rdrand: Fix the RDRAND data reduction

[thirdparty/rng-tools.git] / rdrand_asm.S

diff --git a/rdrand_asm.S b/rdrand_asm.S
index 913b923b445b4efb16ed8bd8bd2771629ef43497..c92996c17dfe7f5b5c0b8fd7e8cdae686701dc56 100644 (file)
--- a/rdrand_asm.S
+++ b/rdrand_asm.S
@@ -58,6 +58,7 @@ ENDPROC(x86_rdrand_bytes)
 #define PTR0   %rdi
 #define PTR1   %rsi
 #define PTR2   %rcx
+#define CTR3   %eax
 #define NPTR2  1       /* %rcx = %r1, only 0-7 valid here */
 
 #elif defined(__i386__)
@@ -97,6 +98,7 @@ ENDPROC(x86_rdrand_bytes)
 #define PTR0   %eax
 #define PTR1   %edx
 #define PTR2   %ecx
+#define CTR3   %esi
 #define NPTR2  1       /* %rcx = %r1 */
 
 #endif
@@ -107,27 +109,37 @@ ENTRY(x86_aes_mangle)
        mov     %esp, %ebp
        movl    8(%ebp), %eax
        movl    12(%ebp), %edx
+       push    %esi
 #endif
+       movl    $512, CTR3      /* Number of rounds */
+       
+       movdqa  (0*16)(PTR1), %xmm0
+       movdqa  (1*16)(PTR1), %xmm1
+       movdqa  (2*16)(PTR1), %xmm2
+       movdqa  (3*16)(PTR1), %xmm3
+       movdqa  (4*16)(PTR1), %xmm4
+       movdqa  (5*16)(PTR1), %xmm5
+       movdqa  (6*16)(PTR1), %xmm6
+       movdqa  (7*16)(PTR1), %xmm7
 
+#ifdef __x86_64__
        SETPTR(aes_round_keys, PTR2)
+1:
+#else
+1:
+       SETPTR(aes_round_keys, PTR2)
+#endif
 
-       movdqa  (0*16)(PTR0), %xmm0
-       movdqa  (1*16)(PTR0), %xmm1
-       movdqa  (2*16)(PTR0), %xmm2
-       movdqa  (3*16)(PTR0), %xmm3
-       movdqa  (4*16)(PTR0), %xmm4
-       movdqa  (5*16)(PTR0), %xmm5
-       movdqa  (6*16)(PTR0), %xmm6
-       movdqa  (7*16)(PTR0), %xmm7
-
-       pxor    (0*16)(PTR1), %xmm0
-       pxor    (1*16)(PTR1), %xmm1
-       pxor    (2*16)(PTR1), %xmm2
-       pxor    (3*16)(PTR1), %xmm3
-       pxor    (4*16)(PTR1), %xmm4
-       pxor    (5*16)(PTR1), %xmm5
-       pxor    (6*16)(PTR1), %xmm6
-       pxor    (7*16)(PTR1), %xmm7
+       /* 8192 = 512 (rounds) * 16 (bytes) */
+       pxor    (0*8192)(PTR0), %xmm0
+       pxor    (1*8192)(PTR0), %xmm1
+       pxor    (2*8192)(PTR0), %xmm2
+       pxor    (3*8192)(PTR0), %xmm3
+       pxor    (4*8192)(PTR0), %xmm4
+       pxor    (5*8192)(PTR0), %xmm5
+       pxor    (6*8192)(PTR0), %xmm6
+       pxor    (7*8192)(PTR0), %xmm7
+       add     $16, PTR0
 
 offset = 0
        .rept 10
@@ -175,16 +187,9 @@ offset = offset + 16
        .byte   0x66,0x0f,0x38,0xdd,0x30+NPTR2  /* aesenclast (PTR2), %xmm6 */
        .byte   0x66,0x0f,0x38,0xdd,0x38+NPTR2  /* aesenclast (PTR2), %xmm7 */
 #endif
-
-       movdqa  %xmm0, (0*16)(PTR0)
-       movdqa  %xmm1, (1*16)(PTR0)
-       movdqa  %xmm2, (2*16)(PTR0)
-       movdqa  %xmm3, (3*16)(PTR0)
-       movdqa  %xmm4, (4*16)(PTR0)
-       movdqa  %xmm5, (5*16)(PTR0)
-       movdqa  %xmm6, (6*16)(PTR0)
-       movdqa  %xmm7, (7*16)(PTR0)
-
+       sub     $1, CTR3
+       jnz     1b
+       
        movdqa  %xmm0, (0*16)(PTR1)
        movdqa  %xmm1, (1*16)(PTR1)
        movdqa  %xmm2, (2*16)(PTR1)
@@ -195,6 +200,7 @@ offset = offset + 16
        movdqa  %xmm7, (7*16)(PTR1)
 
 #ifdef __i386__
+       pop     %esi
        pop     %ebp
 #endif
        ret