly1305_init(poly1305_state *statep, const uint8_t key[32]) {
   uint32_t t0, t1, t2, t3;
 
 #if defined(OPENSSL_ARM)
-  if (CRYPTO_is_NEON_capable()) {
+  if (CRYPTO_is_NEON_functional()) {
     CRYPTO_poly1305_init_neon(statep, key);
     return;
   }
@@ -213,7 +213,7 @@ void CRYPTO_poly1305_update(poly1305_state *statep, const uint8_t *in,
   struct poly1305_state_st *state = (struct poly1305_state_st *)statep;
 
 #if defined(OPENSSL_ARM)
-  if (CRYPTO_is_NEON_capable()) {
+  if (CRYPTO_is_NEON_functional()) {
     CRYPTO_poly1305_update_neon(statep, in, in_len);
     return;
   }
@@ -256,7 +256,7 @@ void CRYPTO_poly1305_finish(poly1305_state *statep, uint8_t mac[16]) {
   uint32_t b, nb;
 
 #if defined(OPENSSL_ARM)
-  if (CRYPTO_is_NEON_capable()) {
+  if (CRYPTO_is_NEON_functional()) {
     CRYPTO_poly1305_finish_neon(statep, mac);
     return;
   }

include/openssl/cpu.h

@@ -91,8 +91,21 @@ OPENSSL_EXPORT char CRYPTO_is_NEON_capable();
 
 /* CRYPTO_set_NEON_capable sets the return value of |CRYPTO_is_NEON_capable|.
  * By default, unless the code was compiled with |-mfpu=neon|, NEON is assumed
- * not to be present. It is not autodetected. */
+ * not to be present. It is not autodetected. Calling this with a zero
+ * argument also causes |CRYPTO_is_NEON_functional| to return false. */
 OPENSSL_EXPORT void CRYPTO_set_NEON_capable(char neon_capable);
+
+/* CRYPTO_is_NEON_functional returns true if the current CPU has a /working/
+ * NEON unit. Some phones have a NEON unit, but the Poly1305 NEON code causes
+ * it to fail. See https://code.google.com/p/chromium/issues/detail?id=341598 */
+OPENSSL_EXPORT char CRYPTO_is_NEON_functional();
+
+/* CRYPTO_set_NEON_functional sets the "NEON functional" flag. For
+ * |CRYPTO_is_NEON_functional| to return true, both this flag and the NEON flag
+ * must be true. By default NEON is assumed to be functional if the code was
+ * compiled with |-mfpu=neon| or if |CRYPTO_set_NEON_capable| has been called
+ * with a non-zero argument. */
+OPENSSL_EXPORT void CRYPTO_set_NEON_functional(char neon_functional);
 #endif  /* OPENSSL_ARM */