1 /******************************************************************************
4 * Guest OS interface to x86 64-bit Xen.
6 * Copyright (c) 2004, K A Fraser
9 #ifndef __XEN_PUBLIC_ARCH_X86_64_H__
10 #define __XEN_PUBLIC_ARCH_X86_64_H__
12 #define __DEFINE_XEN_GUEST_HANDLE(name, type) \
13 typedef struct { type *p; } __guest_handle_ ## name
15 #define DEFINE_XEN_GUEST_HANDLE(name) __DEFINE_XEN_GUEST_HANDLE(name, name)
16 #define XEN_GUEST_HANDLE(name) __guest_handle_ ## name
17 #define set_xen_guest_handle(hnd, val) do { (hnd).p = val; } while (0)
19 #define get_xen_guest_handle(val, hnd) do { val = (hnd).p; } while (0)
23 /* Guest handles for primitive C types. */
24 __DEFINE_XEN_GUEST_HANDLE(uchar, unsigned char);
25 __DEFINE_XEN_GUEST_HANDLE(uint, unsigned int);
26 __DEFINE_XEN_GUEST_HANDLE(ulong, unsigned long);
27 DEFINE_XEN_GUEST_HANDLE(char);
28 DEFINE_XEN_GUEST_HANDLE(int);
29 DEFINE_XEN_GUEST_HANDLE(long);
30 DEFINE_XEN_GUEST_HANDLE(void);
32 typedef unsigned long xen_pfn_t;
33 DEFINE_XEN_GUEST_HANDLE(xen_pfn_t);
37 * SEGMENT DESCRIPTOR TABLES
40 * A number of GDT entries are reserved by Xen. These are not situated at the
41 * start of the GDT because some stupid OSes export hard-coded selector values
42 * in their ABI. These hard-coded values are always near the start of the GDT,
43 * so Xen places itself out of the way, at the far end of the GDT.
45 #define FIRST_RESERVED_GDT_PAGE 14
46 #define FIRST_RESERVED_GDT_BYTE (FIRST_RESERVED_GDT_PAGE * 4096)
47 #define FIRST_RESERVED_GDT_ENTRY (FIRST_RESERVED_GDT_BYTE / 8)
50 * 64-bit segment selectors
51 * These flat segments are in the Xen-private section of every GDT. Since these
52 * are also present in the initial GDT, many OSes will be able to avoid
53 * installing their own GDT.
56 #define FLAT_RING3_CS32 0xe023 /* GDT index 260 */
57 #define FLAT_RING3_CS64 0xe033 /* GDT index 261 */
58 #define FLAT_RING3_DS32 0xe02b /* GDT index 262 */
59 #define FLAT_RING3_DS64 0x0000 /* NULL selector */
60 #define FLAT_RING3_SS32 0xe02b /* GDT index 262 */
61 #define FLAT_RING3_SS64 0xe02b /* GDT index 262 */
63 #define FLAT_KERNEL_DS64 FLAT_RING3_DS64
64 #define FLAT_KERNEL_DS32 FLAT_RING3_DS32
65 #define FLAT_KERNEL_DS FLAT_KERNEL_DS64
66 #define FLAT_KERNEL_CS64 FLAT_RING3_CS64
67 #define FLAT_KERNEL_CS32 FLAT_RING3_CS32
68 #define FLAT_KERNEL_CS FLAT_KERNEL_CS64
69 #define FLAT_KERNEL_SS64 FLAT_RING3_SS64
70 #define FLAT_KERNEL_SS32 FLAT_RING3_SS32
71 #define FLAT_KERNEL_SS FLAT_KERNEL_SS64
73 #define FLAT_USER_DS64 FLAT_RING3_DS64
74 #define FLAT_USER_DS32 FLAT_RING3_DS32
75 #define FLAT_USER_DS FLAT_USER_DS64
76 #define FLAT_USER_CS64 FLAT_RING3_CS64
77 #define FLAT_USER_CS32 FLAT_RING3_CS32
78 #define FLAT_USER_CS FLAT_USER_CS64
79 #define FLAT_USER_SS64 FLAT_RING3_SS64
80 #define FLAT_USER_SS32 FLAT_RING3_SS32
81 #define FLAT_USER_SS FLAT_USER_SS64
83 /* And the trap vector is... */
84 #define TRAP_INSTR "syscall"
86 #define __HYPERVISOR_VIRT_START 0xFFFF800000000000
87 #define __HYPERVISOR_VIRT_END 0xFFFF880000000000
88 #define __MACH2PHYS_VIRT_START 0xFFFF800000000000
89 #define __MACH2PHYS_VIRT_END 0xFFFF804000000000
91 #ifndef HYPERVISOR_VIRT_START
92 #define HYPERVISOR_VIRT_START mk_unsigned_long(__HYPERVISOR_VIRT_START)
93 #define HYPERVISOR_VIRT_END mk_unsigned_long(__HYPERVISOR_VIRT_END)
96 #define MACH2PHYS_VIRT_START mk_unsigned_long(__MACH2PHYS_VIRT_START)
97 #define MACH2PHYS_VIRT_END mk_unsigned_long(__MACH2PHYS_VIRT_END)
98 #define MACH2PHYS_NR_ENTRIES ((MACH2PHYS_VIRT_END-MACH2PHYS_VIRT_START)>>3)
99 #ifndef machine_to_phys_mapping
100 #define machine_to_phys_mapping ((unsigned long *)HYPERVISOR_VIRT_START)
103 /* Maximum number of virtual CPUs in multi-processor guests. */
104 #define MAX_VIRT_CPUS 32
108 typedef unsigned long xen_ulong_t;
111 * int HYPERVISOR_set_segment_base(unsigned int which, unsigned long base)
112 * @which == SEGBASE_* ; @base == 64-bit base address
113 * Returns 0 on success.
116 #define SEGBASE_GS_USER 1
117 #define SEGBASE_GS_KERNEL 2
118 #define SEGBASE_GS_USER_SEL 3 /* Set user %gs specified in base[15:0] */
121 * int HYPERVISOR_iret(void)
122 * All arguments are on the kernel stack, in the following format.
123 * Never returns if successful. Current kernel context is lost.
124 * The saved CS is mapped as follows:
125 * RING0 -> RING3 kernel mode.
126 * RING1 -> RING3 kernel mode.
127 * RING2 -> RING3 kernel mode.
128 * RING3 -> RING3 user mode.
129 * However RING0 indicates that the guest kernel should return to iteself
133 * If flags contains VGCF_IN_SYSCALL:
134 * Restore RAX, RIP, RFLAGS, RSP.
135 * Discard R11, RCX, CS, SS.
137 * Restore RAX, R11, RCX, CS:RIP, RFLAGS, SS:RSP.
138 * All other registers are saved on hypercall entry and restored to user.
140 /* Guest exited in SYSCALL context? Return to guest with SYSRET? */
141 #define VGCF_IN_SYSCALL (1<<8)
142 struct iret_context {
143 /* Top of stack (%rsp at point of hypercall). */
144 uint64_t rax, r11, rcx, flags, rip, cs, rflags, rsp, ss;
145 /* Bottom of iret stack frame. */
149 * Send an array of these to HYPERVISOR_set_trap_table().
150 * N.B. As in x86/32 mode, the privilege level specifies which modes may enter
151 * a trap via a software interrupt. Since rings 1 and 2 are unavailable, we
152 * allocate privilege levels as follows:
153 * Level == 0: Noone may enter
154 * Level == 1: Kernel may enter
155 * Level == 2: Kernel may enter
156 * Level == 3: Everyone may enter
158 #define TI_GET_DPL(_ti) ((_ti)->flags & 3)
159 #define TI_GET_IF(_ti) ((_ti)->flags & 4)
160 #define TI_SET_DPL(_ti,_dpl) ((_ti)->flags |= (_dpl))
161 #define TI_SET_IF(_ti,_if) ((_ti)->flags |= ((!!(_if))<<2))
163 uint8_t vector; /* exception vector */
164 uint8_t flags; /* 0-3: privilege level; 4: clear event enable? */
165 uint16_t cs; /* code selector */
166 unsigned long address; /* code offset */
168 typedef struct trap_info trap_info_t;
169 DEFINE_XEN_GUEST_HANDLE(trap_info_t);
172 /* Anonymous union includes both 32- and 64-bit names (e.g., eax/rax). */
173 #define __DECL_REG(name) union { uint64_t r ## name, e ## name; }
175 /* Non-gcc sources must always use the proper 64-bit name (e.g., rax). */
176 #define __DECL_REG(name) uint64_t r ## name
179 struct cpu_user_regs {
195 uint32_t error_code; /* private */
196 uint32_t entry_vector; /* private */
198 uint16_t cs, _pad0[1];
199 uint8_t saved_upcall_mask;
201 __DECL_REG(flags); /* rflags.IF == !saved_upcall_mask */
203 uint16_t ss, _pad2[3];
204 uint16_t es, _pad3[3];
205 uint16_t ds, _pad4[3];
206 uint16_t fs, _pad5[3]; /* Non-zero => takes precedence over fs_base. */
207 uint16_t gs, _pad6[3]; /* Non-zero => takes precedence over gs_base_usr. */
209 typedef struct cpu_user_regs cpu_user_regs_t;
210 DEFINE_XEN_GUEST_HANDLE(cpu_user_regs_t);
214 typedef uint64_t tsc_timestamp_t; /* RDTSC timestamp */
217 * The following is all CPU context. Note that the fpu_ctxt block is filled
218 * in by FXSAVE if the CPU has feature FXSR; otherwise FSAVE is used.
220 struct vcpu_guest_context {
221 /* FPU registers come first so they can be aligned for FXSAVE/FXRSTOR. */
222 struct { char x[512]; } fpu_ctxt; /* User-level FPU registers */
223 #define VGCF_I387_VALID (1<<0)
224 #define VGCF_HVM_GUEST (1<<1)
225 #define VGCF_IN_KERNEL (1<<2)
226 #define _VGCF_i387_valid 0
227 #define VGCF_i387_valid (1<<_VGCF_i387_valid)
228 #define _VGCF_hvm_guest 1
229 #define VGCF_hvm_guest (1<<_VGCF_hvm_guest)
230 #define _VGCF_in_kernel 2
231 #define VGCF_in_kernel (1<<_VGCF_in_kernel)
232 #define _VGCF_failsafe_disables_events 3
233 #define VGCF_failsafe_disables_events (1<<_VGCF_failsafe_disables_events)
234 #define _VGCF_syscall_disables_events 4
235 #define VGCF_syscall_disables_events (1<<_VGCF_syscall_disables_events)
236 unsigned long flags; /* VGCF_* flags */
237 struct cpu_user_regs user_regs; /* User-level CPU registers */
238 struct trap_info trap_ctxt[256]; /* Virtual IDT */
239 unsigned long ldt_base, ldt_ents; /* LDT (linear address, # ents) */
240 unsigned long gdt_frames[16], gdt_ents; /* GDT (machine frames, # ents) */
241 unsigned long kernel_ss, kernel_sp; /* Virtual TSS (only SS1/SP1) */
242 unsigned long ctrlreg[8]; /* CR0-CR7 (control registers) */
243 unsigned long debugreg[8]; /* DB0-DB7 (debug registers) */
244 unsigned long event_callback_eip;
245 unsigned long failsafe_callback_eip;
246 unsigned long syscall_callback_eip;
247 unsigned long vm_assist; /* VMASST_TYPE_* bitmap */
248 /* Segment base addresses. */
250 uint64_t gs_base_kernel;
251 uint64_t gs_base_user;
253 typedef struct vcpu_guest_context vcpu_guest_context_t;
254 DEFINE_XEN_GUEST_HANDLE(vcpu_guest_context_t);
256 #define xen_pfn_to_cr3(pfn) ((unsigned long)(pfn) << 12)
257 #define xen_cr3_to_pfn(cr3) ((unsigned long)(cr3) >> 12)
259 struct arch_shared_info {
260 unsigned long max_pfn; /* max pfn that appears in table */
261 /* Frame containing list of mfns containing list of mfns containing p2m. */
262 xen_pfn_t pfn_to_mfn_frame_list_list;
263 unsigned long nmi_reason;
266 typedef struct arch_shared_info arch_shared_info_t;
268 struct arch_vcpu_info {
270 unsigned long pad; /* sizeof(vcpu_info_t) == 64 */
272 typedef struct arch_vcpu_info arch_vcpu_info_t;
274 typedef unsigned long xen_callback_t;
276 #endif /* !__ASSEMBLY__ */
279 * Prefix forces emulation of some non-trapping instructions.
280 * Currently only CPUID.
283 #define XEN_EMULATE_PREFIX .byte 0x0f,0x0b,0x78,0x65,0x6e ;
284 #define XEN_CPUID XEN_EMULATE_PREFIX cpuid
286 #define XEN_EMULATE_PREFIX ".byte 0x0f,0x0b,0x78,0x65,0x6e ; "
287 #define XEN_CPUID XEN_EMULATE_PREFIX "cpuid"
298 * indent-tabs-mode: nil