[PATCH 2/8] KVM: x86 emulator: decode extended accumulator explicity
Single-operand MUL and DIV access an extended accumulator: AX for byte instructions, and DX:AX, EDX:EAX, or RDX:RAX for larger-sized instructions. Add support for fetching the extended accumulator. In order not to change things too much, RDX is loaded into Src2, which is already loaded by fastop(). This avoids increasing register pressure on i386. Signed-off-by: Avi Kivity --- arch/x86/kvm/emulate.c | 22 ++ 1 file changed, 22 insertions(+) diff --git a/arch/x86/kvm/emulate.c b/arch/x86/kvm/emulate.c index 18c86b5..aa8516e 100644 --- a/arch/x86/kvm/emulate.c +++ b/arch/x86/kvm/emulate.c @@ -60,6 +60,8 @@ #define OpGS 25ull /* GS */ #define OpMem826ull /* 8-bit zero extended memory operand */ #define OpImm64 27ull /* Sign extended 16/32/64-bit immediate */ +#define OpAccLo 29ull /* Low part of extended acc (AX/AX/EAX/RAX) */ +#define OpAccHi 30ull /* High part of extended acc (-/DX/EDX/RDX) */ #define OpBits 5 /* Width of operand field */ #define OpMask ((1ull << OpBits) - 1) @@ -85,6 +87,7 @@ #define DstMem64(OpMem64 << DstShift) #define DstImmUByte (OpImmUByte << DstShift) #define DstDX (OpDX << DstShift) +#define DstAccLo(OpAccLo << DstShift) #define DstMask (OpMask << DstShift) /* Source operand type. */ #define SrcShift6 @@ -106,6 +109,7 @@ #define SrcImm64(OpImm64 << SrcShift) #define SrcDX (OpDX << SrcShift) #define SrcMem8 (OpMem8 << SrcShift) +#define SrcAccHi(OpAccHi << SrcShift) #define SrcMask (OpMask << SrcShift) #define BitOp (1<<11) #define MemAbs (1<<12) /* Memory operand is absolute displacement */ @@ -154,6 +158,8 @@ #define NoWrite ((u64)1 << 45) /* No writeback */ #define SrcWrite((u64)1 << 46) /* Write back src operand */ +#define DstXacc (DstAccLo | SrcAccHi | SrcWrite) + #define X2(x...) x, x #define X3(x...) X2(x), x #define X4(x...) X2(x), X2(x) @@ -4129,6 +4135,22 @@ static int decode_operand(struct x86_emulate_ctxt *ctxt, struct operand *op, fetch_register_operand(op); op->orig_val = op->val; break; + case OpAccLo: + op->type = OP_REG; + op->bytes = (ctxt->d & ByteOp) ? 2 : ctxt->op_bytes; + op->addr.reg = reg_rmw(ctxt, VCPU_REGS_RAX); + fetch_register_operand(op); + op->orig_val = op->val; + break; + case OpAccHi: + if (ctxt->d & ByteOp) + break; + op->type = OP_REG; + op->bytes = ctxt->op_bytes; + op->addr.reg = reg_rmw(ctxt, VCPU_REGS_RDX); + fetch_register_operand(op); + op->orig_val = op->val; + break; case OpDI: op->type = OP_MEM; op->bytes = (ctxt->d & ByteOp) ? 1 : ctxt->op_bytes; -- 1.8.1.2 -- To unsubscribe from this list: send the line "unsubscribe kvm" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [PATCH 2/8] KVM: x86 emulator: decode extended accumulator explicity
On Sat, Feb 09, 2013 at 11:31:45AM +0200, Avi Kivity wrote: > Single-operand MUL and DIV access an extended accumulator: AX for byte > instructions, and DX:AX, EDX:EAX, or RDX:RAX for larger-sized instructions. > Add support for fetching the extended accumulator. > Where this "extended accumulator" definition is coming from? Spec just says that two registers are used for a result depending on the operand size. > In order not to change things too much, RDX is loaded into Src2, which is > already loaded by fastop(). This avoids increasing register pressure on > i386. > > Signed-off-by: Avi Kivity > --- > arch/x86/kvm/emulate.c | 22 ++ > 1 file changed, 22 insertions(+) > > diff --git a/arch/x86/kvm/emulate.c b/arch/x86/kvm/emulate.c > index 18c86b5..aa8516e 100644 > --- a/arch/x86/kvm/emulate.c > +++ b/arch/x86/kvm/emulate.c > @@ -60,6 +60,8 @@ > #define OpGS 25ull /* GS */ > #define OpMem826ull /* 8-bit zero extended memory operand */ > #define OpImm64 27ull /* Sign extended 16/32/64-bit immediate */ > +#define OpAccLo 29ull /* Low part of extended acc (AX/AX/EAX/RAX) > */ > +#define OpAccHi 30ull /* High part of extended acc (-/DX/EDX/RDX) > */ > > #define OpBits 5 /* Width of operand field */ > #define OpMask ((1ull << OpBits) - 1) > @@ -85,6 +87,7 @@ > #define DstMem64(OpMem64 << DstShift) > #define DstImmUByte (OpImmUByte << DstShift) > #define DstDX (OpDX << DstShift) > +#define DstAccLo(OpAccLo << DstShift) > #define DstMask (OpMask << DstShift) > /* Source operand type. */ > #define SrcShift6 > @@ -106,6 +109,7 @@ > #define SrcImm64(OpImm64 << SrcShift) > #define SrcDX (OpDX << SrcShift) > #define SrcMem8 (OpMem8 << SrcShift) > +#define SrcAccHi(OpAccHi << SrcShift) > #define SrcMask (OpMask << SrcShift) > #define BitOp (1<<11) > #define MemAbs (1<<12) /* Memory operand is absolute displacement > */ > @@ -154,6 +158,8 @@ > #define NoWrite ((u64)1 << 45) /* No writeback */ > #define SrcWrite((u64)1 << 46) /* Write back src operand */ > > +#define DstXacc (DstAccLo | SrcAccHi | SrcWrite) > + > #define X2(x...) x, x > #define X3(x...) X2(x), x > #define X4(x...) X2(x), X2(x) > @@ -4129,6 +4135,22 @@ static int decode_operand(struct x86_emulate_ctxt > *ctxt, struct operand *op, > fetch_register_operand(op); > op->orig_val = op->val; > break; > + case OpAccLo: > + op->type = OP_REG; > + op->bytes = (ctxt->d & ByteOp) ? 2 : ctxt->op_bytes; > + op->addr.reg = reg_rmw(ctxt, VCPU_REGS_RAX); > + fetch_register_operand(op); > + op->orig_val = op->val; > + break; > + case OpAccHi: > + if (ctxt->d & ByteOp) > + break; Why would we set OpAccHi if ByteOp is set in decode tables in the first place? > + op->type = OP_REG; > + op->bytes = ctxt->op_bytes; > + op->addr.reg = reg_rmw(ctxt, VCPU_REGS_RDX); > + fetch_register_operand(op); > + op->orig_val = op->val; > + break; > case OpDI: > op->type = OP_MEM; > op->bytes = (ctxt->d & ByteOp) ? 1 : ctxt->op_bytes; > -- > 1.8.1.2 -- Gleb. -- To unsubscribe from this list: send the line "unsubscribe kvm" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [PATCH 2/8] KVM: x86 emulator: decode extended accumulator explicity
On Sun, Feb 10, 2013 at 02:19:29PM +0200, Gleb Natapov wrote: > On Sat, Feb 09, 2013 at 11:31:45AM +0200, Avi Kivity wrote: > > Single-operand MUL and DIV access an extended accumulator: AX for byte > > instructions, and DX:AX, EDX:EAX, or RDX:RAX for larger-sized instructions. > > Add support for fetching the extended accumulator. > > > Where this "extended accumulator" definition is coming from? Spec just > says that two registers are used for a result depending on the operand > size. > > > In order not to change things too much, RDX is loaded into Src2, which is > > already loaded by fastop(). This avoids increasing register pressure on > > i386. > > > > Signed-off-by: Avi Kivity > > --- > > arch/x86/kvm/emulate.c | 22 ++ > > 1 file changed, 22 insertions(+) > > > > diff --git a/arch/x86/kvm/emulate.c b/arch/x86/kvm/emulate.c > > index 18c86b5..aa8516e 100644 > > --- a/arch/x86/kvm/emulate.c > > +++ b/arch/x86/kvm/emulate.c > > @@ -60,6 +60,8 @@ > > #define OpGS 25ull /* GS */ > > #define OpMem826ull /* 8-bit zero extended memory operand */ > > #define OpImm64 27ull /* Sign extended 16/32/64-bit immediate */ > > +#define OpAccLo 29ull /* Low part of extended acc > > (AX/AX/EAX/RAX) */ > > +#define OpAccHi 30ull /* High part of extended acc > > (-/DX/EDX/RDX) */ > > > > #define OpBits 5 /* Width of operand field */ > > #define OpMask ((1ull << OpBits) - 1) > > @@ -85,6 +87,7 @@ > > #define DstMem64(OpMem64 << DstShift) > > #define DstImmUByte (OpImmUByte << DstShift) > > #define DstDX (OpDX << DstShift) > > +#define DstAccLo(OpAccLo << DstShift) > > #define DstMask (OpMask << DstShift) > > /* Source operand type. */ > > #define SrcShift6 > > @@ -106,6 +109,7 @@ > > #define SrcImm64(OpImm64 << SrcShift) > > #define SrcDX (OpDX << SrcShift) > > #define SrcMem8 (OpMem8 << SrcShift) > > +#define SrcAccHi(OpAccHi << SrcShift) > > #define SrcMask (OpMask << SrcShift) > > #define BitOp (1<<11) > > #define MemAbs (1<<12) /* Memory operand is absolute > > displacement */ > > @@ -154,6 +158,8 @@ > > #define NoWrite ((u64)1 << 45) /* No writeback */ > > #define SrcWrite((u64)1 << 46) /* Write back src operand */ > > > > +#define DstXacc (DstAccLo | SrcAccHi | SrcWrite) > > + > > #define X2(x...) x, x > > #define X3(x...) X2(x), x > > #define X4(x...) X2(x), X2(x) > > @@ -4129,6 +4135,22 @@ static int decode_operand(struct x86_emulate_ctxt > > *ctxt, struct operand *op, > > fetch_register_operand(op); > > op->orig_val = op->val; > > break; > > + case OpAccLo: > > + op->type = OP_REG; > > + op->bytes = (ctxt->d & ByteOp) ? 2 : ctxt->op_bytes; > > + op->addr.reg = reg_rmw(ctxt, VCPU_REGS_RAX); > > + fetch_register_operand(op); > > + op->orig_val = op->val; > > + break; > > + case OpAccHi: > > + if (ctxt->d & ByteOp) > > + break; > Why would we set OpAccHi if ByteOp is set in decode tables in the first > place? > Was trying to create decode tables so that this one will not be necessary, but it will require to have two group3 arrays, one for ByteOp and another for non ByteOp. Not nice. But what I noticed is that current code writes into NULL pointer when ByteOp mul/div is emulated. It happens because src operand stays undecoded, but since SrcWrite is set for the instruction src operand is written back anyway. Since it is undecoded op->type is zero and op->addr.reg is NULL and type zero happens to be OP_REG, so write_register_operand() is called with op->addr.reg == NULL. We should do op->type = OP_NONE if !ByteOp here. Will fix and apply, try to come up with a way to express it via decode tables on top. > > + op->type = OP_REG; > > + op->bytes = ctxt->op_bytes; > > + op->addr.reg = reg_rmw(ctxt, VCPU_REGS_RDX); > > + fetch_register_operand(op); > > + op->orig_val = op->val; > > + break; > > case OpDI: > > op->type = OP_MEM; > > op->bytes = (ctxt->d & ByteOp) ? 1 : ctxt->op_bytes; > > -- > > 1.8.1.2 > > -- > Gleb. > -- > To unsubscribe from this list: send the line "unsubscribe kvm" in > the body of a message to majord...@vger.kernel.org > More majordomo info at http://vger.kernel.org/majordomo-info.html -- Gleb. -- To unsubscribe from this list: send the line "unsubscribe kvm" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html