[PATCH v3 08/16] x86/kvm/emulate: Introduce EM_ASM_3WCL

[Peter Zijlstra]

Replace the FASTOP3WCL instructions.

Signed-off-by: Peter Zijlstra (Intel) <pet...@infradead.org>
---
 arch/x86/kvm/emulate.c |   25 +++++++++++++++++++------
 1 file changed, 19 insertions(+), 6 deletions(-)

--- a/arch/x86/kvm/emulate.c
+++ b/arch/x86/kvm/emulate.c
@@ -302,6 +302,9 @@ static int em_##op(struct x86_emulate_ct
 #define __EM_ASM_2(op, dst, src) \
                __EM_ASM(#op " %%" #src ", %%" #dst " \n\t")
 
+#define __EM_ASM_3(op, dst, src, src2) \
+               __EM_ASM(#op " %%" #src2 ", %%" #src ", %%" #dst " \n\t")
+
 #define EM_ASM_END \
        } \
        ctxt->eflags = (ctxt->eflags & ~EFLAGS_MASK) | (flags & EFLAGS_MASK); \
@@ -371,6 +374,16 @@ static int em_##op(struct x86_emulate_ct
        ON64(case 8: __EM_ASM_2(op##q, rax, cl); break;) \
        EM_ASM_END
 
+/* 3-operand, using "a" (dst), "d" (src) and CL (src2) */
+#define EM_ASM_3WCL(op) \
+       EM_ASM_START(op) \
+       case 1: break; \
+       case 2: __EM_ASM_3(op##w, ax, dx, cl); break; \
+       case 4: __EM_ASM_3(op##l, eax, edx, cl); break; \
+       ON64(case 8: __EM_ASM_3(op##q, rax, rdx, cl); break;) \
+       EM_ASM_END
+
+
 /*
  * fastop functions have a special calling convention:
  *
@@ -1097,8 +1110,8 @@ EM_ASM_1SRC2(imul, imul_ex);
 EM_ASM_1SRC2EX(div, div_ex);
 EM_ASM_1SRC2EX(idiv, idiv_ex);
 
-FASTOP3WCL(shld);
-FASTOP3WCL(shrd);
+EM_ASM_3WCL(shld);
+EM_ASM_3WCL(shrd);
 
 EM_ASM_2W(imul);
 
@@ -4496,14 +4509,14 @@ static const struct opcode twobyte_table
        I(Stack | Src2FS, em_push_sreg), I(Stack | Src2FS, em_pop_sreg),
        II(ImplicitOps, em_cpuid, cpuid),
        I(DstMem | SrcReg | ModRM | BitOp | NoWrite, em_bt),
-       F(DstMem | SrcReg | Src2ImmByte | ModRM, em_shld),
-       F(DstMem | SrcReg | Src2CL | ModRM, em_shld), N, N,
+       I(DstMem | SrcReg | Src2ImmByte | ModRM, em_shld),
+       I(DstMem | SrcReg | Src2CL | ModRM, em_shld), N, N,
        /* 0xA8 - 0xAF */
        I(Stack | Src2GS, em_push_sreg), I(Stack | Src2GS, em_pop_sreg),
        II(EmulateOnUD | ImplicitOps, em_rsm, rsm),
        I(DstMem | SrcReg | ModRM | BitOp | Lock | PageTable, em_bts),
-       F(DstMem | SrcReg | Src2ImmByte | ModRM, em_shrd),
-       F(DstMem | SrcReg | Src2CL | ModRM, em_shrd),
+       I(DstMem | SrcReg | Src2ImmByte | ModRM, em_shrd),
+       I(DstMem | SrcReg | Src2CL | ModRM, em_shrd),
        GD(0, &group15), I(DstReg | SrcMem | ModRM, em_imul),
        /* 0xB0 - 0xB7 */
        I2bv(DstMem | SrcReg | ModRM | Lock | PageTable | SrcWrite, em_cmpxchg),