[PATCH 2/5] KVM: x86 emulator: Use single stage decoding for PUSH/POP XS instructions

[Takuya Yoshikawa <takuya.yoshikawa@gmail.com>]

From: Takuya Yoshikawa <yoshikawa.takuya@oss.ntt.co.jp>

PUSH ES/CS/SS/DS/FS/GS and POP ES/SS/DS/FS/GS are converted.

Signed-off-by: Takuya Yoshikawa <yoshikawa.takuya@oss.ntt.co.jp>
---
 arch/x86/kvm/emulate.c |  111 +++++++++++++++++++++++++++++++-----------------
 1 files changed, 72 insertions(+), 39 deletions(-)

diff --git a/arch/x86/kvm/emulate.c b/arch/x86/kvm/emulate.c
index bd9572a..fcc49ef 100644
--- a/arch/x86/kvm/emulate.c
+++ b/arch/x86/kvm/emulate.c
@@ -2188,6 +2188,67 @@ static int em_push(struct x86_emulate_ctxt *ctxt)
 	return X86EMUL_CONTINUE;
 }
 
+static int em_push_es(struct x86_emulate_ctxt *ctxt)
+{
+	emulate_push_sreg(ctxt, ctxt->ops, VCPU_SREG_ES);
+	return X86EMUL_CONTINUE;
+}
+
+static int em_push_cs(struct x86_emulate_ctxt *ctxt)
+{
+	emulate_push_sreg(ctxt, ctxt->ops, VCPU_SREG_CS);
+	return X86EMUL_CONTINUE;
+}
+
+static int em_push_ss(struct x86_emulate_ctxt *ctxt)
+{
+	emulate_push_sreg(ctxt, ctxt->ops, VCPU_SREG_SS);
+	return X86EMUL_CONTINUE;
+}
+
+static int em_push_ds(struct x86_emulate_ctxt *ctxt)
+{
+	emulate_push_sreg(ctxt, ctxt->ops, VCPU_SREG_DS);
+	return X86EMUL_CONTINUE;
+}
+
+static int em_push_fs(struct x86_emulate_ctxt *ctxt)
+{
+	emulate_push_sreg(ctxt, ctxt->ops, VCPU_SREG_FS);
+	return X86EMUL_CONTINUE;
+}
+
+static int em_push_gs(struct x86_emulate_ctxt *ctxt)
+{
+	emulate_push_sreg(ctxt, ctxt->ops, VCPU_SREG_GS);
+	return X86EMUL_CONTINUE;
+}
+
+static int em_pop_es(struct x86_emulate_ctxt *ctxt)
+{
+	return emulate_pop_sreg(ctxt, ctxt->ops, VCPU_SREG_ES);
+}
+
+static int em_pop_ss(struct x86_emulate_ctxt *ctxt)
+{
+	return emulate_pop_sreg(ctxt, ctxt->ops, VCPU_SREG_SS);
+}
+
+static int em_pop_ds(struct x86_emulate_ctxt *ctxt)
+{
+	return emulate_pop_sreg(ctxt, ctxt->ops, VCPU_SREG_DS);
+}
+
+static int em_pop_fs(struct x86_emulate_ctxt *ctxt)
+{
+	return emulate_pop_sreg(ctxt, ctxt->ops, VCPU_SREG_FS);
+}
+
+static int em_pop_gs(struct x86_emulate_ctxt *ctxt)
+{
+	return emulate_pop_sreg(ctxt, ctxt->ops, VCPU_SREG_GS);
+}
+
 static int em_das(struct x86_emulate_ctxt *ctxt)
 {
 	struct decode_cache *c = &ctxt->decode;
@@ -2471,16 +2532,19 @@ static struct opcode group11[] = {
 static struct opcode opcode_table[256] = {
 	/* 0x00 - 0x07 */
 	I6ALU(Lock, em_add),
-	D(ImplicitOps | Stack | No64), D(ImplicitOps | Stack | No64),
+	I(ImplicitOps | Stack | No64, em_push_es),
+	I(ImplicitOps | Stack | No64, em_pop_es),
 	/* 0x08 - 0x0F */
 	I6ALU(Lock, em_or),
-	D(ImplicitOps | Stack | No64), N,
+	I(ImplicitOps | Stack | No64, em_push_cs), N,
 	/* 0x10 - 0x17 */
 	I6ALU(Lock, em_adc),
-	D(ImplicitOps | Stack | No64), D(ImplicitOps | Stack | No64),
+	I(ImplicitOps | Stack | No64, em_push_ss),
+	I(ImplicitOps | Stack | No64, em_pop_ss),
 	/* 0x18 - 0x1F */
 	I6ALU(Lock, em_sbb),
-	D(ImplicitOps | Stack | No64), D(ImplicitOps | Stack | No64),
+	I(ImplicitOps | Stack | No64, em_push_ds),
+	I(ImplicitOps | Stack | No64, em_pop_ds),
 	/* 0x20 - 0x27 */
 	I6ALU(Lock, em_and), N, N,
 	/* 0x28 - 0x2F */
@@ -2600,12 +2664,14 @@ static struct opcode twobyte_table[256] = {
 	/* 0x90 - 0x9F */
 	X16(D(ByteOp | DstMem | SrcNone | ModRM| Mov)),
 	/* 0xA0 - 0xA7 */
-	D(ImplicitOps | Stack), D(ImplicitOps | Stack),
+	I(ImplicitOps | Stack, em_push_fs),
+	I(ImplicitOps | Stack, em_pop_fs),
 	N, D(DstMem | SrcReg | ModRM | BitOp),
 	D(DstMem | SrcReg | Src2ImmByte | ModRM),
 	D(DstMem | SrcReg | Src2CL | ModRM), N, N,
 	/* 0xA8 - 0xAF */
-	D(ImplicitOps | Stack), D(ImplicitOps | Stack),
+	I(ImplicitOps | Stack, em_push_gs),
+	I(ImplicitOps | Stack, em_pop_gs),
 	N, D(DstMem | SrcReg | ModRM | BitOp | Lock),
 	D(DstMem | SrcReg | Src2ImmByte | ModRM),
 	D(DstMem | SrcReg | Src2CL | ModRM),
@@ -3109,27 +3175,6 @@ special_insn:
 		goto twobyte_insn;
 
 	switch (c->b) {
-	case 0x06:		/* push es */
-		emulate_push_sreg(ctxt, ops, VCPU_SREG_ES);
-		break;
-	case 0x07:		/* pop es */
-		rc = emulate_pop_sreg(ctxt, ops, VCPU_SREG_ES);
-		break;
-	case 0x0e:		/* push cs */
-		emulate_push_sreg(ctxt, ops, VCPU_SREG_CS);
-		break;
-	case 0x16:		/* push ss */
-		emulate_push_sreg(ctxt, ops, VCPU_SREG_SS);
-		break;
-	case 0x17:		/* pop ss */
-		rc = emulate_pop_sreg(ctxt, ops, VCPU_SREG_SS);
-		break;
-	case 0x1e:		/* push ds */
-		emulate_push_sreg(ctxt, ops, VCPU_SREG_DS);
-		break;
-	case 0x1f:		/* pop ds */
-		rc = emulate_pop_sreg(ctxt, ops, VCPU_SREG_DS);
-		break;
 	case 0x40 ... 0x47: /* inc r16/r32 */
 		emulate_1op("inc", c->dst, ctxt->eflags);
 		break;
@@ -3627,12 +3672,6 @@ twobyte_insn:
 	case 0x90 ... 0x9f:     /* setcc r/m8 */
 		c->dst.val = test_cc(c->b, ctxt->eflags);
 		break;
-	case 0xa0:	  /* push fs */
-		emulate_push_sreg(ctxt, ops, VCPU_SREG_FS);
-		break;
-	case 0xa1:	 /* pop fs */
-		rc = emulate_pop_sreg(ctxt, ops, VCPU_SREG_FS);
-		break;
 	case 0xa3:
 	      bt:		/* bt */
 		c->dst.type = OP_NONE;
@@ -3644,12 +3683,6 @@ twobyte_insn:
 	case 0xa5: /* shld cl, r, r/m */
 		emulate_2op_cl("shld", c->src2, c->src, c->dst, ctxt->eflags);
 		break;
-	case 0xa8:	/* push gs */
-		emulate_push_sreg(ctxt, ops, VCPU_SREG_GS);
-		break;
-	case 0xa9:	/* pop gs */
-		rc = emulate_pop_sreg(ctxt, ops, VCPU_SREG_GS);
-		break;
 	case 0xab:
 	      bts:		/* bts */
 		emulate_2op_SrcV_nobyte("bts", c->src, c->dst, ctxt->eflags);
-- 
1.7.1