3 files changed, 144 insertions, 73 deletions

diff --git a/sys/amd64/include/vmm.h b/sys/amd64/include/vmm.h
index 50d879b..f1902d2 100644
--- a/sys/amd64/include/vmm.h
+++ b/sys/amd64/include/vmm.h
@@ -54,6 +54,7 @@ struct vmspace;
 struct vm_object;
 struct pmap;
 
+enum vm_reg_name;
 enum x2apic_state;
 
 typedef int	(*vmm_init_func_t)(int ipinum);
@@ -236,10 +237,11 @@ int vm_exception_pending(struct vm *vm, int vcpuid, struct vm_exception *vme);
 
 void vm_inject_gp(struct vm *vm, int vcpuid); /* general protection fault */
 void vm_inject_ud(struct vm *vm, int vcpuid); /* undefined instruction fault */
+void vm_inject_pf(struct vm *vm, int vcpuid, int error_code, uint64_t cr2);
 
-#endif	/* KERNEL */
+enum vm_reg_name vm_segment_name(int seg_encoding);
 
-#include <machine/vmm_instruction_emul.h>
+#endif	/* KERNEL */
 
 #define	VM_MAXCPU	16			/* maximum virtual cpus */
 
@@ -280,6 +282,7 @@ enum vm_reg_name {
 	VM_REG_GUEST_IDTR,
 	VM_REG_GUEST_GDTR,
 	VM_REG_GUEST_EFER,
+	VM_REG_GUEST_CR2,
 	VM_REG_LAST
 };
 
@@ -318,6 +321,76 @@ struct seg_desc {
 	uint32_t	limit;
 	uint32_t	access;
 };
+#define	SEG_DESC_TYPE(desc)		((desc)->access & 0x001f)
+#define	SEG_DESC_PRESENT(desc)		((desc)->access & 0x0080)
+#define	SEG_DESC_DEF32(desc)		((desc)->access & 0x4000)
+#define	SEG_DESC_GRANULARITY(desc)	((desc)->access & 0x8000)
+#define	SEG_DESC_UNUSABLE(desc)		((desc)->access & 0x10000)
+
+enum vm_cpu_mode {
+	CPU_MODE_COMPATIBILITY,		/* IA-32E mode (CS.L = 0) */
+	CPU_MODE_64BIT,			/* IA-32E mode (CS.L = 1) */
+};
+
+enum vm_paging_mode {
+	PAGING_MODE_FLAT,
+	PAGING_MODE_32,
+	PAGING_MODE_PAE,
+	PAGING_MODE_64,
+};
+
+struct vm_guest_paging {
+	uint64_t	cr3;
+	int		cpl;
+	enum vm_cpu_mode cpu_mode;
+	enum vm_paging_mode paging_mode;
+};
+
+/*
+ * The data structures 'vie' and 'vie_op' are meant to be opaque to the
+ * consumers of instruction decoding. The only reason why their contents
+ * need to be exposed is because they are part of the 'vm_exit' structure.
+ */
+struct vie_op {
+	uint8_t		op_byte;	/* actual opcode byte */
+	uint8_t		op_type;	/* type of operation (e.g. MOV) */
+	uint16_t	op_flags;
+};
+
+#define	VIE_INST_SIZE	15
+struct vie {
+	uint8_t		inst[VIE_INST_SIZE];	/* instruction bytes */
+	uint8_t		num_valid;		/* size of the instruction */
+	uint8_t		num_processed;
+
+	uint8_t		rex_w:1,		/* REX prefix */
+			rex_r:1,
+			rex_x:1,
+			rex_b:1,
+			rex_present:1;
+
+	uint8_t		mod:2,			/* ModRM byte */
+			reg:4,
+			rm:4;
+
+	uint8_t		ss:2,			/* SIB byte */
+			index:4,
+			base:4;
+
+	uint8_t		disp_bytes;
+	uint8_t		imm_bytes;
+
+	uint8_t		scale;
+	int		base_register;		/* VM_REG_GUEST_xyz */
+	int		index_register;		/* VM_REG_GUEST_xyz */
+
+	int64_t		displacement;		/* optional addr displacement */
+	int64_t		immediate;		/* optional immediate operand */
+
+	uint8_t		decoded;	/* set to 1 if successfully decoded */
+
+	struct vie_op	op;			/* opcode description */
+};
 
 enum vm_exitcode {
 	VM_EXITCODE_INOUT,
@@ -335,22 +408,38 @@ enum vm_exitcode {
 	VM_EXITCODE_RENDEZVOUS,
 	VM_EXITCODE_IOAPIC_EOI,
 	VM_EXITCODE_SUSPENDED,
+	VM_EXITCODE_INOUT_STR,
 	VM_EXITCODE_MAX
 };
 
+struct vm_inout {
+	uint16_t	bytes:3;	/* 1 or 2 or 4 */
+	uint16_t	in:1;
+	uint16_t	string:1;
+	uint16_t	rep:1;
+	uint16_t	port;
+	uint32_t	eax;		/* valid for out */
+};
+
+struct vm_inout_str {
+	struct vm_inout	inout;		/* must be the first element */
+	struct vm_guest_paging paging;
+	uint64_t	rflags;
+	uint64_t	cr0;
+	uint64_t	index;
+	uint64_t	count;		/* rep=1 (%rcx), rep=0 (1) */
+	int		addrsize;
+	enum vm_reg_name seg_name;
+	struct seg_desc seg_desc;
+};
+
 struct vm_exit {
 	enum vm_exitcode	exitcode;
 	int			inst_length;	/* 0 means unknown */
 	uint64_t		rip;
 	union {
-		struct {
-			uint16_t	bytes:3;	/* 1 or 2 or 4 */
-			uint16_t	in:1;		/* out is 0, in is 1 */
-			uint16_t	string:1;
-			uint16_t	rep:1;
-			uint16_t	port;
-			uint32_t	eax;		/* valid for out */
-		} inout;
+		struct vm_inout	inout;
+		struct vm_inout_str inout_str;
 		struct {
 			uint64_t	gpa;
 			int		fault_type;
@@ -358,9 +447,7 @@ struct vm_exit {
 		struct {
 			uint64_t	gpa;
 			uint64_t	gla;
-			uint64_t	cr3;
-			enum vie_cpu_mode cpu_mode;
-			enum vie_paging_mode paging_mode;
+			struct vm_guest_paging paging;
 			struct vie	vie;
 		} inst_emul;
 		/*
diff --git a/sys/amd64/include/vmm_dev.h b/sys/amd64/include/vmm_dev.h
index ecafa9c..f094d51 100644
--- a/sys/amd64/include/vmm_dev.h
+++ b/sys/amd64/include/vmm_dev.h
@@ -168,6 +168,15 @@ struct vm_suspend {
 	enum vm_suspend_how how;
 };
 
+struct vm_gla2gpa {
+	int		vcpuid;		/* inputs */
+	int 		prot;		/* PROT_READ or PROT_WRITE */
+	uint64_t	gla;
+	struct vm_guest_paging paging;
+	int		fault;		/* outputs */
+	uint64_t	gpa;
+};
+
 enum {
 	/* general routines */
 	IOCNUM_ABIVERS = 0,
@@ -180,6 +189,7 @@ enum {
 	IOCNUM_MAP_MEMORY = 10,
 	IOCNUM_GET_MEMORY_SEG = 11,
 	IOCNUM_GET_GPA_PMAP = 12,
+	IOCNUM_GLA2GPA = 13,
 
 	/* register/state accessors */
 	IOCNUM_SET_REGISTER = 20,
@@ -289,4 +299,6 @@ enum {
 	_IOR('v', IOCNUM_GET_HPET_CAPABILITIES, struct vm_hpet_cap)
 #define	VM_GET_GPA_PMAP \
 	_IOWR('v', IOCNUM_GET_GPA_PMAP, struct vm_gpa_pte)
+#define	VM_GLA2GPA	\
+	_IOWR('v', IOCNUM_GLA2GPA, struct vm_gla2gpa)
 #endif
diff --git a/sys/amd64/include/vmm_instruction_emul.h b/sys/amd64/include/vmm_instruction_emul.h
index 0901aa2..e4c408b 100644
--- a/sys/amd64/include/vmm_instruction_emul.h
+++ b/sys/amd64/include/vmm_instruction_emul.h
@@ -29,63 +29,7 @@
 #ifndef	_VMM_INSTRUCTION_EMUL_H_
 #define	_VMM_INSTRUCTION_EMUL_H_
 
-enum vie_cpu_mode {
-	CPU_MODE_COMPATIBILITY,		/* IA-32E mode (CS.L = 0) */
-	CPU_MODE_64BIT,			/* IA-32E mode (CS.L = 1) */
-};
-
-enum vie_paging_mode {
-	PAGING_MODE_FLAT,
-	PAGING_MODE_32,
-	PAGING_MODE_PAE,
-	PAGING_MODE_64,
-};
-
-/*
- * The data structures 'vie' and 'vie_op' are meant to be opaque to the
- * consumers of instruction decoding. The only reason why their contents
- * need to be exposed is because they are part of the 'vm_exit' structure.
- */
-struct vie_op {
-	uint8_t		op_byte;	/* actual opcode byte */
-	uint8_t		op_type;	/* type of operation (e.g. MOV) */
-	uint16_t	op_flags;
-};
-
-#define	VIE_INST_SIZE	15
-struct vie {
-	uint8_t		inst[VIE_INST_SIZE];	/* instruction bytes */
-	uint8_t		num_valid;		/* size of the instruction */
-	uint8_t		num_processed;
-
-	uint8_t		rex_w:1,		/* REX prefix */
-			rex_r:1,
-			rex_x:1,
-			rex_b:1,
-			rex_present:1;
-
-	uint8_t		mod:2,			/* ModRM byte */
-			reg:4,
-			rm:4;
-
-	uint8_t		ss:2,			/* SIB byte */
-			index:4,
-			base:4;
-
-	uint8_t		disp_bytes;
-	uint8_t		imm_bytes;
-
-	uint8_t		scale;
-	int		base_register;		/* VM_REG_GUEST_xyz */
-	int		index_register;		/* VM_REG_GUEST_xyz */
-
-	int64_t		displacement;		/* optional addr displacement */
-	int64_t		immediate;		/* optional immediate operand */
-
-	uint8_t		decoded;	/* set to 1 if successfully decoded */
-
-	struct vie_op	op;			/* opcode description */
-};
+#include <sys/mman.h>
 
 /*
  * Callback functions to read and write memory regions.
@@ -111,6 +55,24 @@ int vmm_emulate_instruction(void *vm, int cpuid, uint64_t gpa, struct vie *vie,
 			    mem_region_read_t mrr, mem_region_write_t mrw,
 			    void *mrarg);
 
+int vie_update_register(void *vm, int vcpuid, enum vm_reg_name reg,
+    uint64_t val, int size);
+
+/*
+ * Returns 1 if an alignment check exception should be injected and 0 otherwise.
+ */
+int vie_alignment_check(int cpl, int operand_size, uint64_t cr0,
+    uint64_t rflags, uint64_t gla);
+
+/* Returns 1 if the 'gla' is not canonical and 0 otherwise. */
+int vie_canonical_check(enum vm_cpu_mode cpu_mode, uint64_t gla);
+
+uint64_t vie_size2mask(int size);
+
+int vie_calculate_gla(enum vm_cpu_mode cpu_mode, enum vm_reg_name seg,
+    struct seg_desc *desc, uint64_t off, int length, int addrsize, int prot,
+    uint64_t *gla);
+
 #ifdef _KERNEL
 /*
  * APIs to fetch and decode the instruction from nested page fault handler.
@@ -118,8 +80,18 @@ int vmm_emulate_instruction(void *vm, int cpuid, uint64_t gpa, struct vie *vie,
  * 'vie' must be initialized before calling 'vmm_fetch_instruction()'
  */
 int vmm_fetch_instruction(struct vm *vm, int cpuid,
-			  uint64_t rip, int inst_length, uint64_t cr3,
-			  enum vie_paging_mode paging_mode, struct vie *vie);
+			  struct vm_guest_paging *guest_paging,
+			  uint64_t rip, int inst_length, struct vie *vie);
+
+/*
+ * Translate the guest linear address 'gla' to a guest physical address.
+ *
+ * Returns 0 on success and '*gpa' contains the result of the translation.
+ * Returns 1 if an exception was injected into the guest.
+ * Returns -1 otherwise.
+ */
+int vmm_gla2gpa(struct vm *vm, int vcpuid, struct vm_guest_paging *paging,
+    uint64_t gla, int prot, uint64_t *gpa);
 
 void vie_init(struct vie *vie);
 
@@ -136,7 +108,7 @@ void vie_init(struct vie *vie);
  */
 #define	VIE_INVALID_GLA		(1UL << 63)	/* a non-canonical address */
 int vmm_decode_instruction(struct vm *vm, int cpuid, uint64_t gla,
-			   enum vie_cpu_mode cpu_mode, struct vie *vie);
+			   enum vm_cpu_mode cpu_mode, struct vie *vie);
 #endif	/* _KERNEL */
 
 #endif	/* _VMM_INSTRUCTION_EMUL_H_ */