Enable AMD64 3DES implementation on WIN64

* cipher/des-amd64.S: Enable when
HAVE_COMPATIBLE_GCC_WIN64_PLATFORM_AS defined.
(ELF): New macro to mask lines with ELF specific commands.
* cipher/des.c (USE_AMD64_ASM): Enable when
HAVE_COMPATIBLE_GCC_WIN64_PLATFORM_AS defined.
[HAVE_COMPATIBLE_GCC_WIN64_PLATFORM_AS] (call_sysv_fn): New.
(tripledes_ecb_crypt) [HAVE_COMPATIBLE_GCC_WIN64_PLATFORM_AS]: Call
assembly function through 'call_sysv_fn'.
(tripledes_amd64_ctr_enc, tripledes_amd64_cbc_dec)
(tripledes_amd64_cfb_dec): New wrapper functions for bulk
assembly functions.
--

Signed-off-by: Jussi Kivilinna <jussi.kivilinna@iki.fi>

2 files changed, 75 insertions, 15 deletions

diff --git a/cipher/des-amd64.S b/cipher/des-amd64.S
index e8b2c568..307d2112 100644
--- a/cipher/des-amd64.S
+++ b/cipher/des-amd64.S
@@ -20,7 +20,8 @@
 
 #ifdef __x86_64
 #include <config.h>
-#if defined(USE_DES) && defined(HAVE_COMPATIBLE_GCC_AMD64_PLATFORM_AS)
+#if defined(USE_DES) && (defined(HAVE_COMPATIBLE_GCC_AMD64_PLATFORM_AS) || \
+    defined(HAVE_COMPATIBLE_GCC_WIN64_PLATFORM_AS))
 
 #ifdef __PIC__
 #  define RIP (%rip)
@@ -28,6 +29,12 @@
 #  define RIP
 #endif
 
+#ifdef HAVE_COMPATIBLE_GCC_AMD64_PLATFORM_AS
+# define ELF(...) __VA_ARGS__
+#else
+# define ELF(...) /*_*/
+#endif
+
 .text
 
 #define s1 0
@@ -185,7 +192,7 @@
 
 .align 8
 .globl _gcry_3des_amd64_crypt_block
-.type  _gcry_3des_amd64_crypt_block,@function;
+ELF(.type  _gcry_3des_amd64_crypt_block,@function;)
 
 _gcry_3des_amd64_crypt_block:
 	/* input:
@@ -271,7 +278,7 @@ _gcry_3des_amd64_crypt_block:
 	popq %rbp;
 
 	ret;
-.size _gcry_3des_amd64_crypt_block,.-_gcry_3des_amd64_crypt_block;
+ELF(.size _gcry_3des_amd64_crypt_block,.-_gcry_3des_amd64_crypt_block;)
 
 /***********************************************************************
  * 3-way 3DES
@@ -458,7 +465,7 @@ _gcry_3des_amd64_crypt_block:
 	movl   right##d, 4(io);
 
 .align 8
-.type  _gcry_3des_amd64_crypt_blk3,@function;
+ELF(.type  _gcry_3des_amd64_crypt_blk3,@function;)
 _gcry_3des_amd64_crypt_blk3:
 	/* input:
 	 *  %rdi: round keys, CTX
@@ -528,11 +535,11 @@ _gcry_3des_amd64_crypt_blk3:
 	final_permutation3(RR, RL);
 
 	ret;
-.size _gcry_3des_amd64_crypt_blk3,.-_gcry_3des_amd64_crypt_blk3;
+ELF(.size _gcry_3des_amd64_crypt_blk3,.-_gcry_3des_amd64_crypt_blk3;)
 
 .align 8
 .globl  _gcry_3des_amd64_cbc_dec
-.type   _gcry_3des_amd64_cbc_dec,@function;
+ELF(.type   _gcry_3des_amd64_cbc_dec,@function;)
 _gcry_3des_amd64_cbc_dec:
 	/* input:
 	 *	%rdi: ctx, CTX
@@ -604,11 +611,11 @@ _gcry_3des_amd64_cbc_dec:
 	popq %rbp;
 
 	ret;
-.size _gcry_3des_amd64_cbc_dec,.-_gcry_3des_amd64_cbc_dec;
+ELF(.size _gcry_3des_amd64_cbc_dec,.-_gcry_3des_amd64_cbc_dec;)
 
 .align 8
 .globl  _gcry_3des_amd64_ctr_enc
-.type   _gcry_3des_amd64_ctr_enc,@function;
+ELF(.type   _gcry_3des_amd64_ctr_enc,@function;)
 _gcry_3des_amd64_ctr_enc:
 	/* input:
 	 *	%rdi: ctx, CTX
@@ -682,11 +689,11 @@ _gcry_3des_amd64_ctr_enc:
 	popq %rbp;
 
 	ret;
-.size _gcry_3des_amd64_cbc_dec,.-_gcry_3des_amd64_cbc_dec;
+ELF(.size _gcry_3des_amd64_cbc_dec,.-_gcry_3des_amd64_cbc_dec;)
 
 .align 8
 .globl  _gcry_3des_amd64_cfb_dec
-.type   _gcry_3des_amd64_cfb_dec,@function;
+ELF(.type   _gcry_3des_amd64_cfb_dec,@function;)
 _gcry_3des_amd64_cfb_dec:
 	/* input:
 	 *	%rdi: ctx, CTX
@@ -757,7 +764,7 @@ _gcry_3des_amd64_cfb_dec:
 	popq %rbx;
 	popq %rbp;
 	ret;
-.size _gcry_3des_amd64_cfb_dec,.-_gcry_3des_amd64_cfb_dec;
+ELF(.size _gcry_3des_amd64_cfb_dec,.-_gcry_3des_amd64_cfb_dec;)
 
 .data
 .align 16
diff --git a/cipher/des.c b/cipher/des.c
index d4863d1b..be62763e 100644
--- a/cipher/des.c
+++ b/cipher/des.c
@@ -127,7 +127,8 @@
 
 /* USE_AMD64_ASM indicates whether to use AMD64 assembly code. */
 #undef USE_AMD64_ASM
-#if defined(__x86_64__) && defined(HAVE_COMPATIBLE_GCC_AMD64_PLATFORM_AS)
+#if defined(__x86_64__) && (defined(HAVE_COMPATIBLE_GCC_AMD64_PLATFORM_AS) || \
+    defined(HAVE_COMPATIBLE_GCC_WIN64_PLATFORM_AS))
 # define USE_AMD64_ASM 1
 #endif
 
@@ -771,6 +772,24 @@ extern void _gcry_3des_amd64_cfb_dec(const void *keys, byte *out,
 
 #define TRIPLEDES_ECB_BURN_STACK (8 * sizeof(void *))
 
+#ifdef HAVE_COMPATIBLE_GCC_WIN64_PLATFORM_AS
+static inline void
+call_sysv_fn (const void *fn, const void *arg1, const void *arg2,
+              const void *arg3, const void *arg4)
+{
+  /* Call SystemV ABI function without storing non-volatile XMM registers,
+   * as target function does not use vector instruction sets. */
+  asm volatile ("callq *%0\n\t"
+                : "+a" (fn),
+                  "+D" (arg1),
+                  "+S" (arg2),
+                  "+d" (arg3),
+                  "+c" (arg4)
+                :
+                : "cc", "memory", "r8", "r9", "r10", "r11");
+}
+#endif
+
 /*
  * Electronic Codebook Mode Triple-DES encryption/decryption of data
  * according to 'mode'.  Sometimes this mode is named 'EDE' mode
@@ -784,11 +803,45 @@ tripledes_ecb_crypt (struct _tripledes_ctx *ctx, const byte * from,
 
   keys = mode ? ctx->decrypt_subkeys : ctx->encrypt_subkeys;
 
+#ifdef HAVE_COMPATIBLE_GCC_WIN64_PLATFORM_AS
+  call_sysv_fn (_gcry_3des_amd64_crypt_block, keys, to, from, NULL);
+#else
   _gcry_3des_amd64_crypt_block(keys, to, from);
+#endif
 
   return 0;
 }
 
+static inline void
+tripledes_amd64_ctr_enc(const void *keys, byte *out, const byte *in, byte *ctr)
+{
+#ifdef HAVE_COMPATIBLE_GCC_WIN64_PLATFORM_AS
+  call_sysv_fn (_gcry_3des_amd64_ctr_enc, keys, out, in, ctr);
+#else
+  _gcry_3des_amd64_ctr_enc(keys, out, in, ctr);
+#endif
+}
+
+static inline void
+tripledes_amd64_cbc_dec(const void *keys, byte *out, const byte *in, byte *iv)
+{
+#ifdef HAVE_COMPATIBLE_GCC_WIN64_PLATFORM_AS
+  call_sysv_fn (_gcry_3des_amd64_cbc_dec, keys, out, in, iv);
+#else
+  _gcry_3des_amd64_cbc_dec(keys, out, in, iv);
+#endif
+}
+
+static inline void
+tripledes_amd64_cfb_dec(const void *keys, byte *out, const byte *in, byte *iv)
+{
+#ifdef HAVE_COMPATIBLE_GCC_WIN64_PLATFORM_AS
+  call_sysv_fn (_gcry_3des_amd64_cfb_dec, keys, out, in, iv);
+#else
+  _gcry_3des_amd64_cfb_dec(keys, out, in, iv);
+#endif
+}
+
 #else /*USE_AMD64_ASM*/
 
 #define TRIPLEDES_ECB_BURN_STACK 32
@@ -871,7 +924,7 @@ _gcry_3des_ctr_enc(void *context, unsigned char *ctr, void *outbuf_arg,
     /* Process data in 3 block chunks. */
     while (nblocks >= 3)
       {
-        _gcry_3des_amd64_ctr_enc(ctx->encrypt_subkeys, outbuf, inbuf, ctr);
+        tripledes_amd64_ctr_enc(ctx->encrypt_subkeys, outbuf, inbuf, ctr);
 
         nblocks -= 3;
         outbuf += 3 * DES_BLOCKSIZE;
@@ -926,7 +979,7 @@ _gcry_3des_cbc_dec(void *context, unsigned char *iv, void *outbuf_arg,
     /* Process data in 3 block chunks. */
     while (nblocks >= 3)
       {
-        _gcry_3des_amd64_cbc_dec(ctx->decrypt_subkeys, outbuf, inbuf, iv);
+        tripledes_amd64_cbc_dec(ctx->decrypt_subkeys, outbuf, inbuf, iv);
 
         nblocks -= 3;
         outbuf += 3 * DES_BLOCKSIZE;
@@ -974,7 +1027,7 @@ _gcry_3des_cfb_dec(void *context, unsigned char *iv, void *outbuf_arg,
     /* Process data in 3 block chunks. */
     while (nblocks >= 3)
       {
-        _gcry_3des_amd64_cfb_dec(ctx->encrypt_subkeys, outbuf, inbuf, iv);
+        tripledes_amd64_cfb_dec(ctx->encrypt_subkeys, outbuf, inbuf, iv);
 
         nblocks -= 3;
         outbuf += 3 * DES_BLOCKSIZE;