Merge branch 'master' of https://github.com/MoonchildProductions/UXP into pm_url_1

1 files changed, 122 insertions, 49 deletions

diff --git a/security/nss/lib/freebl/gcm.c b/security/nss/lib/freebl/gcm.c
index f1e16da78..0fdb0fd48 100644
--- a/security/nss/lib/freebl/gcm.c
+++ b/security/nss/lib/freebl/gcm.c
@@ -17,50 +17,18 @@
 
 #include <limits.h>
 
+#ifdef NSS_X86_OR_X64
+#include <wmmintrin.h> /* clmul */
+#endif
+
 /* Forward declarations */
-SECStatus gcm_HashInit_hw(gcmHashContext *ghash);
-SECStatus gcm_HashWrite_hw(gcmHashContext *ghash, unsigned char *outbuf);
 SECStatus gcm_HashMult_hw(gcmHashContext *ghash, const unsigned char *buf,
                           unsigned int count);
-SECStatus gcm_HashZeroX_hw(gcmHashContext *ghash);
 SECStatus gcm_HashMult_sftw(gcmHashContext *ghash, const unsigned char *buf,
                             unsigned int count);
 SECStatus gcm_HashMult_sftw32(gcmHashContext *ghash, const unsigned char *buf,
                               unsigned int count);
 
-/* Stub definitions for the above *_hw functions, which shouldn't be
- * used unless NSS_X86_OR_X64 is defined */
-#ifndef NSS_X86_OR_X64
-SECStatus
-gcm_HashWrite_hw(gcmHashContext *ghash, unsigned char *outbuf)
-{
-    PORT_SetError(SEC_ERROR_LIBRARY_FAILURE);
-    return SECFailure;
-}
-
-SECStatus
-gcm_HashMult_hw(gcmHashContext *ghash, const unsigned char *buf,
-                unsigned int count)
-{
-    PORT_SetError(SEC_ERROR_LIBRARY_FAILURE);
-    return SECFailure;
-}
-
-SECStatus
-gcm_HashInit_hw(gcmHashContext *ghash)
-{
-    PORT_SetError(SEC_ERROR_LIBRARY_FAILURE);
-    return SECFailure;
-}
-
-SECStatus
-gcm_HashZeroX_hw(gcmHashContext *ghash)
-{
-    PORT_SetError(SEC_ERROR_LIBRARY_FAILURE);
-    return SECFailure;
-}
-#endif /* NSS_X86_OR_X64 */
-
 uint64_t
 get64(const unsigned char *bytes)
 {
@@ -78,8 +46,6 @@ get64(const unsigned char *bytes)
 SECStatus
 gcmHash_InitContext(gcmHashContext *ghash, const unsigned char *H, PRBool sw)
 {
-    SECStatus rv = SECSuccess;
-
     ghash->cLen = 0;
     ghash->bufLen = 0;
     PORT_Memset(ghash->counterBuf, 0, sizeof(ghash->counterBuf));
@@ -87,7 +53,17 @@ gcmHash_InitContext(gcmHashContext *ghash, const unsigned char *H, PRBool sw)
     ghash->h_low = get64(H + 8);
     ghash->h_high = get64(H);
     if (clmul_support() && !sw) {
-        rv = gcm_HashInit_hw(ghash);
+#ifdef NSS_X86_OR_X64
+        ghash->ghash_mul = gcm_HashMult_hw;
+        ghash->x = _mm_setzero_si128();
+        /* MSVC requires __m64 to load epi64. */
+        ghash->h = _mm_set_epi32(ghash->h_high >> 32, (uint32_t)ghash->h_high,
+                                 ghash->h_low >> 32, (uint32_t)ghash->h_low);
+        ghash->hw = PR_TRUE;
+#else
+        PORT_SetError(SEC_ERROR_LIBRARY_FAILURE);
+        return SECFailure;
+#endif /* NSS_X86_OR_X64 */
     } else {
 /* We fall back to the software implementation if we can't use / don't
          * want to use pclmul. */
@@ -99,7 +75,7 @@ gcmHash_InitContext(gcmHashContext *ghash, const unsigned char *H, PRBool sw)
         ghash->x_high = ghash->x_low = 0;
         ghash->hw = PR_FALSE;
     }
-    return rv;
+    return SECSuccess;
 }
 
 #ifdef HAVE_INT128_SUPPORT
@@ -307,17 +283,102 @@ gcm_HashMult_sftw32(gcmHashContext *ghash, const unsigned char *buf,
 }
 #endif /* HAVE_INT128_SUPPORT */
 
+SECStatus
+gcm_HashMult_hw(gcmHashContext *ghash, const unsigned char *buf,
+                unsigned int count)
+{
+#ifdef NSS_X86_OR_X64
+    size_t i;
+    pre_align __m128i z_high post_align;
+    pre_align __m128i z_low post_align;
+    pre_align __m128i C post_align;
+    pre_align __m128i D post_align;
+    pre_align __m128i E post_align;
+    pre_align __m128i F post_align;
+    pre_align __m128i bin post_align;
+    pre_align __m128i Ci post_align;
+    pre_align __m128i tmp post_align;
+
+    for (i = 0; i < count; i++, buf += 16) {
+        bin = _mm_set_epi16(((uint16_t)buf[0] << 8) | buf[1],
+                            ((uint16_t)buf[2] << 8) | buf[3],
+                            ((uint16_t)buf[4] << 8) | buf[5],
+                            ((uint16_t)buf[6] << 8) | buf[7],
+                            ((uint16_t)buf[8] << 8) | buf[9],
+                            ((uint16_t)buf[10] << 8) | buf[11],
+                            ((uint16_t)buf[12] << 8) | buf[13],
+                            ((uint16_t)buf[14] << 8) | buf[15]);
+        Ci = _mm_xor_si128(bin, ghash->x);
+
+        /* Do binary mult ghash->X = Ci * ghash->H. */
+        C = _mm_clmulepi64_si128(Ci, ghash->h, 0x00);
+        D = _mm_clmulepi64_si128(Ci, ghash->h, 0x11);
+        E = _mm_clmulepi64_si128(Ci, ghash->h, 0x01);
+        F = _mm_clmulepi64_si128(Ci, ghash->h, 0x10);
+        tmp = _mm_xor_si128(E, F);
+        z_high = _mm_xor_si128(tmp, _mm_slli_si128(D, 8));
+        z_high = _mm_unpackhi_epi64(z_high, D);
+        z_low = _mm_xor_si128(_mm_slli_si128(tmp, 8), C);
+        z_low = _mm_unpackhi_epi64(_mm_slli_si128(C, 8), z_low);
+
+        /* Shift one to the left (multiply by x) as gcm spec is stupid. */
+        C = _mm_slli_si128(z_low, 8);
+        E = _mm_srli_epi64(C, 63);
+        D = _mm_slli_si128(z_high, 8);
+        F = _mm_srli_epi64(D, 63);
+        /* Carry over */
+        C = _mm_srli_si128(z_low, 8);
+        D = _mm_srli_epi64(C, 63);
+        z_low = _mm_or_si128(_mm_slli_epi64(z_low, 1), E);
+        z_high = _mm_or_si128(_mm_or_si128(_mm_slli_epi64(z_high, 1), F), D);
+
+        /* Reduce */
+        C = _mm_slli_si128(z_low, 8);
+        /* D = z_low << 127 */
+        D = _mm_slli_epi64(C, 63);
+        /* E = z_low << 126 */
+        E = _mm_slli_epi64(C, 62);
+        /* F = z_low << 121 */
+        F = _mm_slli_epi64(C, 57);
+        /* z_low ^= (z_low << 127) ^ (z_low << 126) ^ (z_low << 121); */
+        z_low = _mm_xor_si128(_mm_xor_si128(_mm_xor_si128(z_low, D), E), F);
+        C = _mm_srli_si128(z_low, 8);
+        /* D = z_low >> 1 */
+        D = _mm_slli_epi64(C, 63);
+        D = _mm_or_si128(_mm_srli_epi64(z_low, 1), D);
+        /* E = z_low >> 2 */
+        E = _mm_slli_epi64(C, 62);
+        E = _mm_or_si128(_mm_srli_epi64(z_low, 2), E);
+        /* F = z_low >> 7 */
+        F = _mm_slli_epi64(C, 57);
+        F = _mm_or_si128(_mm_srli_epi64(z_low, 7), F);
+        /* ghash->x ^= z_low ^ (z_low >> 1) ^ (z_low >> 2) ^ (z_low >> 7); */
+        ghash->x = _mm_xor_si128(_mm_xor_si128(
+                                     _mm_xor_si128(_mm_xor_si128(z_high, z_low), D), E),
+                                 F);
+    }
+    return SECSuccess;
+#else
+    PORT_SetError(SEC_ERROR_LIBRARY_FAILURE);
+    return SECFailure;
+#endif /* NSS_X86_OR_X64 */
+}
+
 static SECStatus
 gcm_zeroX(gcmHashContext *ghash)
 {
-    SECStatus rv = SECSuccess;
-
     if (ghash->hw) {
-        rv = gcm_HashZeroX_hw(ghash);
+#ifdef NSS_X86_OR_X64
+        ghash->x = _mm_setzero_si128();
+        return SECSuccess;
+#else
+        PORT_SetError(SEC_ERROR_LIBRARY_FAILURE);
+        return SECFailure;
+#endif /* NSS_X86_OR_X64 */
     }
 
     ghash->x_high = ghash->x_low = 0;
-    return rv;
+    return SECSuccess;
 }
 
 /*
@@ -442,10 +503,15 @@ gcmHash_Final(gcmHashContext *ghash, unsigned char *outbuf,
     }
 
     if (ghash->hw) {
-        rv = gcm_HashWrite_hw(ghash, T);
-        if (rv != SECSuccess) {
-            goto cleanup;
-        }
+#ifdef NSS_X86_OR_X64
+        uint64_t tmp_out[2];
+        _mm_storeu_si128((__m128i *)tmp_out, ghash->x);
+        WRITE64(tmp_out[0], T + 8);
+        WRITE64(tmp_out[1], T);
+#else
+        PORT_SetError(SEC_ERROR_LIBRARY_FAILURE);
+        return SECFailure;
+#endif /* NSS_X86_OR_X64 */
     } else {
         WRITE64(ghash->x_low, T + 8);
         WRITE64(ghash->x_high, T);
@@ -529,7 +595,14 @@ GCM_CreateContext(void *context, freeblCipherFunc cipher,
     if (gcm == NULL) {
         return NULL;
     }
-    ghash = PORT_ZNewAligned(gcmHashContext, 16, mem);
+    /* aligned_alloc is C11 so we have to do it the old way. */
+    ghash = PORT_ZAlloc(sizeof(gcmHashContext) + 15);
+    if (ghash == NULL) {
+        PORT_SetError(SEC_ERROR_NO_MEMORY);
+        goto loser;
+    }
+    ghash->mem = ghash;
+    ghash = (gcmHashContext *)(((uintptr_t)ghash + 15) & ~(uintptr_t)0x0F);
 
     /* first plug in the ghash context */
     gcm->ghash_context = ghash;